A method and apparatus for optimizing a power and thermal management system fuel control law

Abstract

The application provides a power and heat management system fuel control law optimization method and device, and relates to the technical field of power system control technology. The method comprises the following steps: acquiring real-time working condition data of the power and heat management system; inputting the real-time working condition data into an MLP neural network model to predict control stage feedback control PID parameters and control stage switching endpoint state values; correcting a feedforward base quantity based on a feedforward correction factor to obtain a final feedforward control quantity; determining a control stage according to a current rotating speed and the control stage switching endpoint state values, and selecting corresponding PID parameters to calculate a feedback control quantity; and combining the final feedforward control quantity and the final feedback control quantity to generate a fuel quantity control instruction output. The MLP neural network model is used to perform online optimization on the PID parameters, the switching endpoint and the feedforward correction factor in the fuel control law, so that the feedforward control and the feedback control work cooperatively, and the adaptive capacity, the response performance and the operation stability of the fuel control are improved.

Description

Technical Field

[0001] This application relates to the field of power system control technology, and in particular to a method and apparatus for optimizing fuel control laws in a power and thermal management system. Background Technology

[0002] In aero-engines and their power and thermal management systems, fuel control laws are used to achieve stable engine operation and performance regulation. Fuel also serves as a crucial heat sink in the thermal management system, absorbing the heat load generated by onboard equipment. Therefore, fuel control not only affects engine thrust output but also directly relates to the overall thermal management capability of the aircraft. Current technologies generally employ a segmented PID control strategy, dividing the operation into start-up, acceleration transition, and steady-state stages based on engine speed, and using different control methods in each stage. For example, open-loop control is often used in the start-up stage, while closed-loop PID control is used in the acceleration and steady-state stages. While this approach is simple to implement, it has significant shortcomings. Firstly, PID control parameters are usually fixed or empirically tuned. Under varying flight altitudes, speeds, and thermal management loads, the engine's dynamic characteristics change, making it difficult for fixed parameters to consistently maintain optimal control, leading to decreased control performance. Secondly, current stage switching relies heavily on fixed speed thresholds, lacking adaptability to real-time operating conditions, and prone to response lag, overshoot, or decreased stability under complex operating conditions. Furthermore, existing fuel feedforward control typically employs simple proportional relationships for compensation, failing to adequately consider the impact of changes in flight environment and system load. This results in limited feedforward compensation accuracy and a tendency for control mismatch issues to arise under complex operating conditions, affecting overall control performance. Therefore, existing fuel control laws still lack sufficient adaptive capability and control accuracy when facing complex flight envelopes and variable thermal management loads. Summary of the Invention

[0003] In view of this, this application provides a method and apparatus for optimizing the fuel control law of a power and thermal management system, in order to solve the problems of fixed segmented PID control parameters, insufficient feedforward control accuracy, and difficulty in adapting to complex and variable operating conditions.

[0004] Specifically, this application is implemented through the following technical solution:

[0005] The first aspect of this application provides a method for optimizing fuel control laws in a power and thermal management system, the method comprising:

[0006] Acquire real-time operating data of the power and thermal management system;

[0007] The real-time operating data is input into the MLP neural network model to predict the feedback control PID parameters of the control stage, the switching endpoint state value of the control stage, and the feedforward correction factor. The feedforward correction factor is used to correct individual differences in the power and thermal management system. The control stage includes at least one of the speed closed-loop PID control stage and the acceleration closed-loop PID control stage.

[0008] The feedforward base quantity is corrected based on the aforementioned feedforward correction factor to obtain the final feedforward control quantity;

[0009] Based on the relationship between the current speed and the state value of the control stage switching endpoint, the current control stage is determined. Based on the current control stage, a target PID control parameter is selected from the feedback control PID parameters of the control stage. Based on the target PID control parameter, the final feedback control quantity corresponding to the real-time operating data is calculated.

[0010] The final feedforward control quantity and the final feedback control quantity are combined to generate a fuel quantity control command output.

[0011] A second aspect of this application provides a fuel control law optimization device for a power and thermal management system, the device comprising an acquisition module, a prediction module, and a calculation module, wherein:

[0012] The acquisition module is used to acquire real-time operating data of the power and thermal management system;

[0013] The prediction module is used to input the real-time operating condition data into the MLP neural network model, predict the feedback control PID parameters, the switching endpoint state values ​​and the feedforward correction factor in the control stage. The feedforward correction factor is used to correct individual differences in the power and thermal management system. The control stage includes at least one of the speed closed-loop PID control stage and the acceleration closed-loop PID control stage.

[0014] The calculation module is used to correct the feedforward base quantity based on the feedforward correction factor to obtain the final feedforward control quantity.

[0015] The calculation module is used to determine the current control stage based on the relationship between the current speed and the state value of the control stage switching endpoint, select a target PID control parameter from the feedback control PID parameters of the control stage according to the current control stage, and calculate the final feedback control quantity corresponding to the real-time operating data based on the target PID control parameter.

[0016] The calculation module is used to generate a fuel quantity control command output based on merging the final feedforward control quantity and the final feedback control quantity.

[0017] This application provides a method and apparatus for optimizing fuel control laws in a power and thermal management system. Compared to traditional PID control with fixed endpoint values ​​using multiple control strategies (where the timing of control strategy application is manually fixed), this application directly uses an MLP neural network for adaptive prediction. The MLP neural network can adaptively predict the switching endpoints for different switching states, as well as the feedforward correction factor and feedback control quantity, based on the state conditions. On the one hand, by comprehensively predicting relevant information from feedforward and feedback control variables, the coordination between feedforward and feedback control is improved, leading to more scientific control of the power and thermal management system. On the other hand, by adapting to the system's own predicted control strategy switching endpoints, customized configurations of one value per machine or multiple endpoints are achieved, thus ensuring the scientific nature of control strategy switching through customized endpoint values. This application can improve the adaptability, control accuracy, and robustness of fuel control laws under complex flight conditions and variable thermal management loads, achieving dynamic optimization and stable control of fuel. Attached Figure Description

[0018] Figure 1 A flowchart of an embodiment of the fuel control law optimization method for the power and thermal management system provided in this application;

[0019] Figure 2 A schematic diagram illustrating the switching relationship of the operating modes of the power and thermal management system, as shown in an exemplary embodiment of this application;

[0020] Figure 3 A schematic diagram of the structure of the fuel control law optimization device for the power and thermal management system provided in this application. Detailed Implementation

[0021] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0022] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0023] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0024] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0025] Figure 1 This is a flowchart of an embodiment of the fuel control law optimization method for the power and thermal management system provided in this application. Please refer to... Figure 1 The method provided in this embodiment may include:

[0026] S101. Obtain real-time operating data of the power and thermal management system.

[0027] Specifically, the method described in this application is applied to an engine control system related to integrated power and thermal management control. The power and thermal management system includes at least an engine control unit, a sensor acquisition unit, and a fuel actuator. The engine control unit is used to comprehensively calculate and control the engine's operating status and thermal management status; the sensor acquisition unit is used to collect real-time operating condition data during system operation and send this data to the engine control unit; the fuel actuator is used to adjust the fuel supply according to the fuel control commands output by the engine control unit.

