A neural network-based hybrid engine driving switching control optimization design method

Abstract

A hybrid engine driving switching control optimization design method based on neural network, comprising the following steps: step one: mechanism modeling of turboshaft engine; step two: mechanism modeling of electric motor; step three: based on the output parameters of the engine and the electric motor, the engine and the electric motor are realized to work cooperatively; step four: building BP neural network, output mode instruction and torque distribution ratio; step five: programming GA genetic algorithm and optimizing HEPS output parameters; the BP neural network responds to real-time working conditions quickly, the GA algorithm is calibrated globally periodically, and the cascade PI controller is corrected in real time, so that the performance of the HEPS is stable under the scene of load mutation and working condition switching. Under the premise of ensuring the safety margin of the system, the switching impact is effectively reduced, the switching smoothness is improved, and the multi-objective optimization of fuel economy, emission index and smoothness is realized.

Description

Technical Field

[0001] This invention belongs to the field of hybrid engine drive switching control technology, and specifically relates to an optimized design method for hybrid engine drive switching control based on neural networks. Background Technology

[0002] Limited by current technological bottlenecks in battery energy density and electric motor performance, existing battery technology cannot yet support the mission requirements of all-electric flight for high-thrust aircraft. Currently, the energy density of commercial lithium batteries can reach 200Wh / kg, with some laboratory levels reaching 500Wh / kg, while the energy density of fuel is 13kWh / kg, dozens of times higher than that of batteries. Therefore, all-electric propulsion systems cannot completely replace traditional fuel propulsion systems and are only applied to some small general aviation electric aircraft, failing to achieve a leapfrog innovation in propulsion systems. Consequently, pure electric aircraft struggle to overcome technological barriers in real-world scenarios such as long-distance, long-range flights, and cannot meet the diverse requirements of aircraft for range, payload, and reliability. Against this backdrop, hybrid electric propulsion systems (HEPS) have emerged as an innovative and practical aviation power solution.

[0003] HEPS (Hybrid Power Assist System) typically consists of a power supply system, a propulsion system, and an energy storage system. This coupled design of "traditional aircraft engine + electric system" inevitably increases the complexity of the original power system, thus placing stringent requirements on the rationality and accuracy of the control method. Existing HEPS control methods still have significant technical shortcomings: First, speed / torque matching deviations are prone to occur during mode switching, leading to power connection gaps or switching shocks (such as power fluctuations caused by sudden torque changes when the motor switches from generator mode to assist mode). Second, there is insufficient adaptability to sudden load changes; when flight conditions change abruptly (such as rapid climb or instantaneous acceleration), the adjustment response of the fuel-electric power distribution ratio is delayed or excessively large, easily leading to fluctuations in total thrust. Third, existing technologies do not consider the synergistic optimization of fuel consumption, emissions, and smoothness, often sacrificing one performance aspect for the optimal performance of another. For example, reducing fuel consumption to limit motor assist amplifies switching shocks and thrust fluctuations, creating a chain reaction of "single optimization leading to overall imbalance." These shortcomings not only reduce flight smoothness and increase operating costs but may also threaten flight safety.

[0004] Existing HEPS mode switching control systems mostly employ PID control algorithms, triggering switching through preset speed / torque thresholds. The power distribution ratio is calculated based on empirical formulas, but the fixed PID parameters cannot adapt to sudden load changes. Furthermore, the empirical formulas do not consider the coordination between fuel consumption and smoothness, resulting in larger switching shocks, higher fuel consumption, and increased emissions. Summary of the Invention

[0005] To overcome the above technical problems, the present invention aims to provide an optimized design method for hybrid engine drive switching control based on neural networks, which effectively reduces switching impact and improves switching smoothness while ensuring system safety margin, and achieves multi-objective optimization of fuel economy, emission indicators and smoothness.

[0006] To achieve the above objectives, the technical solution adopted by the casting system of the present invention is as follows:

[0007] An optimization design method for hybrid engine drive switching control based on neural networks, characterized by the following steps;

[0008] Step 1: Perform mechanism modeling of the turboshaft engine and output the engine's speed, temperature, and fuel consumption parameters;

[0009] Specifically, it provides three core parameters (gas turbine speed, power turbine speed, and exhaust temperature) out of the six inputs to the BP neural network in step four; and provides the basis for multi-objective optimization calculations of fuel consumption and emissions for the GA genetic algorithm in step five.

[0010] Each equation specifically corresponds to: the gas turbine speed and power turbine speed output from the gas turbine / power turbine rotor power balance equation; the combustion chamber outlet temperature (exhaust temperature) output from the combustion chamber outlet enthalpy-temperature conversion equation; the fuel consumption index output from the fuel consumption rate (BSFC) equation; and the NOx output from the emission index equation. x Emission parameters.

[0011] Step 2: Perform motor mechanism modeling and output the parameters of motor torque and speed constraints;

[0012] This step establishes a mapping relationship between the motor's control commands and power output, providing physical constraints on the motor side for torque distribution optimization and ensuring that control commands can be implemented.

[0013] Each equation specifically corresponds to: the electromagnetic torque equation, which outputs the real-time torque and upper limit of the motor; the motor motion equation, which outputs the motor speed and dynamic speed constraints; the three-phase voltage / magnetic flux equation, which outputs electromagnetic characteristic constraints; and the power equation, which outputs energy efficiency and power boundary constraints.

[0014] Step 3: Based on the output parameters of the engine and electric motor, the engine and electric motor work together by controlling the coordinated distribution and mode switching of oil and electric power.

[0015] Step 4: Build a BP neural network to output mode commands and torque distribution ratios;

[0016] Step 3 provides control decisions and command inputs. Based on the mode command and torque distribution ratio output from Step 4, Step 3 executes the coordinated distribution of oil and electric power and mode switching control. The two constitute a coordinated control relationship of decision-making and execution.

[0017] The purpose of this step is to construct a nonlinear mapping model between system state parameters and control commands, quickly generate drive modes and torque distribution decisions that adapt to real-time operating conditions, and achieve adaptive and precise control in hybrid operating conditions.

[0018] The purpose of this step is to output mode commands and torque distribution ratios in real time, provide core control parameters for drive switching, and provide a high-quality initial solution for subsequent global optimization.

[0019] Specifically, this manifests in the following ways: 1. Rapid response to operating conditions (outputting control commands in real time based on state parameters (speed, load, SOC, temperature, etc.), solving the problems of slow response and poor adaptability of traditional control). 2. Providing core control output (directly outputting mode commands and torque distribution ratios, serving as the decision-making core for the entire hybrid switching control). 3. Fitting nonlinear characteristics (handling the strong coupling and nonlinear characteristics of the HEPS system, ensuring control accuracy and smooth switching). 4. Providing initial solutions for global optimization (the output results serve as the optimization basis for the GA genetic algorithm, avoiding the algorithm from getting trapped in local optima).

[0020] Step 5: Global optimization using the GA genetic algorithm directly serves the hybrid engine drive switching control. The optimized target speed of the gas turbine, the ratio of oil and electric power distribution, and the mode switching trigger threshold are all core control parameters for hybrid engine drive switching, ensuring that the drive switching process is shock-free, fast-responding, and achieves optimal fuel consumption and emissions, thus realizing multi-objective optimization control of hybrid engine drive switching.

[0021] Step one specifically involves:

[0022] 1. Gas turbine power balance equation

[0023]

[0024] Indicates the output power of the gas turbine. Indicates the compressor drive power. Gas turbine mechanical loss power;

[0025] 2. Gas turbine rotor power balance equation:

[0026]

[0027] This represents the moment of inertia of the gas turbine rotor. Indicates the rotational speed of the gas turbine rotor. This represents the angular acceleration of the gas turbine rotor. Represents the deviation term in the equation;

[0028] 3. Power balance equation for a power turbine:

[0029]

[0030] This represents the total power absorbed by the system (which must satisfy both load demand and energy loss). This indicates the power demand of the load (such as the propulsion power demand of an electric motor driving a propeller). Indicates system energy loss;

[0031] 4. Power balance equation for the power turbine rotor:

[0032]

[0033] Indicates the output power of the power turbine. This represents the moment of inertia of the power turbine rotor. Indicates the rotational speed of the power turbine rotor. This represents the angular acceleration of the power turbine rotor. Represents the deviation term in the equation;

[0034] 5. The enthalpy at the combustion chamber outlet is:

[0035]

[0036] Indicates the enthalpy at the combustion chamber outlet. Indicates fuel flow rate. Indicates the calorific value of fuel oil. Indicates combustion chamber efficiency. Indicates the airflow rate at the combustion chamber inlet. Indicates the enthalpy of the air at the combustion chamber inlet. This indicates the total flow rate of gas exiting the combustion chamber;

[0037] 6. The combustion chamber outlet temperature is:

[0038]

[0039] Indicates the total temperature at the combustion chamber outlet. This represents the enthalpy-temperature conversion function, used to calculate the corresponding temperature based on a given oil-to-gas ratio and enthalpy value. Indicates the air-fuel ratio at the combustion chamber outlet;

[0040] 7. Fuel Consumption Rate (BSFC):

[0041]

[0042] This represents the engine's output power. This equation quantifies the fuel consumption per unit power output and is a core quantitative indicator for optimizing fuel economy.