[0028] Specifically, the real-time operating condition data is acquired in real time by the sensor acquisition unit and input to the engine control unit for processing. The real-time operating condition data includes at least engine status parameters, flight environment parameters, and thermal management system load parameters. Specifically, the engine status parameters include at least engine speed, compressor outlet pressure, total intake air temperature, and fuel flow rate; the flight environment parameters include at least flight altitude, Mach number, and ambient temperature; and the thermal management system load parameters include at least bleed air load, power generation load, and liquid cooling system power requirements.

[0029] Furthermore, the sensor acquisition unit performs periodic acquisition according to a preset sampling period, or event-triggered acquisition when the engine operating conditions change, and performs time synchronization processing on the data from each channel to form a unified real-time operating condition data vector. It should be noted that the engine control unit normalizes the real-time operating condition data to eliminate the influence of different units on subsequent calculations.

[0030] S102. Input the real-time operating condition data into the MLP neural network model to predict the feedback control PID parameters, the switching endpoint state values, and the feedforward correction factor in the control stage. The feedforward correction factor is used to correct individual differences in the power and thermal management system. The control stage includes at least one of the speed closed-loop PID control stage and the acceleration closed-loop PID control stage.

[0031] Specifically, the feedforward correction factor is used to correct for individual differences in the power and thermal management system, which characterizes the differences in dynamic response between different engine individuals due to manufacturing, assembly, or long-term operation.

[0032] Specifically, the power and thermal management system divides the operation into different control stages based on the relationship between the current engine speed and the switching endpoint state value, with different fuel control strategies employed in each stage. Specifically, at lower engine speeds, an open-loop control strategy is used to ensure ignition and initial stability; as the engine speed increases, an acceleration closed-loop PID control strategy is used to improve the engine's dynamic response performance; and once the engine speed reaches a predetermined range, a speed closed-loop PID control strategy is used to ensure system stability and control accuracy. It should be noted that the division of control stages is illustrative, and the specific number and division method can be adjusted according to actual control requirements. Furthermore, the switching between different control stages is determined based on the current operating state, and the control parameters for each stage can be adjusted according to actual operating conditions.

[0033] Specifically, the control phase switching endpoint state value is used to delineate the critical speed between the acceleration closed-loop PID control stage and the speed closed-loop PID control stage, and serves as a criterion for switching between different control strategies. When the current engine speed is less than the control phase switching endpoint state value, the system is in the acceleration closed-loop PID control stage. The controller uses the deviation between the target acceleration and the actual acceleration as input to adjust fuel, thereby improving the engine's dynamic response capability. When the current engine speed is greater than or equal to the control phase switching endpoint state value, the system switches to the speed closed-loop PID control stage. The controller uses the deviation between the target speed and the actual speed as input to adjust fuel, thereby ensuring speed control accuracy and system operational stability. Furthermore, the control phase switching endpoint state value is adaptively output by the MLP neural network model based on the current operating parameters. Compared to using a fixed speed threshold for stage division, this method can dynamically adjust the stage switching position according to changes in the flight environment, engine status, and load, thereby avoiding the switching lag or premature problems caused by a fixed threshold and improving the rationality and smoothness of control phase switching.

[0034] Specifically, the MLP neural network model performs forward calculations on the input real-time operating data and outputs control law parameters that match the current operating conditions. These control law parameters include at least the PID parameters for the control phase feedback control, the state values ​​of the control phase switching endpoints, and a feedforward correction factor. The output control law parameters are used in subsequent feedback control calculations, control phase determinations, and feedforward control correction processes, thereby providing a parameter basis for the generation of fuel control commands in the power and thermal management system.

[0035] Optionally, in one possible implementation, the MLP neural network model takes flight environment parameters, engine state parameters, and thermal management system state parameters as inputs, and outputs control phase feedback control PID parameters, control phase switching endpoint state values, and feedforward correction factors.

[0036] Specifically, the MLP neural network model is used to perform nonlinear mapping processing on real-time operating data to achieve adaptive generation of control law parameters. The real-time operating data is used to characterize the comprehensive operating status of the power and thermal management system under different flight and load conditions, thereby providing an input basis for subsequent control parameter prediction and realizing online optimization of control parameters.

[0037] (1) The MLP neural network model includes an input layer, at least one hidden layer and an output layer.

[0038] Specifically, the input layer receives feature vectors composed of real-time operating condition data, and the hidden layer performs nonlinear mapping and feature fusion processing on the input features to extract the coupling relationship between different operating condition parameters. In this embodiment, the MLP neural network model includes two hidden layers: the first hidden layer has 12 neurons, the second hidden layer has 8 neurons, and the activation function is ReLU; the output layer outputs control law optimization parameters that match the current operating condition.

[0039] (2) The flight environment parameters include at least flight altitude and Mach number, the engine status parameters include at least speed and compressor outlet pressure, and the thermal management system status parameters include at least one or more of power generation, liquid cooling power and bleed air load.

[0040] Specifically, the flight environment parameters characterize the impact of changes in the external flight environment on the operating state of the power system, the engine state parameters characterize changes in the internal operating conditions of the engine, and the thermal management system state parameters characterize the impact of changes in airborne heat load demand on the fuel control process. The input feature vector formed by combining these multi-source parameters can comprehensively characterize the real-time operating state of the power and thermal management systems, thereby enabling the MLP neural network model to adapt to complex operating condition changes.

[0041] Optionally, in one possible implementation, the MLP neural network model is trained in the following manner:

[0042] (1) Select multiple operating points within the preset flight envelope range.

[0043] Specifically, the preset flight envelope range is used to characterize the entire typical operating range of the engine within its design operating boundary, including the operating state intervals under different combinations of altitude, Mach number, and engine speed. Within this flight envelope range, multiple operating points are selected according to a preset discretization strategy. These operating points characterize the system's steady-state or quasi-steady-state operating state under different typical flight and load conditions. For example, representative operating points can be selected in low-altitude low-speed, high-altitude cruise, and cross-speed domain transition regions to ensure that the training samples cover the entire engine operating domain.

[0044] It should be noted that the selection method of the operating condition points is not limited to regular grid division. Latin hypercube sampling or clustering selection based on historical flight data can also be used to improve the uniformity and representativeness of operating condition coverage.

[0045] (2) Based on the full digital simulation model, for each operating point, the optimization algorithm is used to optimize the PID parameters of the feedback control in the control stage, the state values ​​of the switching endpoints in the control stage, and the feedforward correction factor, so as to minimize the preset loss function and obtain the corresponding optimal parameter set.

[0046] Specifically, the fully digital simulation model is used to perform high-fidelity modeling of the dynamic operation process of the power and thermal management system under different operating conditions, so as to simulate the dynamic characteristics of the system under engine speed response, fuel supply changes and thermal management load, and to evaluate the system control effect under different combinations of control parameters.

[0047] Furthermore, each operating point is represented by a multi-dimensional feature vector consisting of engine state parameters, flight environment parameters, and thermal management load parameters, and serves as input conditions for the simulation model to drive the operation of the fully digital simulation model. It should be noted that at each operating point, the PID parameters of the control phase feedback control, the state values ​​of the control phase switching endpoints, and the feedforward correction factor are input as optimization variables into the optimization algorithm. The initial values ​​of these optimization variables can be determined based on prior empirical values ​​or historical calibration data, and their search range is pre-set according to the engine's stable operating boundary and thermal safety constraints. The optimization algorithm includes one or more of the following: swarm intelligence-based optimization algorithms, gradient optimization algorithms, or heuristic search algorithms. By iteratively optimizing different parameter combinations and evaluating them in conjunction with the system response results output by the simulation model, the parameter combinations are gradually updated until the convergence condition is met, thus obtaining the optimal parameter set corresponding to that operating point. The convergence condition includes the preset loss function value no longer decreasing after multiple consecutive iterations, or the number of iterations reaching a preset upper limit.

[0048] Optionally, in one possible implementation, the preset loss function is composed of a weighted average of multiple control performance indicators, which at least include response time, overshoot, surge margin loss, exhaust temperature overrun penalty, and steady-state error.

[0049] Wherein, (A) the response time is the time elapsed from the issuance of the control command to the engine speed reaching the target speed preset ratio.

[0050] Specifically, the response time is determined based on the engine speed response curve, and is specifically the time elapsed from the issuance of the control command to the engine's actual speed first reaching a preset proportion of the target speed. This time is used to characterize the system's dynamic response speed to fuel adjustment commands. In this embodiment, the preset proportion is set to 90%.

[0051] (B) The overshoot is the maximum deviation of the engine speed from the target speed.

[0052] Specifically, the overshoot is calculated based on the deviation between the actual speed curve and the target speed during the speed response process. Specifically, it is the ratio of the maximum deviation when the speed exceeds the target value to the target speed, and is used to characterize the degree of dynamic overshoot caused by fuel regulation during the control process.

[0053] (C) The surge margin loss is the deviation of the actual surge margin from the minimum allowable surge margin.

[0054] Specifically, the surge margin loss is calculated based on the distance between the compressor operating point and the surge boundary, and is used to characterize the degree of deviation of the compressor operating stability margin from the safety boundary under the current control action.

[0055] (D) The exhaust temperature over-limit penalty is the penalty amount corresponding to the power turbine outlet temperature exceeding the preset temperature limit.

[0056] Specifically, the exhaust temperature over-limit penalty is determined based on the difference between the turbine outlet temperature monitoring value and the preset upper temperature limit. When the temperature exceeds the limit, a corresponding penalty amount is introduced to constrain the safe operation of the hot-end components.

[0057] (E) The steady-state error is the absolute value of the deviation of the engine speed from the target speed in steady state.

[0058] Specifically, the steady-state error is calculated based on the absolute value of the deviation between the average rotational speed after the system enters steady state and the target rotational speed, and is used to characterize the steady-state control accuracy of the system.

[0059] (F) Multiply each of the control performance indicators by its corresponding weight coefficient and sum them to obtain the loss function value.

[0060] Specifically, each control performance index is multiplied by its corresponding weighting coefficient, and then a weighted sum is taken to obtain the loss function value. Each weighting coefficient is used to balance the influence of dynamic performance, stability margin, and steady-state accuracy on the overall control objective. The expression for the loss function is:

[0061] ;

[0062] in, For the corresponding time, For overshoot, For surge margin loss, The outlet temperature of the power turbine. For steady-state error, These are weighting coefficients used to balance the influence of different control performance indicators on the overall control objective. These weighting coefficients are set based on simulation optimization results or historical calibration data, with response time and overshoot having higher weighting priority than steady-state error and thermal constraints, in order to prioritize the system's rapid response capability and dynamic stability.

[0063] (3) Using the working condition feature parameters corresponding to each working condition point as input and the optimal parameter group as label, a training sample set is constructed.

[0064] Specifically, the operating condition feature parameters are used to characterize the input state information of the power and thermal management system under different operating conditions, including flight environment parameters, engine state parameters, and thermal management system state parameters. Further, the operating condition feature parameters corresponding to each operating condition are used as input vectors, and the optimal parameter set obtained through the optimization algorithm is used as output labels, thereby constructing supervised learning training sample pairs. The optimal parameter set includes PID parameters for control phase feedback control, control phase switching endpoint state values, and feedforward correction factors, used to characterize the optimal control law configuration result under that operating condition. It should be noted that the training sample set covers multiple flight envelope operating conditions, thereby ensuring that the training data has good generalization ability and operating condition adaptability.

[0065] (4) Supervised learning training of the MLP neural network model based on the training sample set.

[0066] Specifically, the MLP neural network model is trained using sample pairs based on input operating condition characteristics and corresponding optimal control parameters, enabling the model to learn the nonlinear mapping relationship between operating conditions and control parameters. During training, an error backpropagation algorithm is used to iteratively update the network parameters. The error function characterizes the deviation between the predicted output and the target output, and its form includes a mean square error function, an absolute error function, or an error function constructed based on a weighted average of control performance indicators. When the training error meets a preset convergence condition or reaches a preset number of iterations, training stops, and the trained MLP neural network model is deployed to the engine control unit for online control parameter prediction.

[0067] Optionally, in one possible implementation, the supervised learning training of the MLP neural network model based on the training sample set further includes:

[0068] (A) Calculate the change between the parameter values ​​of the current working condition and the parameter values ​​of the previous working condition, wherein the change includes the change value and the change direction.

[0069] Specifically, during the supervised learning training process, for the continuously sampled operating condition data of the engine operation, the sequence of operating condition parameters within two adjacent control cycles is obtained, where the current operating condition is represented by time. The state vector, where the previous operating condition is represented by time t. The state vector. The change in each parameter is obtained by calculating the dimension-by-dimensional difference between the current operating condition parameters and the previous operating condition parameters. The change includes both the numerical value of the change and the direction of the change. When When the change is greater than or equal to zero, the direction of change is defined as upward; when When the value is less than zero, the direction of change is defined as the downward direction, thus forming a two-dimensional change representation that includes the magnitude and trend of change.

[0070] (B) Generate a change description based on the change value and the change direction.

[0071] Specifically, the variation characteristics of each operating condition parameter are structured and encoded based on the change amount to form a change description vector. This change description vector is used to characterize the dynamic evolution trend of the operating condition in the time dimension. Each dimension contains the numerical change amount and direction identifier of the corresponding parameter, thereby transforming the original operating condition changes into a unified trend expression form for subsequent neural network input.

[0072] Specifically, based on the changes in parameters under various operating conditions, the change characteristics are structured and encoded to generate a change description vector. This change description vector can be represented as:

[0073] ;

[0074] Among them, the Dimensional elements The change in the corresponding parameter value The direction of change is jointly determined. In one implementation method, The following method can be used to construct it: change the numerical value After normalization, it is combined with the change direction identifier for encoding. Wherein, when Time direction indicator ,when Time direction indicator This yields information that includes both the magnitude and trend of change. Alternatively, the numerical change value can be directly concatenated with the direction indicator to form a two-dimensional feature pair. As the first The description of parameter changes. Through the above method, the time-dimensional changes of the original operating parameters are transformed into a unified vector representation. This allows the change description to reflect not only the magnitude of each parameter's change but also its trend, thus providing auxiliary dynamic feature input for the subsequent MLP neural network model. Furthermore, the change description vector can be concatenated with the original operating parameter vector as extended input features to the MLP neural network model, enhancing the model's ability to perceive trends in operating conditions.

[0075] (C) The description of the change is used as a prompt and input into the MLP neural network model along with the training sample set.