[0043] 8. Emissions Index Formula

[0044]

[0045] NO represents the actual working conditions x Emission index, Indicates NO under standard ground conditions x Emission index;

[0046] Through the interrelation and constraints of the above formulas, a complete mechanism model of the turboshaft engine is formed; the power balance equation and rotor dynamics equation determine the dynamic relationship between engine speed, torque, and power; the combustion chamber outlet enthalpy and temperature equation determines the engine's thermal state and safety boundaries; the fuel consumption rate equation quantifies economic indicators; and the emission index equation quantifies environmental indicators. The final outputs are engine speed, exhaust temperature, fuel consumption, and NOx. x Emission parameters serve as the basis for judging engine operating status, making mode switching decisions, and allocating torque in hybrid engine drive switching control.

[0047] Step two specifically involves:

[0048] The motor is selected as a three-phase permanent magnet AC synchronous motor. In a three-phase permanent magnet AC synchronous motor, the voltage can be decomposed into three-phase voltage vectors.

[0049] 1. Equations of the three-phase voltage vector in the natural coordinate system:

[0050]

[0051] In the formula: This refers to the three-phase voltage of the motor. It is a three-phase resistor. It is a three-phase current. For three-phase winding flux linkage;

[0052] Phase separation after disassembly

[0053]

[0054] 2. Three-phase winding flux linkage equation:

[0055]

[0056] In the formula Inductance for a three-phase winding; For three-phase winding flux linkage;

[0057] 3. Phase-separated flux linkage function

[0058]

[0059] This function simplifies the calculation of flux linkage equations, clarifies the independent calculation relationship of flux linkage in each phase, and simplifies the formula of flux linkage equation for three-phase windings.

[0060] 4. The electromagnetic torque equation of an electric motor:

[0061]

[0062] The relationship between quantified magnetic flux and electromagnetic torque represents the power output of the motor;

[0063] 5. Equation of motion of an electric motor:

[0064]

[0065]

[0066]

[0067]

[0068] In the formula: Let be the mechanical angular velocity of the electric motor. This represents the load torque of the motor. This is the damping coefficient of the motor. Let be the moment of inertia of the motor. Electromagnetic angular velocity, This refers to the motor speed;

[0069] 6. Equations for the input and output power of an electric motor:

[0070]

[0071]

[0072] The energy conversion efficiency of the motor (input electrical power → output mechanical power) is quantified, including loss calculation, to ensure that the torque distribution ratio optimized by the neural network meets the energy efficiency requirements;

[0073] Through the mutual coupling and constraints of the above formulas, a mechanistic model of a three-phase permanent magnet synchronous motor is constructed: the voltage equation and flux linkage equation clarify the electromagnetic constraints of the motor; the electromagnetic torque equation establishes the relationship between flux linkage and power output; the motion equation determines the dynamic response law of speed and torque; and the input and output power equation quantifies the energy conversion efficiency. The final output parameters—motor torque, motor speed, power constraints, and dynamic response characteristics—provide the motor-side power output boundary and execution basis for the hybrid engine drive switching control.

[0074] Step three specifically involves:

[0075] The generator converts the mechanical energy output by the turboshaft engine into electrical energy, which is then transmitted to the electric motor via a power bus. The generator is considered an auxiliary component of the engine, and the two together form a turbine power generation system. The generator's design speed is different from that of the engine, hence the following:

[0076] Based on the engine speed, power, and temperature parameters output in step one, and the motor torque, speed, and power constraint parameters output in step two, engine-motor power coupling and coordinated control are performed to achieve the underlying execution and stable operation of hybrid engine drive switching.

[0077] The generator converts the mechanical energy output by the turboshaft engine into electrical energy, which is then transmitted to the electric motor via a power bus. The generator serves as an auxiliary component of the engine, and the two together form a turbine power generation system. Since the generator's design speed does not match the engine's power shaft speed, a generator speed-torque matching equation is introduced to achieve a coordinated match between speed and torque. This equation takes the engine dynamics characteristics from step one as input and the electromagnetic and dynamic constraints of the electric motor from step two as output boundaries, efficiently converting the engine's mechanical energy into usable electrical energy for the electric motor. This provides a power transmission basis for the coordinated operation of the engine and electric motor and the switching of hybrid drive.

[0078] Step three specifically involves:

[0079] 1. Generator speed-torque matching equation:

[0080]

[0081]

[0082] In the formula: The speed at the output of the reducer. The speed at the input end of the reducer. The output torque of the reducer. The input torque of the reducer. The reduction ratio of the reducer. This is for the mechanical efficiency of the reducer. The input end is connected to the engine power shaft, and the output end is connected to the generator working shaft.

[0083] All-electric operating mode

[0084]

[0085] Charging mode

[0086]

[0087] 2. The cascaded PI controller takes the following form;

[0088]

[0089] The cascaded PI controller is the core execution unit for "real-time correction and smooth transition" during HEPS mode switching. Its control parameters and target values ​​are dynamically optimized through a BP neural network and a GA genetic algorithm, forming a closed-loop collaborative control chain of "GA+BP+PI". The specific interaction logic is as follows:

[0090] Outer loop (speed loop): The "gas turbine target speed" obtained by global optimization using the GA genetic algorithm is used as the outer loop setpoint; the feedback value of the outer loop is the real-time speed of the engine power turbine; the output result of the outer loop is used as the upper limit of the torque reference of the inner loop (torque loop), ensuring that the speed is stable is the premise of torque distribution;

[0091] Inner loop (torque loop): It consists of two parts based on the inner loop set value: 1. Torque distribution ratio output by BP neural network; 2. Compensation torque for speed deviation of outer loop; The feedback value of the inner loop is the real-time output torque of the motor, which is finally adjusted by PI to output the motor torque control signal to achieve accurate torque tracking.

[0092] The BP neural network takes the load demand power change rate, battery SOC, and exhaust temperature as input features, and outputs PI parameter correction coefficients adapted to the current operating conditions through a trained nonlinear mapping model, and updates the PI controller parameters in real time.

[0093] Among them, the load demand power change rate is calculated by combining the motor motion equation and input / output power equation in the motor mechanism modeling in step two with the real-time flight condition load demand, reflecting the dynamic change of the motor side load.

[0094] The battery SOC is calculated in real time by the hybrid power system energy storage battery module using the coulomb counting method, and is used to determine the battery's charging and discharging capacity and the conditions for switching drive modes.

[0095] The exhaust temperature is calculated in real time by the combustion chamber outlet temperature equation in the turboshaft engine mechanism modeling step one, reflecting the engine thermal state and safe operating boundary.

[0096] 3. Basic voltage equations for a three-phase motor in synchronous rotational coordinates along the dq axis:

[0097]

[0098] This equation transforms the motor model from a natural coordinate system to a synchronous rotating coordinate system, simplifying control calculations. At the same time, it improves the real-time performance of motor control, adapting to the fast response requirements of neural networks.

[0099] 4. SOC Calculation

[0100]

[0101] Charging limit:

[0102] SOC≤0.95;

[0103] Lower discharge limit:

[0104] SOC ≥ 0.2;

[0105] 5. Switching between different flight phases

[0106] Hybrid power system shared power supply mode;

[0107] When an aircraft is in a high-load phase such as taxiing, takeoff, or climb, and the power demand exceeds the output power of the turbine generator system, the energy storage battery pack needs to discharge to compensate for the insufficient power of the engine and keep the engine running at its highest efficiency point.

[0108]

[0109]

[0110]

[0111] To meet the power requirements of the electric motor, This refers to the generator's output power. This refers to the battery's output power.

[0112] When the aircraft is in low-to-medium load conditions such as cruise or descent, if the SOC is high and the energy storage battery pack has sufficient energy, the battery can provide power independently.

[0113] All-electric operating mode

[0114]

[0115]

[0116]

[0117] Battery charging working mode

[0118] Power required when the aircraft is in the cruise phase Power supplied by generator It is closer to but slightly smaller than the latter. In order to maintain the internal power balance of the system, the energy storage battery pack charges the battery through a bidirectional DC / DC converter.

[0119]

[0120]

[0121]

[0122] The generator speed-torque matching equation and the basic voltage equation of the three-phase motor in the synchronous rotation coordinate of the dq axis mainly provide physical constraints; the fixed physical boundary of power output is the underlying basis of all control and determines how the power energy is output.

[0123] Through the mutual coupling and constraints of the above five parts, a complete engine-motor power coupling and coordinated control mechanism is formed: the generator speed-torque matching equation uses the engine modeling parameters from step one as input and the motor constraints from step two as boundaries, laying the foundation for power transmission; the three-phase motor dq-axis voltage equation is derived based on the motor mechanism from step two, limiting the physical limits of motor power output; the cascade PI controller uses the first two parts as a reference and basis to achieve real-time correction of speed and torque; SOC calculation provides energy safety boundaries and constrains drive mode switching conditions; the mode switching of different flight stages combines operating conditions and SOC to determine the switching logic of engine drive, motor drive, and hybrid drive. Finally, stable, smooth, and safe hybrid engine drive switching control commands and torque distribution commands are output, realizing engine-motor coordinated work and smooth switching of multi-mode drive.

[0124] Specifically, this manifests as follows:

[0125] 1. Generator speed-torque matching equation

[0126] It takes the engine power turbine speed and output power output from step one as input, and the rated speed and rated torque of the motor output from step two as boundary constraints, to solve the problem of speed mismatch between the engine and generator, and stably convert the engine's mechanical energy into electrical energy, providing a speed reference and torque upper limit for all subsequent control.

[0127] 2. Basic voltage equations of a three-phase motor in synchronous rotational coordinates along the dq axis

[0128] It is derived from the three-phase voltage equation, flux linkage equation, and electromagnetic torque equation in step two, which model the motor mechanism. It clarifies the electromagnetic relationship between motor voltage, current, flux linkage, and torque, and limits the physical limits of motor power output. On the one hand, it constrains the motor-side output of the generator speed-torque matching equation; on the other hand, it provides the basis for the execution of torque commands for the cascade PI controller. All control commands must satisfy the constraints of this equation.

[0129] 3. Cascade PI controller

[0130] It receives constraints and benchmarks from the first two parts: the outer speed loop uses the target speed given by the generator speed-torque matching equation as the set value and the real-time engine speed output in step one as feedback; the inner torque loop uses the torque distribution ratio output by the BP neural network as the set value and the real-time motor torque output in step two as feedback; at the same time, the PI controller parameters are corrected in real time by the BP neural network according to the operating conditions, so as to achieve stable speed, no torque shock, and precise power tracking during the hybrid engine drive switching process.

[0131] 4. SOC Calculation

[0132] It calculates the remaining battery charge in real time using the coulomb counting method and sets the charging and discharging safety boundaries. The State of Charge (SOC) directly constrains the action of the cascade PI controller: when the SOC is too low, the PI controller limits the motor assist and increases the engine output; the SOC directly constrains the generator speed-torque matching relationship: when charging is required, the engine speed and generator output are increased according to the matching equation; the SOC is one of the core judgment conditions for mode switching, and together with the flight conditions, it determines the driving mode.

[0133] 5. Switching between different flight phases

[0134] It takes into account flight conditions, SOC status, engine exhaust temperature in step one, and motor load demand in step two to decide whether to enter hybrid power supply, all-electric, or charging mode.

[0135] The generator speed-torque matching equation resolves the contradiction between the engine and generator speed mismatch and clarifies the transmission efficiency of the engine's mechanical energy to electrical energy conversion. Is this equation the load-side input to the three-phase motor's dq-axis voltage equation, or the speed reference for the cascade PI controller?

[0136] The basic voltage equation of a three-phase motor in synchronous rotating coordinates along the dq axis quantifies the electromagnetic relationship between voltage, current, flux linkage, and speed, and clarifies the torque output limit of the motor. This equation serves as the motor-side constraint for the generator speed-torque matching equation and is also the basis for the execution of the cascade PI controller.

[0137] The cascaded PI controller and SOC calculation mainly adjust the power precisely within the constraints according to the instructions of the upper layer, while managing the battery power (SOC).

[0138] The speed loop reference of the cascade PI controller comes from the generator speed-torque matching equation, and the current command must satisfy the dq axis voltage equation. When the trigger mode changes during flight phase switching, the PI controller will dynamically adjust the parameters to adapt to the torque requirements of the new operating condition.

[0139] When calculating the State of Charge (SOC), if SOC ≥ 95%, the all-electric operating mode is triggered; if SOC ≤ 20%, the engine power supply + battery charging mode is triggered. When SOC is too low, the PI controller will limit the motor's assist torque to prioritize battery safety (related to the PI controller); the engine torque distribution ratio increases, the generator speed increases, and the charging current increases (related to the generator speed-torque matching equation).

[0140] Switching between different flight phases: Observe flight conditions and SOC, determine the working mode (hybrid / all-electric / charging), and send instructions to the lower level.

[0141] The switching between different flight phases means that the speed loop setpoint and torque distribution ratio of the PI controller will be adjusted synchronously in different modes (associated with the cascade PI controller); at the same time, SOC is one of the core thresholds for mode switching (associated with SOC); the power output in different modes must conform to the physical laws of the generator speed-torque matching equation and the dq axis voltage equation.

[0142] Step four specifically involves:

[0143] The BP neural network model consists of an input layer, hidden layers, and an output layer, with a purely linear transfer function used between the hidden layers and the output layer.

[0144] (1) Data preparation:

[0145] Prepare a dataset for training and testing the neural network; extract the component-level mechanism model of the turboshaft engine established in step one and the mechanism model of the electric motor established in step two, and combine the engine-electric motor power coupling relationship to construct the HEPS hybrid engine drive switching dynamic model. Based on the above model, extract full-condition state data to form the training set and test set required for the BP neural network.

[0146] The input parameters, namely the input layer of the BP neural network, include the turbine speed of the turboshaft engine, the power turbine speed, the power distribution ratio of oil and electricity, the load demand power, the battery SOC, and the exhaust temperature.

[0147] The output parameters, namely the output layer of the BP neural network, include engine operating mode commands, motor operating mode commands, and torque distribution ratios, which are used to control the mode switching and power output of the engine and motor.

[0148] (2) Determine the network architecture:

[0149] A BP neural network consists of an input layer, a hidden layer, and an output layer. The input layer has six parameters: the turbine speed of the turboshaft engine, the power turbine speed, the power distribution ratio between the fuel cell and electric power, the load demand power, the battery SOC, and the exhaust temperature.

[0150] After comparing the network prediction results with different numbers of nodes in the hidden layer, it was finally determined that the black box function was best fitted when there were 11 hidden layer nodes. Therefore, 11 hidden layer nodes were selected.

[0151] The output layer outputs three parameters, which correspond to the precise control of the engine and motor's working mode switching and torque distribution.

[0152] (3) Initialize weights and thresholds:

[0153] In the programming software, the rand function is used to randomly initialize the weight matrix V between the input layer and the hidden layer, the weight matrix W between the hidden layer and the output layer, as well as the hidden layer threshold b1 and the output layer threshold b2, with an initialization range of [0,1].

[0154] (4) Select activation function

[0155] The ReLU function is chosen as the activation function for the neural network, and its functional form is:

[0156] ;

[0157] in The linear input values ​​for the hidden layer neurons are, i.e. (Input parameter * corresponding weight) + threshold (b1);

[0158] (5) Forward propagation:

[0159] This involves passing input data to the network, calculating the output of each neuron through the neurons in each layer, and the propagation process of the BP neural network;

[0160] The input layer is denoted by L1, the output layer by L2, the weights between L1 and the hidden layer are denoted by V, and the matrix size is 11*6; the weights between L2 and the hidden layer are denoted by W, and the matrix size is 3*11.

[0161]

[0162] Where X is the input parameter vector (6×1) and H is the hidden layer output vector (11×1).

[0163]

[0164] Where Y is the output parameter vector (3×1);

[0165] (6) Back propagation:

[0166] The parameters are updated using a weighted multi-objective loss function, with the gradients of the weights and biases set based on priority. Backpropagation uses gradient descent or its variants to adjust the parameters to minimize the loss. Backpropagation enables the neural network to initially learn the parameter mapping relationships under multi-objective constraints, providing a high-quality initial solution for the subsequent global optimization of the GA genetic algorithm.

[0167]

[0168] Torque fluctuation loss; This is due to fuel consumption loss; This is for emissions losses.