[0076] Specifically, the change description vector is concatenated with the current operating condition state vector to obtain an extended input vector. This extended input vector is then input into the MLP neural network model. The prompt information is used to introduce first-order time difference features, enabling the model to simultaneously map the current operating condition state and model the trend of operating condition changes, thereby improving its ability to represent non-stationary operating conditions.

[0077] (D) The MLP neural network model makes a first direction prediction based on the input in the training sample set.

[0078] Specifically, the MLP neural network model performs forward computation based on the extended input vector and outputs a first direction prediction result. This first direction prediction result characterizes the evolution direction of the operating state in the next control cycle, including an upward trend, a downward trend, or a steady-state trend. The first direction prediction output adopts a softmax probability output form and is trained using a classification loss function. It should be noted that the first direction prediction result does not directly participate in the control output but is used for subsequent solving of spatial constraints.

[0079] (E) Define the solution space based on the difference between the first direction and the changing direction. Specifically, compare the predicted result of the first direction with the actual changing direction, and dynamically adjust the solution space for subsequent parameter predictions based on the degree of consistency between the two. Here, the predicted direction and the actual changing direction are respectively represented as... and When the signs of the two parameters are the same, they are determined to be in the same direction; when the signs are opposite, they are determined to be in different directions. The solution space refers to the range of candidate values ​​for the control parameters, which is composed of the upper and lower limit intervals of each control parameter, including the value intervals of the PID parameters, the state values ​​of the control stage switching endpoints, and the feedforward correction factor. Its boundaries are determined by the engine physical operating constraints and the historical calibration range.

[0080] Specifically, when the predicted result matches the actual direction, the solution space for the control parameters is reduced to a local neighborhood centered on the current parameter. For example, centered on the current parameter value, the original value interval is reduced by a preset shrinkage coefficient. ( The search range is narrowed down proportionally to limit it to a local neighborhood, thereby improving parameter search efficiency and convergence speed. In one optional implementation, the shrinkage coefficient... It can be set to 0.1 to 0.2, meaning the shrunk search range is within ±5% to ±10% of the current parameter value.

[0081] Specifically, when the predicted result differs from the actual direction, the solution space is expanded. For example, the solution space is expanded by a preset expansion coefficient based on the original value range. ( The parameters can be amplified or directly restored to a preset global physical constraint boundary range to enhance the globality of the parameter search and avoid getting trapped in local optima. In one exemplary implementation, the expansion coefficients... It can be set to 2 or 3, or directly restored to the global optimal parameter range obtained from the engine bench calibration.

[0082] (F) Predict the changed operating condition parameter values ​​in the solution space, and calculate the output of the MLP neural network model based on the changed operating condition parameter values.

[0083] Specifically, within the defined solution space, the operating condition state vector for the next control cycle is predicted to obtain a predicted operating condition state vector. This predicted operating condition state vector is then input into the MLP neural network model to obtain the corresponding control parameter output results. The output results include the PID parameters for the control phase feedback control, the state values ​​of the control phase switching endpoints, and the feedforward correction factor, which are used for subsequent fuel control law calculations. By introducing a joint mechanism of "trend constraint + solution space limitation," the model maintains stable parameter prediction capabilities even under rapid changes or disturbances in operating conditions, and reduces extrapolation errors under non-stationary operating conditions.

[0084] By incorporating the change description as a prompt into the input of the MLP neural network model, the model can not only base its parameter predictions on the current operating condition but also perceive the changing trend of the operating condition over time. Compared to methods that rely solely on static operating condition parameters for prediction, this embodiment can guide the direction of parameter adjustment in advance when the operating condition changes rapidly, thereby reducing invalid search paths, lowering computational time costs, and improving the accuracy and stability of control parameter output. Furthermore, the change description forms a guiding constraint on the parameter update direction, causing the parameter search process to preferentially focus on regions consistent with the changing trend of the operating condition, thus reducing the solution space while improving the stability and reliability of the prediction results. Through the above methods, a synergistic optimization of computational efficiency and control performance is achieved.

[0085] S103. Based on the feedforward correction factor, correct the feedforward base quantity to obtain the final feedforward control quantity.

[0086] Specifically, fuel feedforward control is used to generate a basic fuel compensation amount based on the current change in flight conditions when switching flight modes or experiencing rapid changes in flight conditions. This basic fuel compensation amount is then superimposed with the feedback control output to form the final fuel control command, thereby improving the system's response speed and control accuracy under transitional conditions. The fuel feedforward control includes a steady-state feedforward component and a dynamic feedforward component. The steady-state feedforward component characterizes the basic fuel demand under steady-state operating conditions. The dynamic feedforward component consists of a first dynamic feedforward fuel correction amount caused by operating mode switching and a second dynamic feedforward fuel correction amount caused by changes in external operating conditions.

[0087] Optionally, in one possible implementation, the final feedforward control quantity includes a steady-state feedforward fuel flow rate, and the method for obtaining the steady-state feedforward fuel flow rate includes:

[0088] (1) Obtain the engine operating status parameters under the current steady-state conditions. The engine operating status parameters include at least the compressor outlet pressure, the engine inlet total temperature and the current speed.

[0089] Specifically, the system acquires the operating parameters of the engine during its steady-state operation. The steady-state operation refers to the operating state where the engine speed fluctuation is below a preset threshold and the system has entered a stable operating range. The compressor outlet pressure, engine inlet total temperature, and current engine speed are acquired in real time through a sensor acquisition unit. The compressor outlet pressure characterizes the outlet pressure level after the compressor pressurizes the air, reflecting the engine's intake pressurization state; the engine inlet total temperature characterizes the total temperature of the airflow entering the engine core, which is affected by flight altitude, Mach number, and ambient temperature; and the current engine speed characterizes the engine's rotational state under the current steady-state operating conditions. It should be noted that the above parameters are processed using a moving average within a preset time window to reduce the impact of transient disturbances on steady-state identification, thereby improving the stability of the steady-state feedforward calculation.

[0090] (2) The steady-state feedforward fuel flow rate is calculated by using a preset function relationship based on the compressor outlet pressure, engine inlet total temperature and current speed.

[0091] Specifically, the compressor outlet pressure, engine inlet total temperature, and current engine speed are input into a preset steady-state feedforward function to calculate the steady-state feedforward fuel flow rate. The preset function relationship is expressed as follows:

[0092] ;

[0093] in, This is the steady-state feedforward fuel flow rate, used to characterize the benchmark value of fuel demand under the current steady-state operating conditions; This refers to the compressor outlet pressure. Total engine inlet temperature; The current rotational speed; The steady-state feedforward mapping function is used to establish a nonlinear mapping relationship between the three parameters mentioned above and the steady-state fuel flow rate. This steady-state feedforward mapping function can be implemented in the following ways: based on engine bench test data or high-fidelity simulation data, it is calibrated using a lookup table function, piecewise fitting function, or regression model, thereby outputting the corresponding fuel reference value under different steady-state operating condition parameter combinations. Through this method, this embodiment can quickly calculate the reference fuel flow rate based on the real-time engine operating state under steady-state conditions, serving as the basic fuel supply quantity for the fuel control system. This provides an accurate feedforward basis quantity for subsequent dynamic feedforward compensation and feedback control, thereby reducing the adjustment burden of closed-loop feedback control.