[0169] Step five specifically involves:

[0170] The GA genetic algorithm, based on a multi-objective optimization function, obtains the optimal core control parameters, including:

[0171] 1. Target speed of the gas turbine;

[0172] 2. Power distribution ratio between oil and electricity;

[0173] 3. Mode switching trigger threshold;

[0174] Based on the GA genetic algorithm, evolutionary operators such as selection, crossover, and mutation are used to generate new combinations of design parameters and evaluate their performance to guide the search for the next generation;

[0175] (1) Definition of fitness function

[0176] In the process of HEPS multi-objective collaborative optimization, it is required that the smoothness of the power mode switching of the hybrid aero engine meets the preset threshold of aviation flight safety, while minimizing the aviation fuel consumption rate and NOx exhaust emission concentration, and optimizing the core adjustable control parameters of HEPS so that the overall operating efficiency of the hybrid aero engine under all flight conditions reaches the optimal level.

[0177] In the construction of the fitness function, conditional statements are used to pre-filter the optimization output results.

[0178] First, the pre-established mathematical model of HEPS power switching dynamics and energy consumption emission theory is imported for calculation, and the power switching smoothness corresponding to the parameter combination to be evaluated is verified first to meet the preset safety threshold requirements.

[0179] When the smoothness index meets the preset requirements, the fuel consumption rate, exhaust emission concentration and comprehensive operating efficiency corresponding to the parameter combination are further calculated, and the comprehensive operating efficiency value that integrates fuel consumption and emissions is output. In subsequent iterations, the core control input parameter corresponding to the maximum comprehensive operating efficiency is selected as the optimal solution. If the smoothness index does not meet the preset requirements, the fitness value of the parameter combination is directly set to a minimum invalid value and eliminated.

[0180] During the iterative process of finding the optimal solution, the convergence and computational accuracy of the genetic algorithm are ensured by limiting the output range of the fitness function to negative numbers.

[0181] (2) The chromosome is a column vector that stores the three core adjustable control parameters of HEPS. The three core control parameters are: the target speed of the turboshaft engine gas turbine, the power distribution ratio of oil and electricity, and the trigger threshold for engine / motor mode switching. Each element of the column vector corresponds to a specific value of a control parameter, and the value range conforms to the safety operation boundary constraints of HEPS.

[0182] Construct a zero matrix of appropriate dimensions to store the parent chromosome data during the iteration process of the genetic algorithm;

[0183] The selection operator uses the roulette wheel selection method, the crossover operator uses the single-point crossover method, and the mutation operator uses the basic bit mutation method. New combinations of design parameters are generated through evolutionary operators such as selection, crossover, and mutation, and their performance is evaluated to guide the search of the next generation. The parent population is continuously updated in multiple iterations to finally obtain the optimal combination of control parameters.

[0184] (3) Based on the global search capability of the GA genetic algorithm, the parent population is updated iteratively through selection, crossover, and mutation. On the basis of the "key state parameter-control parameter" mapping model constructed by the BP neural network, the optimal solution for multiple objectives is screened.

[0185] The multi-objective optimal solution is physically represented as a set of control parameters that enable the hybrid propulsion system to achieve the highest overall physical operating efficiency under the current flight load and battery SOC state. This optimal solution directly affects the control system through the following three core physical variables: the three core control parameters corresponding to chromosome coding, namely, the target speed of the turboshaft engine gas turbine, the fuel-electric power distribution ratio, and the engine / motor mode switching trigger threshold.

[0186] 1. Optimal target speed of turboshaft engine gas turbine: This parameter physically represents the best thermodynamic and mechanical operating state that the engine should maintain under this condition. It is the direct physical setting benchmark of the outer loop (speed loop) of the cascade PI controller. The BP neural network compares this target speed with the current actual speed to determine at the source whether to adjust the fuel supply or introduce electric motor assistance to ensure stable speed and improve gas thermal efficiency.

[0187] 2. Optimal fuel-electric power distribution ratio: This parameter quantifies the specific physical ratio of the mechanical output power of the turboshaft engine to the electromagnetic output power of the electric motor under the current load demand, guiding the BP neural network to output precise torque distribution commands; under the premise of meeting the total propulsion power of the aircraft, the characteristics of the fast torque response of the electric motor are used to reduce the dynamic torque fluctuation of the engine and maintain the engine operating point in the low fuel consumption range.

[0188] 3. Optimal mode switching trigger threshold: This parameter defines the physical critical conditions for the system to safely switch between hybrid, all-electric, and charging drive modes. It transforms real-time physical states such as load demand power, battery SOC, and exhaust temperature into clear switching limits, thus avoiding sudden changes in drive shaft torque or power interruption caused by improper switching timing at the execution level.

[0189] In summary, the multi-objective optimal solution finally obtained by the genetic algorithm is the optimal target speed of the turboshaft engine gas turbine, the optimal oil-electric power distribution ratio, and the optimal mode switching trigger threshold, which simultaneously meet the requirements of drive switching smoothness, fuel economy, and low emissions.

[0190] The specific implementation is as follows:

[0191] 1. Selection operation: The roulette wheel selection method is adopted. The parent chromosome (target speed of turboshaft engine, power distribution ratio of oil and electricity, mode switching trigger threshold) is input and substituted into the fitness function to calculate the overall operating efficiency. The selection probability of a single parent is equal to the overall operating efficiency of that parent / the sum of the overall operating efficiencies of all parents. The dominant parent is randomly selected according to the probability to ensure the inheritance of superior genes.

[0192] 2. Crossover operation: The rand function generates a crossover point of 1-2, and at the crossover point, the two sets of parent chromosomes are truncated and parameter fragments are exchanged to achieve gene recombination and expand the search range;

[0193] 3. Mutation operation: The parameters are perturbed by a normal distribution. After perturbation, the parameters must satisfy the HEPS safety boundary. If they exceed the boundary, they are corrected to the nearest boundary value to avoid invalid solutions.

[0194] 4. Iteration Termination: When the change in the optimal fitness value over 10 consecutive generations is ≤10⁻ 4 Alternatively, the iteration can stop after 100 iterations, and the optimal combination of control parameters can be output.

[0195] The beneficial effects of the present invention.

[0196] This invention uses a BP neural network to accurately fit the complex nonlinear mapping relationship between parameters such as speed, torque, and load in the HEPS power system. Combined with the mechanistic modeling constraints of the turboshaft engine and the electric motor, it achieves dynamic speed / torque matching, theoretically eliminating the matching deviation problem caused by fixed parameters in traditional PID control. According to the model derivation, the speed / torque matching deviation can be controlled within ±2%, and the torque surge amplitude during mode switching is theoretically reduced to below 2 N·m (the theoretical switching impact peak of existing PID control schemes is about 8 N·m due to the limitations of empirical formulas). This can fundamentally avoid power continuity gaps, improve switching smoothness, and meet the power continuity requirements of aviation flight.

[0197] By employing the ReLU activation function to avoid the vanishing gradient problem and combining it with the gradient descent algorithm to optimize the efficiency of neural network parameter updates, the system's response speed to load changes is significantly improved. Combined with the real-time correction function of the cascaded PI controller, the power response delay can theoretically be controlled within 0.1s (existing PID control, due to the inability to dynamically adjust parameters, typically has a response delay ≥0.5s to sudden load changes). For sudden high-load conditions such as takeoff and rapid climb, this invention can rapidly adjust the fuel-electric power distribution ratio and mode switching strategy through a neural network, fundamentally avoiding total thrust fluctuations and adapting to the dynamic requirements of complex flight conditions.

[0198] The BP+GA algorithm and cascade PI control logic designed in this invention both follow the calculation rules for modifying control parameters (for example, replacing the fixed parameters of the original PID with dynamic parameters output by the algorithm; for example, GA first sets the Kp and Ki baselines for different operating conditions (such as cruise and takeoff), BP monitors the operating conditions in real time and dynamically corrects them (increasing Kp during sudden load changes and decreasing it during smooth cruise), and the cascade PI controller directly uses these dynamic parameters to calculate torque commands, requiring only software logic modifications without hardware changes. The original PID parameters are "hard-coded constants," and control follows the same rules regardless of changes in operating conditions; while this invention uses variables calculated by the algorithm, and the parameters change accordingly with the operating conditions, always using the rules most suitable for the current scenario for control). Therefore, it does not require replacing engine, motor, or other hardware, nor does it require modifying hardware interfaces, and is naturally compatible with the existing HEPS hardware architecture.

[0199] Meanwhile, the existing hybrid engine software platform supports the replacement of control algorithm modules. The algorithm of this invention only needs to be packaged into an independent module according to the platform's input and output interface specifications (such as receiving parameters such as speed and SOC, and outputting torque commands) to be directly connected without reconstructing the entire software system. Therefore, the modification cycle is short and the cost is low.