[0094] Optionally, in one possible implementation, the final feedforward control quantity further includes a first dynamic feedforward fuel correction quantity caused by the switching of operating modes, wherein the method for obtaining the first dynamic feedforward fuel correction quantity includes:

[0095] (1) When the engine working mode is detected to be switched, the target speed change before and after the working mode switch and the steady-state fuel reference value before the switch are obtained.

[0096] Figure 2 This is a schematic diagram illustrating the switching relationship of the power and thermal management system operating modes, as shown in an exemplary embodiment of this application. Please refer to... Figure 2 , Figure 2 The power and thermal management system is shown to have at least four operating modes: steady-state mode, main start mode, maintenance mode, and cooling mode. The steady-state mode characterizes stable engine operation; the main start mode is used for starting and powering the external main engine; the maintenance mode is used for low-load or maintenance operation; and the cooling mode is used for controlled power reduction operation to lower thermal load, in which fuel supply is gradually reduced until it reaches zero. The operating modes can be switched bidirectionally using the steady-state mode as a pivot, with changes in the target speed accompanying the mode switching process. It should be noted that the switching of operating modes causes a change in the system control target, specifically a change in the target speed. This change in target speed serves as a key input parameter for the first dynamic feedforward fuel correction, used to calculate the transient fuel compensation required during mode switching to reduce speed deviation and system overshoot during the switching process, thereby achieving rapid response and smooth transition during mode switching.

[0097] It should be noted that the self-start mode is used for the engine to transition from a standstill or low speed to a steady-state operating state. In this mode, fuel control employs a staged control strategy. For details on the staged control process in self-start mode, please refer to the relevant description; it will not be repeated here. After self-start is complete, the system enters steady-state mode. Because... Figure 2This section primarily describes the mode switching relationships during steady-state operation; therefore, the self-starting mode is not shown.

[0098] Specifically, when the engine control unit detects a change in operating mode, it triggers the first dynamic feedforward fuel correction calculation process. The target speed change is... This is the difference between the target speed after the switch and the target speed before the switch. When This indicates an increase in load demand. This indicates a decrease in load demand.

[0099] Specifically, the steady-state fuel reference value before the switch refers to the fuel flow rate when the system is in steady-state operation before the mode switch. It can be an actual measured value or a reference value calculated by the steady-state feedforward model. This value is used as a reference for the first dynamic feedforward fuel correction amount.

[0100] (2) Calculate the first dynamic feedforward fuel correction amount based on the target speed change and the steady-state feedforward fuel flow before the switch using a preset functional relationship.

[0101] Specifically, will Compared with the steady-state fuel baseline value before the switch Input to preset function relationship In the process, the first dynamic feedforward fuel correction amount is obtained. :

[0102] ;

[0103] in, This is the first dynamic feedforward fuel correction amount, used to compensate for transient fuel demand changes during the switching of operating modes; This is a mode switching mapping function used to characterize the correspondence between the target speed change and the steady-state fuel reference value. In a preferred embodiment, the... Linear interpolation is used. Specifically, based on the direction and magnitude of the change in the target speed before and after the switch, and combined with the steady-state fuel reference value before the switch, the feedforward value is linearly determined within a preset correction range. In one exemplary embodiment, the correction range can be set to within ±50% of the steady-state fuel reference value before the switch.

[0104] It should be noted that the first dynamic feedforward fuel correction amount takes effect within a preset time window after the mode switch occurs, and the time window is used to limit the duration of the feedforward compensation. In one exemplary embodiment, the time window can be set to within 5 seconds after the mode switch occurs.

[0105] Optionally, in one possible implementation, the final feedforward control quantity further includes a second dynamic feedforward fuel correction quantity caused by changes in operating parameters, wherein the calculation process of the second dynamic feedforward fuel correction quantity includes:

[0106] (1) The baseline dynamic feedforward gain factor is calculated based on the flight environment parameters and system load parameters.

[0107] Specifically, the change in operating parameters refers to the change in fuel demand caused by abrupt changes in flight environment parameters or system load parameters. The flight environment parameters include flight altitude and Mach number; the system load parameters include power generation and liquid cooling power. These parameters serve not only as input features to the MLP neural network model for predicting the feedforward correction factor, but also as input to the baseline dynamic feedforward model for calculating the basic feedforward compensation strength under the current operating condition.

[0108] Specifically, changes in the aforementioned parameters can lead to changes in engine intake conditions, power requirements, or thermal management needs, thereby necessitating dynamic adjustments to fuel flow. Unlike the first dynamic feedforward fuel correction caused by operating mode switching, the second dynamic feedforward fuel correction is used to respond to abrupt changes in continuous operating parameters, enabling rapid compensation for transient disturbances. The first dynamic feedforward is calculated based on changes in the target engine speed, while the second dynamic feedforward is calculated based on changes in operating parameters. Both compensate for different sources of disturbance, operating independently yet collaboratively.

[0109] Furthermore, the second dynamic feedforward fuel correction amount is calculated using a "baseline model + correction factor" method. The baseline dynamic feedforward gain factor is... The basic feedforward compensation strength under calibration conditions is calculated through a preset functional relationship; based on this, a feedforward correction factor output by the MLP neural network model is introduced. The reference gain is multiplicatively corrected to obtain the actual dynamic feedforward gain, thereby improving the adaptability to individual differences and complex operating conditions.

[0110] Specifically, the operating parameters at adjacent time points are calculated to determine whether the operating parameters have changed. When a change in operating parameters is detected, the current flight altitude, Mach number, power generation, and liquid cooling power are acquired and input into a preset functional relationship to calculate the reference dynamic feedforward gain factor.

[0111] ;

[0112] in, is the baseline dynamic feedforward gain factor, which is a dimensionless coefficient used to characterize the feedforward compensation strength under the current operating conditions. Flight altitude; It is the Mach number; Power generation capacity; This refers to the liquid cooling power; This is a preset functional relationship used to establish the mapping relationship between operating parameters and feedforward gain. In practical applications, this preset function... It can be achieved through multidimensional table lookup, interpolation calculation or piecewise function. Its core purpose is to provide appropriate feedforward compensation strength according to different working conditions.

[0113] In one exemplary embodiment, the response time of the second dynamic feedforward fuel correction amount can be set to within 2 seconds after a sudden change in operating parameters.

[0114] (2) Multiply the reference dynamic feedforward gain factor with the feedforward correction factor to obtain the actual dynamic feedforward gain factor.

[0115] Specifically, the benchmark dynamic feedforward gain factor With the feedforward correction factor output by the MLP neural network model By combining these factors, we obtain the actual dynamic feedforward gain factor:

[0116] ;

[0117] in, is the actual dynamic feedforward gain factor, which is a dimensionless coefficient and is the gain value ultimately used to calculate the second dynamic feedforward fuel correction amount; This is a feedforward correction factor used to adaptively correct the reference gain to compensate for the effects of individual engine differences or changes in operating conditions. It should be noted that the above multiplicative correction method can effectively adapt to individual differences between different engines and performance changes in the same system due to wear, aging, or calibration deviations during long-term operation. When there is a deviation between the individual characteristics of the system and the calibration reference, the MLP neural network model can learn to output a corresponding correction factor, enabling the actual gain factor to dynamically adapt to the actual response characteristics of the current engine.

[0118] (3) Determine the second dynamic feedforward fuel correction amount based on the fuel flow rate at the previous stable moment and the actual dynamic feedforward gain factor.

[0119] Specifically, the actual dynamic feedforward gain factor fuel flow rate compared to the previous steady-state time Multiplying these together yields the second dynamic feedforward fuel correction amount:

[0120] ;

[0121] in, This is the second dynamic feedforward fuel correction amount, used to compensate for transient fuel demand changes caused by sudden changes in operating parameters; The fuel flow rate at the previous stable moment, i.e., the actual measured fuel flow rate when the system is in steady-state operation before the sudden change in operating parameters, is used as a reference benchmark for dynamic compensation. It should be noted that the second dynamic feedforward fuel correction amount takes effect within a preset time window after detecting a sudden change in operating parameters, in order to achieve a rapid response to changes in fuel demand. In one exemplary embodiment, the time window can be set to within 2 seconds after the sudden change occurs.

[0122] In this embodiment, when flight environment parameters or system load parameters change abruptly, the feedforward compensation intensity can be dynamically calculated according to the current operating conditions, and the compensation amount can be adaptively adjusted by the correction factor output by the MLP neural network model, thereby achieving adaptation to the individual differences of different engines and improving the accuracy and robustness of dynamic feedforward compensation.

[0123] It should be noted that the operating modes, operating parameters, and control stages involved in this application correspond to different levels of system state description. Specifically, the operating modes characterize the task operation state of the power and thermal management system and are discrete state variables, such as steady-state mode, main start mode, maintenance mode, and cooling mode; the operating parameters characterize the flight environment and system load conditions and are continuously changing physical quantities, such as flight altitude, Mach number, power generation, and liquid cooling power; the control stages characterize the execution stage of the specific control strategy adopted under a specific operating state, such as the acceleration closed-loop control stage and the speed closed-loop control stage. Based on this, switching operating modes typically causes discrete changes in the system's target speed, corresponding to the first dynamic feedforward fuel correction; changes in operating parameters reflect continuous disturbances in the environment and load, corresponding to the second dynamic feedforward fuel correction. These two types of dynamic feedforward compensate for disturbances from different sources and, combined with the feedforward correction factor output by the MLP neural network model, adaptively correct the baseline feedforward, thus forming a complete dynamic feedforward control mechanism.

[0124] Furthermore, the final feedforward control quantity is composed of both steady-state feedforward fuel flow and dynamic feedforward fuel correction. The dynamic feedforward fuel correction includes a first dynamic feedforward fuel correction caused by operating mode switching and a second dynamic feedforward fuel correction caused by changes in operating parameters. Specifically, the dynamic feedforward fuel correction is the sum of the first and second dynamic feedforward fuel corrections.

[0125] ;

[0126] The total feedforward fuel flow rate is the sum of the steady-state feedforward fuel flow rate and the dynamic feedforward fuel correction amount:

[0127] ;

[0128] in, For total feedforward fuel flow, For steady-state feedforward fuel flow, This is the dynamic feedforward fuel correction amount. This is the first dynamic feedforward fuel correction amount. This is the second dynamic feedforward fuel correction amount.

[0129] By employing the above method, steady-state fuel demand is integrated with dynamic fuel compensation caused by mode switching and changes in operating conditions, resulting in a final feedforward control quantity. This final quantity is then combined with the feedback control quantity to generate fuel control commands. This embodiment achieves unified fusion control of steady-state and dynamic feedforward, enabling the system to maintain an accurate fuel supply benchmark under steady-state conditions and to quickly respond to changes in fuel demand during mode switching or sudden changes in operating conditions. This improves the system's dynamic response performance and transient control accuracy at the open-loop control level.

[0130] S104. Determine the current control stage based on the relationship between the current speed and the state value of the control stage switching endpoint. Select a target PID control parameter from the feedback control PID parameters of the control stage based on the current control stage. Calculate the final feedback control quantity corresponding to the real-time operating data based on the target PID control parameter.

[0131] Specifically, the engine fuel feedback control employs a phased control strategy, with different control objectives and PID control parameters corresponding to different control stages. The control stages include at least a startup stage, an acceleration closed-loop PID control stage, and a speed closed-loop PID control stage. These control stages are distinguished by the state values ​​of the control stage switching endpoints. These control stage switching endpoint state values ​​are output by the MLP neural network model and are used to determine the critical switching conditions between the acceleration closed-loop PID control stage and the speed closed-loop PID control stage.

[0132] Specifically, the current engine speed is acquired and compared with the control phase switching endpoint state value. When the current speed is less than a preset start-up phase upper limit, it is determined that the engine is currently in the start-up phase; when the current speed is greater than the preset start-up phase upper limit but less than the control phase switching endpoint state value, it is determined that the engine is currently in the acceleration closed-loop PID control phase; when the current speed is greater than or equal to the control phase switching endpoint state value, it is determined that the engine is currently in the speed closed-loop PID control phase. In one exemplary embodiment, the preset start-up phase upper limit can be set to 10% of the engine's rated speed.

[0133] Furthermore, after determining the current control stage, target PID control parameters corresponding to the current control stage are selected from the control stage feedback control PID parameters output by the MLP neural network model. These control stage feedback control PID parameters include at least the proportional and integral coefficients for the acceleration closed-loop PID control stage, and the proportional and integral coefficients for the speed closed-loop PID control stage.

[0134] Furthermore, after determining the target PID control parameters, a feedback control error is constructed based on the control objective corresponding to the current control stage. The control objective is determined by a preset control objective or updated by the output of an MLP neural network model under specific operating conditions. Specifically, in the acceleration closed-loop PID control stage, the control objective is the acceleration target value; in the steady-state closed-loop PID control stage, the control objective is the engine speed target value. Based on the deviation between the control objective and the actual value, the feedback control quantity is calculated using the target PID control parameters. Specifically, in the acceleration closed-loop PID control stage, the deviation between the target acceleration and the actual acceleration is used as the control error, and the feedback control quantity is calculated using the acceleration closed-loop PID control parameters to ensure the engine quickly and safely passes through the resonance zone and low-speed zone; in the engine speed closed-loop PID control stage, the deviation between the target engine speed and the current engine speed is used as the control error, and the feedback control quantity is calculated using the steady-state closed-loop PID control parameters to ensure engine speed control accuracy and meet power generation and bleed air requirements.

[0135] It should be noted that the aforementioned phased control strategy is primarily applied in the self-starting mode. In this mode, fuel control is divided into a start-up phase, an acceleration closed-loop PID control phase, and a speed closed-loop PID control phase based on engine speed. When the system is in main start-up mode, maintenance mode, or cooling mode, the system has typically completed the self-starting process and entered or is close to a steady-state operating state. At this time, fuel control adopts a speed closed-loop control strategy, that is, the PID controller makes feedback adjustments based on the deviation between the target speed and the actual speed to ensure speed control accuracy and system stability. In this operating mode, the control process no longer involves switching between the start-up phase and the acceleration closed-loop PID control phase.

[0136] In this way, the MLP neural network model outputs the state values ​​of the control stage switching endpoints and the PID control parameters of each stage in a unified manner, realizing the coordinated optimization of control stage division and control parameters. Furthermore, the state machine-based segmented control strategy enables adaptive switching between different operating stages, thereby improving the control stability, response speed, and robustness of the system under non-stationary conditions.

[0137] S105. Combine the final feedforward control quantity and the final feedback control quantity to generate a fuel quantity control command output.