[0200] This invention utilizes the nonlinear fitting capability of neural networks and the global optimization characteristics of GA genetic algorithms to dynamically adapt to changes in load requirements, battery SOC status, and engine operating parameters (such as exhaust temperature) during different flight phases (takeoff, cruise, and descent). Theoretically, this improves the system's control robustness under complex variables and avoids the performance degradation problem caused by insufficient adaptation to operating conditions in existing technologies. Attached Figure Description

[0201] Figure 1 This is a schematic diagram of the overall process.

[0202] Figure 2 This is a schematic diagram of a turboshaft engine.

[0203] Figure 3 It is a three-phase voltage vector.

[0204] Figure 4 This is a schematic diagram of a cascaded PI controller.

[0205] Figure 5 Schematic diagram of the energy flow direction of the system

[0206] Figure 6 This is a diagram of the BP neural network architecture.

[0207] Figure 7 This is a schematic diagram of the crossover process in the genetic algorithm iteration.

[0208] Figure 8 This is a curve showing the coordinated control of hybrid engine battery SOC, flight conditions, and energy modes. Detailed Implementation

[0209] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0210] like Figure 1 As shown, Step 1: Turboshaft engine mechanism modeling:

[0211] This invention employs a theoretical modeling method for aero-engines, which is a component-characteristic-based modeling approach. The advantage of component-characteristic-based modeling lies in its mature development and ability to effectively characterize the characteristics of various engine components. This invention focuses solely on modeling key components and their coupling relationships that are strongly correlated with the neural network input parameters. It aims to characterize the key characteristics of core components such as the gas turbine, power turbine, and combustion chamber (e.g., speed-power mapping, fuel quantity-exhaust temperature correlation), and offers high modeling efficiency with minimal redundant computation, meeting the input parameter requirements of a BP neural network.

[0212] In the engine thermodynamic model, the system can be simplified to a Brayton cycle, which includes processes such as isentropic compression of the compressor (providing a high-pressure environment for combustion and generating parameters related to the gas turbine speed), constant-pressure heating of the combustion chamber (supporting the quantitative calculation of exhaust temperature and fuel consumption rate through the coupling of fuel supply and combustion efficiency), and isentropic expansion of the turbine (realizing the conversion of thermal energy into mechanical energy and relating to the mapping relationship between power turbine speed and output torque). The whole process is a continuous process.

[0213] The modeling object in this study is the cross-sectional structure of a turboshaft engine, as shown below. Figure 2 As shown.

[0214] Figure 2 In the diagram, 1 is the engine intake, 2 is the compressor, 3 is the combustion chamber, 4 is the tail nozzle, 5 is the gas turbine, 6 is the power turbine, 7 is the engine's built-in reducer, and 8 is the power output shaft (free turbine shaft).

[0215] In the modeling, the cross-sections of the engine components are numbered as follows: 1-engine intake, 2-compressor inlet, 3-compressor outlet, 4-combustion chamber outlet (gas turbine inlet), 4.5-power turbine inlet, 5-power turbine outlet, 8-exhaust nozzle outlet.

[0216] 1. Gas turbine power balance equation

[0217]

[0218] Quantify the "input power - output power" balance relationship of gas turbine (input: combustion chamber thermal power; output: compressor drive power + mechanical losses).

[0219] 2. Gas turbine rotor power balance equation:

[0220]

[0221] This describes the dynamic balance of a gas turbine rotor (torque balance determines speed stability), where changes in speed directly reflect changes in power distribution. This formula supports the gas turbine speed (neural network input), the power balance state determines speed stability, and changes in speed reflect changes in power distribution.

[0222] 3. Power balance equation for a power turbine:

[0223]

[0224] The quantification of the "absorbed power - output power" balance of the power turbine (input: gas energy discharged from the gas turbine; output: load drive power + mechanical losses) is used. This formula correlates the input "power turbine speed" and the output "torque distribution ratio" of the neural network to establish the energy transfer logic between the two.

[0225] 4. Power balance equation for the power turbine rotor:

[0226]

[0227] Supports the turbine speed (neural network input) and torque distribution ratio (neural network output). Clearly defines the non-linear mapping relationship between speed and torque.

[0228] 5. The enthalpy at the combustion chamber outlet is:

[0229]

[0230] The outlet enthalpy directly reflects the combustion efficiency and is used to connect the thermodynamic relationship between fuel supply and exhaust temperature (the input part of the neural network).

[0231] 6. The combustion chamber outlet temperature is:

[0232]

[0233] The specific value of exhaust temperature (the input part of the neural network) is directly output. Exhaust temperature is a key safety parameter for judging the engine's operating status. The neural network needs to adjust the mode switching strategy based on this parameter (such as reducing fuel supply and increasing electric motor assistance when the temperature is too high).

[0234] 7. Fuel Consumption Rate (BSFC):

[0235]

[0236] Quantifying fuel consumption per unit of power is a core quantitative indicator for fuel economy optimization. The calculation of fuel consumption loss in the multi-objective loss function of neural networks supports multi-objective optimization.

[0237] 8. Emissions Index Formula

[0238]

[0239] Quantify pollutants in exhaust gas (such as The emission concentration of ) This is a correction term related to humidity-saturated vapor pressure. This is the ratio of the engine's operating ambient temperature to the standard atmospheric temperature. It is the ratio of the engine's operating environmental pressure to the standard environmental pressure, serving as the basis for calculating emission loss in the multi-objective loss function of the neural network, and supporting multi-objective optimization.

[0240] This step aims to characterize the key operational features of the turboshaft engine. It establishes a quantitative relationship between state parameters and performance indicators, providing reliable input data and physical constraints for subsequent neural networks and optimization algorithms.

[0241] For example:

[0242] By using the power balance equations of the gas turbine / power turbine and the rotor dynamics equations, the nonlinear mapping relationship between speed and torque is clarified, providing a calculation basis for the neural network input (gas turbine speed and power turbine speed), while constraining the physical boundary of torque distribution (to avoid exceeding the mechanical limits of the engine).

[0243] By using formulas for combustion chamber outlet temperature, fuel consumption rate (BSFC), and emission index, exhaust temperature (safety parameter), fuel consumption (economic indicator), and pollutant emissions (environmental indicator) are quantified. This provides key state parameters for the neural network input layer and a quantitative calculation basis for multi-objective optimization (smoothness + fuel consumption + emissions).

[0244] Step 2: Motor Mechanism Modeling

[0245] The modeling object selected for this study is a three-phase permanent magnet AC synchronous motor. This motor has two advantages: first, it is energy-saving and highly efficient, with minimal efficiency loss under light loads; second, it has a high torque density, which enables small size and lightweight design, a crucial factor for hybrid power systems.

[0246] In a three-phase permanent magnet AC synchronous motor, voltage can be decomposed into three-phase voltage vectors, such as... Figure 3 .

[0247] 1. Equations of the three-phase voltage vector in the natural coordinate system:

[0248]

[0249] In the formula: This refers to the three-phase voltage of the motor. It is a three-phase resistor. For three-phase current, This represents the three-phase winding flux linkage. This equation describes the three-phase decomposition relationship of the motor stator voltage and is fundamental to motor modeling. It provides constraints for calculating motor torque and speed, ensuring that motor control conforms to circuit laws.

[0250] Phase separation after disassembly

[0251]

[0252] 2. Three-phase winding flux linkage equation:

[0253]

[0254] In the formula Inductance for a three-phase winding; It is a three-phase winding flux linkage.

[0255] This equation quantifies the relationship between the magnetic flux linkage, current, and voltage of the stator winding. It is a prerequisite for calculating electromagnetic torque, supports the derivation of the electromagnetic torque formula, and provides physical constraints on the motor side for the output part (torque distribution ratio) of the neural network.

[0256] 3. Phase-separated flux linkage function

[0257]

[0258] This function simplifies the calculation of flux linkage equations, clarifies the independent calculation relationship of flux linkage in each phase, simplifies the formula of three-phase winding flux linkage equations, and ensures the operability of flux linkage calculation.

[0259] 4. The electromagnetic torque equation of an electric motor:

[0260]

[0261] The relationship between quantified magnetic flux linkage and electromagnetic torque represents the power output of the motor. The supporting neural network output (torque distribution ratio) ensures that the motor-side torque output conforms to electromagnetic laws.

[0262] 5. Equation of motion of an electric motor:

[0263]

[0264]

[0265]

[0266]

[0267] In the formula: Let be the mechanical angular velocity of the electric motor. This represents the load torque of the motor. This is the damping coefficient of the motor. Let be the moment of inertia of the motor. Electromagnetic angular velocity, Let represent the motor speed. This equation describes the dynamic balance of the motor rotor (electromagnetic torque → speed change), clarifying the relationship between motor speed and torque. It supports the coordinated control of motor speed and torque, ensuring that the torque distribution ratio output by the neural network can be converted into actual speed.

[0268] 6. Equations for the input and output power of an electric motor:

[0269]

[0270]

[0271] Quantify the energy conversion efficiency of the motor (input electrical power → output mechanical power), including loss calculation. Ensure that the torque distribution ratio optimized by the neural network meets the energy efficiency requirements.

[0272] This step clarifies the electrical and dynamic characteristics of the three-phase permanent magnet AC synchronous motor, establishes the mapping relationship between the motor's control commands and power output, provides physical constraints on the motor side for torque distribution optimization, and ensures that control commands can be implemented.

[0273] For example:

[0274] By using the three-phase voltage vector equation, the three-phase winding flux linkage equation, and the electromagnetic torque equation, the relationship between voltage, current, flux linkage, and torque is quantified, and the output of the neural network (torque distribution ratio) is constrained to conform to the electromagnetic laws of the motor (such as avoiding torque exceeding the rated output of the motor).

[0275] By defining the dynamic response relationship between motor speed and torque through the motor motion equation and input / output power equation, the coordinated control of speed and torque is supported, ensuring that the output part of the neural network (torque distribution ratio) can be converted into the actual speed. At the same time, it ensures a smooth transition of motor power output during mode switching (such as avoiding sudden torque changes).

[0276] Step 3: Engine and electric motor work together:

[0277] The generator converts the mechanical energy output by the turboshaft engine into electrical energy, which is then transmitted to the electric motor via a power bus. The generator serves as an auxiliary component of the engine, and the two together form a turbine-generator system. The generator's design speed differs from the engine's, therefore...

[0278] 1. Generator speed-torque matching equation:

[0279]

[0280]

[0281] In the formula: The speed at the output end of the reducer. The speed at the input end of the reducer. The output torque of the reducer. The input torque of the reducer. The reduction ratio of the reducer. This equation represents the mechanical efficiency of the reducer. The input is connected to the engine's power shaft, and the output is connected to the generator's working shaft. This equation quantifies the conversion relationship between engine speed, generator speed, and motor speed, resolving the speed mismatch problem between the engine and generator. Simultaneously, this equation clarifies the logic of power transfer from the engine to the motor, supporting the neural network output section (power distribution ratio).

[0282] All-electric operating mode

[0283]

[0284] Charging mode

[0285]

[0286] 2. The cascaded PI controller takes the following form (cooperative control logic with BP neural network and GA algorithm): see Figure 4

[0287]

[0288] The cascaded PI controller is the core execution unit for "real-time correction and smooth transition" during HEPS mode switching. Its control parameters and target values ​​are dynamically optimized through a BP neural network and a GA genetic algorithm, forming a closed-loop collaborative control chain of "GA+BP+PI". The specific interaction logic is as follows:

[0289] Outer Loop (Speed ​​Loop): The "target gas turbine speed" obtained by global optimization using the GA genetic algorithm is used as the outer loop setpoint (this target speed takes into account both fuel efficiency and smooth switching, and is the optimal solution of GA under multi-objective constraints); the feedback value of the outer loop is the real-time speed of the engine power turbine (one of the input parameters of the BP neural network, which is calculated in real time through mechanism modeling); the output result of the outer loop serves as the upper limit of the torque reference for the inner loop (torque loop), ensuring that stable speed is a prerequisite for torque distribution.

[0290] Inner loop (torque loop): It consists of two parts based on the inner loop set value: 1. Torque distribution ratio output by BP neural network; 2. Compensation torque for speed deviation of outer loop (calculated by outer loop based on speed deviation, used to correct the basic torque distribution ratio); The feedback value of inner loop is the real-time output torque of motor (calculated by electromagnetic torque equation in motor mechanism modeling), and finally the output motor torque control signal is achieved by PI regulation to realize accurate torque tracking.

[0291] The proportional and integral coefficients of the cascaded PI controller are not fixed values, but are dynamically adjusted by the BP neural network according to real-time operating conditions. The BP neural network takes "load demand power change rate, battery SOC, and exhaust temperature" as input features, and outputs PI parameter correction coefficients adapted to the current operating conditions through a trained nonlinear mapping model, updating the PI controller parameters in real time and avoiding the response lag problem of traditional fixed PI parameters when the load changes suddenly.

[0292] 3. Basic voltage equations for a three-phase motor in synchronous rotational coordinates along the dq axis:

[0293]

[0294] This equation transforms the motor model from a natural coordinate system to a synchronous rotating coordinate system, simplifying control calculations. At the same time, it improves the real-time performance of motor control, adapting to the fast response requirements of neural networks.

[0295] 4. SOC Calculation

[0296]

[0297] Charging limit:

[0298] SOC≤0.95

[0299] Lower discharge limit:

[0300] SOC ≥ 0.2

[0301] This equation quantifies the remaining battery capacity based on the coulomb counting method, clearly defining the charge and discharge boundaries. It provides a calculation basis for the neural network input part (battery SOC) and provides constraints for the network output part (mode switching instructions) (such as disabling boost at low SOC).

[0302] 5. For the switching between different flight phases, see [link / reference]. Figure 5 Hybrid power system shared power supply mode:

[0303] When an aircraft is in a high-load phase such as taxiing, takeoff, or climb, and the power demand exceeds the output power of the turbine generator system, the energy storage battery pack needs to discharge to compensate for the insufficient power of the engine and keep the engine running at its highest efficiency point.

[0304]

[0305]

[0306]

[0307] To meet the power requirements of the electric motor, This refers to the generator's output power. This refers to the battery output power.

[0308] When the aircraft is in low to medium load conditions such as cruise or descent, if the SOC is high and the energy storage battery pack has sufficient energy, the battery can provide power independently, quickly consume the power, improve the overall energy utilization of the system, and reduce fuel consumption.

[0309] All-electric operating mode

[0310]

[0311]

[0312]

[0313] Battery charging working mode

[0314] Power required when the aircraft is in the cruise phase Power supplied by generator It is closer to but slightly smaller than the latter. In order to maintain the internal power balance of the system, the energy storage battery pack charges the battery through a bidirectional DC / DC converter.

[0315]

[0316]

[0317]

[0318] This step addresses the power coupling issue between the engine and the electric motor, establishing a collaborative working logic and a real-time correction mechanism. For example:

[0319] By using the generator speed-torque matching equation, the problem of speed mismatch between the engine and the generator is solved, the logic of power transmission from the engine to the motor is clarified, and the output part of the neural network (the power distribution ratio between oil and electricity) is supported (e.g., the efficient conversion of engine mechanical energy → electrical energy → motor power).

[0320] The design incorporates a "GA+BP+PI" cascade closed-loop control chain: the outer loop (speed loop) uses the gas turbine target speed optimized by GA as a benchmark to ensure speed stability; the inner loop (torque loop) combines the torque distribution ratio output by BP with the speed deviation compensation torque, and achieves precise torque tracking through PI regulation to avoid power shock during mode switching.

[0321] Based on flight conditions and battery SOC status, the switching logic for three modes—hybrid power supply, all-electric operation, and battery charging—is defined, providing scenario-based constraints for the neural network's output of "mode switching commands." For example... Figure 8 As shown.

[0322] The PI controller parameters are dynamically adjusted (optimized by a BP neural network based on real-time operating conditions) to solve the problem of delayed response to load changes in traditional fixed PI parameters, thereby improving system adaptability.

[0323] Step 4: Building the BP neural network:

[0324] A backpropagation (BP) neural network model consists of an input layer, hidden layers, and an output layer, with a purely linear transfer function between the hidden and output layers. (See...) Figure 6 .

[0325] (1) Data preparation:

[0326] A dataset was prepared for training and testing the neural network. By extracting component-level models of turboshaft engines and HEPS power switching dynamic models, this invention collected data from 30 sets of engines, covering typical flight profiles such as takeoff, cruise, and descent.

[0327] The input parameters, namely the input layer of the BP neural network, include six key state parameters: turbine speed of the turboshaft engine, power turbine speed, power distribution ratio of oil and electricity, load demand power, battery SOC, and exhaust temperature.

[0328] The output parameters, namely the output layer of the BP neural network, include three control parameters: engine operating mode command, motor operating mode command, and torque distribution ratio. These parameters are used to control the mode switching and power output of the engine and motor.

[0329] The 30 sets of data were divided into 20 training sets and 10 prediction sets. The input matrix of the 20 training sets was (6×20) and the output matrix was (3×20); the input matrix of the 10 prediction sets was (6×10) and the output matrix was (3×10).

[0330] (2) Determine the network architecture:

[0331] The BP neural network constructed in this invention includes an input layer, a hidden layer, and an output layer, with 6 parameters in the input layer;

[0332] After comparing the network prediction results with different numbers of nodes in the hidden layer, it was finally determined that the black box function was best fitted when there were 11 hidden layer nodes. Therefore, 11 hidden layer nodes were selected.