[0138] Specifically, the final feedforward control variable and the final feedback control variable are combined to generate the fuel flow control command. Its mathematical expression is as follows:

[0139] ;

[0140] in, This is a fuel quantity control command. For the final feedforward control quantity, This is the final feedback control variable.

[0141] Furthermore, the synthesized fuel quantity control command is constrained to ensure it meets the engine's physical boundary conditions and dynamic safety requirements. Specifically, the fuel quantity control command is first limited to a preset range between minimum and maximum fuel supply to avoid unstable combustion caused by fuel shortages or over-fueling. Then, the rate of change of the fuel command is limited to restrict the variation in fuel flow between adjacent control cycles, preventing compressor surge or turbine overheating risks caused by sudden changes in fuel flow. After these constraints, the final fuel quantity control command is output to the fuel metering device, which adjusts the opening of the fuel metering valve according to the control command, thereby controlling the fuel flow into the combustion chamber and achieving real-time adjustment of engine thrust or power.

[0142] It should be noted that each processing step of the above fuel control method is executed cyclically within each control cycle. In one exemplary embodiment, the control cycle is 20ms. Within each control cycle, the system collects engine and flight condition data in real time, updates control parameters based on the condition data using an MLP neural network model, calculates feedforward control and feedback control quantities respectively, and outputs fuel quantity control commands after synthesizing and constraining the two, thereby achieving real-time closed-loop regulation of fuel supply.

[0143] Optionally, in one possible implementation, after generating the fuel quantity control command output, the method further includes:

[0144] (1) Obtain real-time operating data after the fuel quantity control command is implemented.

[0145] Specifically, after the fuel quantity control command is output to the fuel metering device and executed, real-time engine operating status data is continuously collected. This real-time operating data includes at least key status parameters used for control phase determination and control parameter updates, such as the current engine speed and compressor outlet pressure. Furthermore, based on the rate of change of real-time operating data in adjacent cycles and the real-time operating data itself, it is determined whether a change has occurred in the current phase.

[0146] (2) Based on the comparison between the real-time operating data and the control phase switching endpoint, determine whether the current control phase has changed.

[0147] Specifically, the current engine speed is compared with the state value of the control phase switching endpoint, and a comprehensive judgment is made in conjunction with the current control mode or target speed change to determine whether the current control phase has changed. When the real-time speed crosses the switching endpoint state value, it is determined that the control phase has switched; otherwise, it is determined that the control phase remains unchanged. Preferably, the rate of change and real-time operating data are comprehensively judged. The rate of change is used to assess the stability level of the real-time operating data on the one hand, and to assess the magnitude relationship between the real-time operating data of the next cycle and the control phase switching endpoint on the other hand. Therefore, if the real-time speed does not exceed the switching endpoint state value, the stability level is determined based on the rate of change, the real-time operating data of the next cycle is estimated based on the rate of change, and the magnitude relationship between the real-time operating data of the next cycle and the control phase switching endpoint is used to determine whether the control phase will change in the next cycle. The current control phase is determined based on the comprehensive judgment results of the two judgments.

[0148] It should be noted that, compared to the hard threshold judgment method based solely on the current speed and the state value of the switching endpoint, this embodiment introduces the rate of change and the prediction results of the next cycle's operating conditions to assist in the judgment of the control stage, which can comprehensively consider the trend of operating condition changes and the stability characteristics of the system. Specifically, when the operating condition is close to the switching boundary but has not yet stably crossed the threshold, possible stage changes are judged in advance by trend identification, thereby avoiding erroneous switching of the control stage and reducing switching jitter caused by threshold critical fluctuations; at the same time, the stability level of the operating condition is evaluated by combining the rate of change, reducing the frequency of unnecessary stage switching judgment updates during the stable operating phase. Through the above methods, the control stage switching not only depends on the instantaneous state, but also introduces the prediction information of future trends, thereby improving the continuity and stability of the control stage switching, reducing the control output fluctuations caused by stage switching, and improving the control accuracy and dynamic response performance of the system under non-stationary operating conditions.

[0149] (3) If the current control stage does not change, the target PID control parameter of the current control round is fed back from the control stage control PID parameter.

[0150] Specifically, when the control phase remains unchanged, the target PID control parameters corresponding to the current control cycle are directly used for feedback control calculation, without needing to re-call the MLP neural network model, thereby maintaining the stability of the control parameters.

[0151] (4) If the current control stage changes, the real-time operating condition data is input into the MLP neural network model for re-solution.

[0152] Specifically, when changes occur in the control phase, the MLP neural network model regenerates the state values ​​of the control phase switching endpoints and the corresponding PID control parameters based on the current real-time operating data, thus achieving online updates of the control parameters. This event-triggered update mechanism ensures that control parameters are refreshed only when changes occur in the control phase, reducing unnecessary model calculations while maintaining control performance and improving the system's real-time response capability and operational stability.

[0153] It should be noted that by replacing periodic recalculation with event-triggered methods, the frequency of calling the MLP neural network model is reduced, thereby reducing the consumption of computing resources. At the same time, the PID parameters are kept stable during the control phase to avoid fluctuations in control output caused by frequent parameter changes. Furthermore, the control parameters are updated in a timely manner when the control phase switches, enabling the system to adapt to the control requirements under different operating conditions, thereby achieving a balance between computational efficiency and control performance.

[0154] This application introduces a multilayer perceptron (MLP) neural network to model and analyze multi-source operating condition information of the power and thermal management system, enabling unified adaptive optimization and adjustment of key parameters of the fuel control law. Compared to traditional fixed parameter or empirical tuning methods, this application can update control parameters online based on changes in flight environment, engine status, and thermal management load, thereby improving the adaptability of the control strategy to complex operating conditions. Furthermore, this application integrates feedforward control and feedback control, enabling feedforward control to respond quickly to changes in operating conditions, while feedback control is used to eliminate steady-state errors. This improves the dynamic response speed of the system while ensuring steady-state control accuracy, reducing the risk of overshoot and control deviation. In addition, by using the control stage switching endpoint state values ​​based on the output of the MLP neural network, adaptive division of the control stage is achieved, allowing acceleration closed-loop control and speed closed-loop control to switch smoothly under different operating conditions. This avoids the lag and abrupt changes caused by traditional fixed threshold switching methods, thereby improving the control stability and robustness of the system under complex flight envelopes and variable thermal management loads. In summary, this application improves the adaptive capability, dynamic response performance, and operational stability of the fuel control system while ensuring control accuracy.

[0155] Corresponding to the aforementioned embodiment of a fuel control law optimization method for a power and thermal management system, this application also provides an embodiment of a fuel control law optimization device for a power and thermal management system.

[0156] Figure 3 This is a schematic diagram of the structure of an embodiment of the fuel control law optimization device for the power and thermal management system provided in this application. Please refer to... Figure 3 The apparatus provided in this embodiment includes an acquisition module 301, a prediction module 302, and a calculation module 303, wherein:

[0157] The acquisition module 301 is used to acquire real-time operating data of the power and thermal management system.

[0158] The prediction module 302 is used to input the real-time operating condition data into the MLP neural network model, predict the feedback control PID parameters, the switching endpoint state value and the feedforward correction factor in the control stage. The feedforward correction factor is used to correct individual differences in the power and thermal management system. The control stage includes at least one of the speed closed-loop PID control stage and the acceleration closed-loop PID control stage.