[0333] The output layer outputs three parameters, which correspond to the precise control of the engine and motor's working mode switching and torque distribution.

[0334] (3) Initialize weights and thresholds:

[0335] In the programming software, the rand function is used to randomly initialize the weight matrix V (size 11×6) between the input layer and the hidden layer, the weight matrix W (size 3×11) between the hidden layer and the output layer, as well as the hidden layer threshold b1 and the output layer threshold b2, with the initialization range being [0,1].

[0336] (4) Select activation function

[0337] In this invention, the activation function of the neural network is the ReLU function. The ReLU function is a nonlinear function that, through the combination of multiple neurons, can accurately fit the complex nonlinear mapping between parameters such as high-pressure rotor speed and load demand in the HEPS power system and mode switching and torque output. ReLU has greater advantages in avoiding gradient vanishing and reducing real-time control latency, and is suitable for parameters with a large dynamic range (such as a gas turbine speed of 50,000-165,000 rpm, exhaust temperature ≤1003K, and load power demand of 1.11-12.97kW). Therefore, the ReLU function is chosen as the activation function of the neural network, and its functional form is as follows:

[0338] (in The linear input values ​​for the hidden layer neurons are, i.e. (Input parameter * corresponding weight) + threshold (b1)

[0339] (5) Forward propagation:

[0340] This involves passing input data into the network, calculating the output of each neuron through the neurons in each layer, and the propagation process of the BP neural network.

[0341] The input layer is denoted by L1, the output layer by L2, the weights between L1 and the hidden layer are denoted by V, and the matrix size is 11*6; the weights between L2 and the hidden layer are denoted by W, and the matrix size is 3*11.

[0342] Where X is the input parameter vector (6×1) and H is the hidden layer output vector (11×1). Where Y is the output parameter vector (3×1).

[0343] (6) Back propagation:

[0344] The parameters are updated using a weighted multi-objective loss function, with gradients applied to the weights and biases to reduce the loss function's value. Weight coefficients are set based on priority (e.g., ride comfort weight 0.5, fuel consumption rate weight 0.3, emissions weight 0.2). Backpropagation uses gradient descent or its variants to adjust the parameters to minimize the loss. Backpropagation allows the neural network to initially learn the parameter mapping relationships under multi-objective constraints, providing a high-quality initial solution for the subsequent global optimization by the GA genetic algorithm.

[0345]

[0346] Torque fluctuation loss; This is due to fuel consumption loss; This is for emissions losses.

[0347] This step breaks through the limitations of traditional linear control, accurately fits the complex nonlinear characteristics of HEPS, and achieves dynamic matching of real-time state perception and precise control output, providing a preliminary optimization scheme for mode switching and torque distribution.

[0348] A nonlinear mapping model is constructed for inputs (key parameters of engine / motor / battery / load) and outputs (mode commands, torque distribution ratio). Through the ReLU activation function and an architecture with 11 hidden nodes, the complex characteristics of HEPS under different operating conditions are accurately captured, and preliminary control commands are quickly output. At the same time, multi-objective optimization logic is incorporated through a weighted multi-objective loss function (smoothness 0.5, fuel consumption 0.3, emissions 0.2).

[0349] Step 5: Program the GA genetic algorithm and use it to optimize the HEPS output parameters: (e.g.) Figure 7 As shown, the GA genetic algorithm, based on a multi-objective optimization function, obtains a set of optimal "core control parameters," including:

[0350] 1. (Target speed of gas turbine (balancing efficiency and smoothness, it is a reference benchmark for gas turbine speed. The BP neural network determines whether fuel supply and electric motor assistance are needed by comparing the target speed of gas turbine with the current gas turbine speed (input parameter))

[0351] 2. Hybrid power distribution ratio (balancing fuel consumption and power, serving as the basis for allocating input parameters (load demand power, battery SOC) to the neural network; the BP neural network outputs the designed torque distribution command based on the input parameters and the hybrid power distribution ratio optimized by GA).

[0352] 3. Mode switching trigger threshold (balancing smoothness and response speed, it is the switching condition of BP neural network load demand power, battery SOC, and exhaust temperature. The neural network monitors whether these three values ​​have reached the switching trigger threshold and outputs mode switching command to make adjustments accordingly).

[0353] The three parameters optimized in this process can achieve multi-objective optimization while avoiding the inefficiency and tendency to get trapped in local optima in the GA optimization process. GA determines the optimal core control parameters through global search and inputs them into the mapping model of the BP neural network, so that the control parameters output by BP meet both the real-time response requirements and the multi-objective optimization objectives.

[0354] Based on the GA genetic algorithm, evolutionary operators such as selection, crossover, and mutation are used to generate new combinations of design parameters and evaluate their performance to guide the search for the next generation. During the multiple iterations of this GA genetic algorithm, the parent population is continuously updated. The key lies in leveraging its global search and multi-objective optimization capabilities to generate the optimal engine / motor operating mode switching strategy and torque output combination based on the nonlinear mapping model of "HEPS key state parameters - output control parameters" established by the BP neural network. This simultaneously achieves multi-objective optimization of fuel economy, emissions, and ride comfort, addressing the technical shortcomings of existing technologies such as large impacts from mode switching, poor adaptability to sudden load changes, and imbalances caused by single optimization methods.

[0355] (1) Definition of fitness function

[0356] In the process of HEPS multi-objective collaborative optimization, the smoothness of the power mode switching of the hybrid aero engine is required to meet the preset threshold of aviation flight safety (such as peak torque fluctuation ≤ 2 N·m, power response delay ≤ 0.1s), while minimizing aviation fuel consumption rate and NOx exhaust emission concentration, and optimizing the core adjustable control parameters of HEPS (i.e., target speed of turboshaft engine, fuel-electric power distribution ratio, engine / motor mode switching trigger threshold) so that the comprehensive operating efficiency of the hybrid aero engine under all flight conditions reaches the optimal level.

[0357] In the construction of the fitness function, conditional statements are used to pre-screen the optimized output results. First, a pre-established mathematical model of HEPS power switching dynamics and energy consumption and emission theory is imported for calculation. Priority is given to verifying whether the power switching smoothness corresponding to the parameter combination to be evaluated meets the preset safety threshold requirements. When the smoothness index meets the preset requirements, the fuel consumption rate, exhaust emission concentration and comprehensive operating efficiency corresponding to the parameter combination are further calculated, and the comprehensive operating efficiency value integrating fuel consumption and emissions is output. In subsequent iterations, the core control input parameter corresponding to the maximum comprehensive operating efficiency is selected as the optimal solution. If the smoothness index does not meet the preset requirements, the fitness value of the parameter combination is directly set to a minimum invalid value and eliminated.

[0358] In the iterative process of finding the optimal solution, in order to avoid stability problems in numerical calculations (such as iteration divergence, weight imbalance, and gradient vanishing) and to ensure the convergence and computational accuracy of the genetic algorithm, the output range of the fitness function is limited to the negative range.

[0359] (2) Genetic Algorithm Parameter Settings

[0360] 1. Chromosome Encoding: The chromosome is a column vector that stores the three core adjustable control parameters of HEPS. The three core control parameters are: target speed of the turboshaft engine gas turbine, power distribution ratio of oil and electricity, and engine / motor mode switching trigger threshold (based on load demand power and battery SOC). Each element of the column vector corresponds to a specific value of a control parameter, and the value range conforms to the safety operation boundary constraints of HEPS (e.g., gas turbine speed range of 50,000-165,000 rpm, power distribution ratio of oil and electricity range of 0-1).

[0361] 2. Population initialization: Construct a zero matrix of appropriate dimensions to store the parent chromosome data during the genetic algorithm iteration process. Initialize the population size to 50 and the number of iterations to 100.

[0362] 3. Evolutionary operator settings: The selection operator adopts the roulette wheel selection method, the crossover operator adopts the single-point crossover method with a crossover probability of 0.7, and the mutation operator adopts the basic bit mutation method with a mutation probability of 0.05. New combinations of design parameters are generated through evolutionary operators such as selection, crossover, and mutation, and their performance is evaluated to guide the search of the next generation. The parent population is continuously updated in multiple iterations to finally obtain the optimal combination of control parameters.

[0363] (3) Genetic algorithm programming: Genetic algorithm iteration includes selection, crossover, and mutation.

[0364] Based on the global search capability of the GA genetic algorithm, the parent population is iteratively updated through selection, crossover, and mutation. On the basis of a "key state parameter-control parameter" mapping model constructed by a BP neural network, the optimal solution for multiple objectives is selected. The specific implementation is as follows:

[0365] 1. Selection operation: The roulette wheel selection method is adopted. The parent chromosome (target speed of turboshaft engine, power distribution ratio of oil and electricity, mode switching trigger threshold) is input and substituted into the fitness function to calculate the overall operating efficiency. The selection probability of a single parent is equal to the overall operating efficiency of that parent / the sum of the overall operating efficiencies of all parents. The dominant parent is randomly selected according to the probability to ensure the inheritance of superior genes.