[0159] The calculation module 303 is used to correct the feedforward base quantity based on the feedforward correction factor to obtain the final feedforward control quantity.

[0160] The calculation module 303 is used to determine the current control stage based on the relationship between the current speed and the state value of the control stage switching endpoint, select a target PID control parameter from the feedback control PID parameters of the control stage according to the current control stage, and calculate the final feedback control quantity corresponding to the real-time operating data based on the target PID control parameter.

[0161] The calculation module 303 is used to generate a fuel quantity control command output based on merging the final feedforward control quantity and the final feedback control quantity.

[0162] The apparatus of this embodiment can be used to perform... Figure 1 The steps of the method embodiment shown are similar in principle and process, and will not be repeated here.

[0163] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0164] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0165] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for optimizing fuel control laws in a power and thermal management system, characterized in that, include: Acquire real-time operating data of the power and thermal management system; The real-time operating data is input into the MLP neural network model to predict the feedback control PID parameters of the control stage, the switching endpoint state value of the control stage, and the feedforward correction factor. The feedforward correction factor is used to correct individual differences in the power and thermal management system. The control stage includes at least one of the speed closed-loop PID control stage and the acceleration closed-loop PID control stage. The feedforward base quantity is corrected based on the aforementioned feedforward correction factor to obtain the final feedforward control quantity; Based on the relationship between the current speed and the state value of the control stage switching endpoint, the current control stage is determined. Based on the current control stage, a target PID control parameter is selected from the feedback control PID parameters of the control stage. Based on the target PID control parameter, the final feedback control quantity corresponding to the real-time operating data is calculated. The final feedforward control quantity and the final feedback control quantity are combined to generate a fuel quantity control command output. After generating the fuel quantity control command output, it also includes: Obtain real-time operating data after the fuel quantity control command is implemented; Based on the comparison between the real-time operating data and the control phase switching endpoint, it is determined whether the current control phase has changed. If the current control phase does not change, the target PID control parameter for the current control round is fed back from the control phase control PID parameter. If the current control stage changes, the real-time operating condition data is input into the MLP neural network model for re-solution; The MLP neural network model is trained in the following way: Select multiple operating points within the preset flight envelope; Based on the fully digital simulation model, for each operating point, the optimization algorithm is used to optimize the PID parameters of the feedback control in the control stage, the state values ​​of the switching endpoints in the control stage, and the feedforward correction factor, so as to minimize the preset loss function and obtain the corresponding optimal parameter set. Using the working condition feature parameters corresponding to each working condition point as input, and the optimal parameter set as label, a training sample set is constructed. The MLP neural network model is trained using the training sample set under supervised learning conditions. The preset loss function is composed of a weighted average of multiple control performance indicators, which at least include response time, overshoot, surge margin loss, exhaust temperature over-limit penalty, and steady-state error; wherein... The response time is the time elapsed from the issuance of the control command to the engine speed reaching the preset proportion of the target speed; The overshoot is the maximum deviation of the engine speed from the target speed. The surge margin loss is the deviation of the actual surge margin from the minimum allowable surge margin; The exhaust temperature over-limit penalty is the penalty amount corresponding to when the power turbine outlet temperature exceeds the preset temperature limit; The steady-state error is the absolute value of the deviation of the engine speed from the target speed in a steady state; The loss function value is obtained by multiplying each of the control performance indicators by its corresponding weight coefficient and then summing the results.

2. The method according to claim 1, characterized in that, The MLP neural network model takes flight environment parameters, engine status parameters and thermal management system status parameters as inputs, and outputs PID parameters for feedback control in the control phase, state values ​​of switching endpoints in the control phase and feedforward correction factors. The MLP neural network model includes an input layer, at least one hidden layer, and an output layer; The flight environment parameters include at least flight altitude and Mach number, the engine status parameters include at least engine speed and compressor outlet pressure, and the thermal management system status parameters include at least one or more of the following: power generation, liquid cooling power, and bleed air load.

3. The method according to claim 1, characterized in that, The supervised learning training of the MLP neural network model based on the training sample set further includes: Calculate the change between the parameter values ​​of the current working condition and the parameter values ​​of the previous working condition, wherein the change includes the change value and the change direction; A change description is generated based on the change value and the change direction; The description of the change is used as a prompt and input into the MLP neural network model along with the training sample set; The MLP neural network model makes a first direction prediction based on the input in the training sample set; The solution space is defined based on the difference between the first direction and the direction of change; Predict the changed operating condition parameter values ​​in the solution space, and calculate the output of the MLP neural network model based on the changed operating condition parameter values.

4. The method according to claim 1, characterized in that, The final feedforward control quantity includes the steady-state feedforward fuel flow rate, and the method for obtaining the steady-state feedforward fuel flow rate includes: Obtain engine operating status parameters under the current steady-state operating conditions. The engine operating status parameters include at least the compressor outlet pressure, engine inlet total temperature, and current speed. The steady-state feedforward fuel flow rate is calculated based on the compressor outlet pressure, engine inlet total temperature, and current speed using a preset functional relationship.

5. The method according to claim 1, characterized in that, The final feedforward control quantity also includes a first dynamic feedforward fuel correction quantity caused by the switching of the operating mode, and the method for obtaining the first dynamic feedforward fuel correction quantity includes: When an engine operating mode switch is detected, the target speed change before and after the operating mode switch and the steady-state fuel reference value before the switch are obtained. Based on the target speed change and the steady-state fuel reference value before switching, the first dynamic feedforward fuel correction amount is calculated through a preset functional relationship.

6. The method according to claim 1, characterized in that, The final feedforward control quantity also includes a second dynamic feedforward fuel correction quantity caused by changes in operating parameters. The calculation process for the second dynamic feedforward fuel correction quantity includes: The baseline dynamic feedforward gain factor is calculated based on flight environment parameters and system load parameters. Multiply the reference dynamic feedforward gain factor by the feedforward correction factor to obtain the actual dynamic feedforward gain factor; The second dynamic feedforward fuel correction amount is determined based on the fuel flow rate at the previous stable moment and the actual dynamic feedforward gain factor.

7. A fuel control law optimization device for a power and thermal management system, characterized in that, The device is used to execute the fuel control law optimization method for the power and thermal management system according to any one of claims 1-6, and the device includes an acquisition module, a prediction module, and a calculation module, wherein: The acquisition module is used to acquire real-time operating data of the power and thermal management system; The prediction module is used to input the real-time operating condition data into the MLP neural network model, predict the feedback control PID parameters, the switching endpoint state values ​​and the feedforward correction factor in the control stage. The feedforward correction factor is used to correct individual differences in the power and thermal management system. The control stage includes at least one of the speed closed-loop PID control stage and the acceleration closed-loop PID control stage. The calculation module is used to correct the feedforward base quantity based on the feedforward correction factor to obtain the final feedforward control quantity. The calculation module is used to determine the current control stage based on the relationship between the current speed and the state value of the control stage switching endpoint, select a target PID control parameter from the feedback control PID parameters of the control stage based on the current control stage, and calculate the final feedback control quantity corresponding to the real-time operating data based on the target PID control parameter. The calculation module is used to merge the final feedforward control quantity and the final feedback control quantity to generate a fuel quantity control command output.

Images