[0366] 2. Crossover operation: Set the crossover probability to 0.7, generate a crossover point of 1-2 using the rand function (no crossover occurs when returning to 3), truncate the two sets of parent chromosomes at the crossover point and exchange parameter fragments to achieve gene recombination and expand the search range.

[0367] 3. Mutation operation: Set the mutation probability to 0.05, and perform normal distribution perturbation on the parameters with a standard deviation of 0.03. After perturbation, the parameters must meet the HEPS safety boundary (speed 50000-165000rpm, distribution ratio 0-1, etc.). If the parameters exceed the boundary, they are corrected to the nearest boundary value to avoid invalid solutions.

[0368] 4. Iteration Termination: When the change in the optimal fitness value over 10 consecutive generations is ≤10⁻ 4 Alternatively, the iteration can stop after 100 iterations, and the optimal combination of control parameters can be output.

[0369] This step addresses the issue of BP neural networks potentially getting stuck in local optima. The GA genetic algorithm optimizes three core control parameters (gas turbine target speed, electric power distribution ratio, and mode switching trigger threshold) through global search, ensuring that the control strategy simultaneously meets multiple objectives such as smoothness, fuel economy, and emissions. The optimized parameters are then fed back to the BP neural network and the cascaded PI controller to complete the precise calibration of the control commands.

[0370] Steps one and two provide physical constraints and data support for the entire scheme.

[0371] Step 1 (engine modeling) and Step 2 (electric motor modeling) quantify the characteristics of the two core power sources of HEPS, which together constitute the physical model library of the hybrid power system;

[0372] The outputs of both (such as engine speed, temperature, and fuel consumption, and motor torque and speed constraints) are the core data source for all subsequent steps: the collaborative work in step three requires the design of coupling logic based on the characteristics of both, the input layer parameters of the BP neural network in step four are directly taken from the modeling results of both, and the GA optimization in step five also requires the physical constraints of both as boundary conditions.

[0373] Step 3: Connect contact modeling and algorithm optimization.

[0374] Step 3 is based on the modeling results of Steps 1 and 2: because the generator speed-torque matching equation needs to be designed based on the engine speed range and motor speed requirements, and the parameter constraints of the cascade PI controller need to meet the dynamic response characteristics of the engine and motor.

[0375] Step three provides a collaborative logic framework for steps four and five: the mode commands and torque distribution ratios output by the BP neural network need to be executed in the three working modes and cascaded PI control chains defined in step three; the core parameters optimized by the GA genetic algorithm also need to be transformed into actual power output through the collaborative mechanism of step three.

[0376] Meanwhile, the cascaded PI controller in step three will receive the optimization results from steps four and five (such as BP dynamically adjusting PI parameters and GA providing target speed), realizing real-time correction of output → controller execution → operating condition feedback.

[0377] Steps four and five represent the iterative algorithmic relationship of fitting → optimization → closure.

[0378] Step four (BP neural network) builds a training set (30 sets of data covering the entire flight profile) based on the modeling data from steps one and two, and designs input / output parameters (such as input battery SOC and output mode commands) based on the collaborative logic from step three.

[0379] Step 5 (GA Genetic Algorithm): The output of Step 4 is used as the initial solution of the model. The three core parameters of GA optimization need to be added to the mapping model of the BP neural network to verify its optimization effect. Conversely, the parameters optimized by GA will be fed back to the BP neural network to correct its nonlinear mapping relationship and avoid local optima.

[0380] The two ultimately form a closed loop through the collaborative mechanism in step three: the BP neural network responds quickly to real-time operating conditions, the GA algorithm performs periodic global calibration, and the cascaded PI controller performs real-time correction, ensuring the stable performance of HEPS under scenarios such as load changes and operating condition switching.

Claims

1. An optimization design method for hybrid engine drive switching control based on neural networks, characterized in that, Includes the following steps; Step 1: Perform mechanism modeling of the turboshaft engine and output the engine's speed, temperature, and fuel consumption parameters; Step 2: Perform motor mechanism modeling and output the parameters of motor torque and speed constraints; Step 3: Based on the output parameters of the engine and electric motor, the engine and electric motor work together by controlling the coordinated distribution and mode switching of oil and electric power. Step 4: Build a BP neural network to output mode commands and torque distribution ratios; Step 5: Through global optimization using the GA genetic algorithm, the hybrid engine drive switching control is directly applied. The optimized target speed of the gas turbine, the ratio of oil and electric power distribution, and the mode switching trigger threshold are used to achieve multi-objective optimization control of the hybrid engine drive switching.

2. The optimization design method for hybrid engine drive switching control based on neural networks according to claim 1, characterized in that, Step one specifically involves: (1). Gas turbine power balance equation ； Indicates the output power of the gas turbine. Indicates the compressor drive power. Gas turbine mechanical loss power; (2). Power balance equation of gas turbine rotor: ； This represents the moment of inertia of the gas turbine rotor. Indicates the rotational speed of the gas turbine rotor. This represents the angular acceleration of the gas turbine rotor. Represents the deviation term in the equation; (3). Power balance equation of the power turbine: ； This represents the total power absorbed by the system (which must satisfy both load demand and energy loss). This indicates the power demand of the load (such as the propulsion power demand of an electric motor driving a propeller). Indicates system energy loss; (4). Power balance equation of the power turbine rotor: ； Indicates the output power of the power turbine. This represents the moment of inertia of the power turbine rotor. Indicates the rotational speed of the power turbine rotor. This represents the angular acceleration of the power turbine rotor. Represents the deviation term in the equation; (5). The enthalpy at the combustion chamber outlet is: ； Indicates the enthalpy at the combustion chamber outlet. Indicates fuel flow rate. Indicates the calorific value of fuel oil. Indicates combustion chamber efficiency. Indicates the airflow rate at the combustion chamber inlet. Indicates the enthalpy of the air at the combustion chamber inlet. This indicates the total flow rate of gas exiting the combustion chamber; (6). The combustion chamber outlet temperature is: ； Indicates the total temperature at the combustion chamber outlet. This represents the enthalpy-temperature conversion function, used to calculate the corresponding temperature based on a given oil-to-gas ratio and enthalpy value. Indicates the air-fuel ratio at the combustion chamber outlet; (7) Fuel consumption rate: ； This represents the engine's output power; this equation quantifies the fuel consumption per unit power, and is the core quantitative indicator for optimizing fuel economy. (8). Emissions Index Formula ； NO represents the actual working conditions. x Emission index, NO indicates the ground standard environment x Emission index; By interrelation and constraint of the above formulas, a complete mechanism model of the turboshaft engine is formed; the power balance equation and rotor dynamics equation determine the dynamic relationship between engine speed, torque and power. The combustion chamber outlet enthalpy and temperature equation determines the engine's thermal state and safety boundaries; the fuel consumption rate equation quantifies economic indicators; the emission index equation quantifies environmental indicators; and the final outputs are engine speed, exhaust temperature, fuel consumption, and NO. x Emission parameters serve as the basis for judging engine operating status, making mode switching decisions, and allocating torque in hybrid engine drive switching control.

3. The optimization design method for hybrid engine drive switching control based on neural networks according to claim 1, characterized in that, Step two specifically involves: The motor is selected as a three-phase permanent magnet AC synchronous motor. In the three-phase permanent magnet AC synchronous motor, the voltage is decomposed into three-phase voltage vectors. (1). Equations of the three-phase voltage vector in the natural coordinate system: ； In the formula: This refers to the three-phase voltage of the motor. It is a three-phase resistor. It is a three-phase current. For three-phase winding flux linkage; Phase separation after disassembly ； (2). Three-phase winding flux linkage equation: ； In the formula Inductance for a three-phase winding; For three-phase winding flux linkage; (3). Phase-separated flux linkage function ； This function simplifies the calculation of flux linkage equations, clarifies the independent calculation relationship of flux linkage in each phase, and simplifies the formula of flux linkage equation for three-phase windings. (4). The electromagnetic torque equation of the motor: ； The relationship between quantified magnetic flux and electromagnetic torque represents the power output of the motor; (5). Equation of motion of the electric motor: ； ； ； ； In the formula: Let be the mechanical angular velocity of the electric motor. This represents the load torque of the motor. This is the damping coefficient of the motor. Let be the moment of inertia of the motor. Electromagnetic angular velocity, This refers to the motor speed; (6). Equations for the input and output power of the electric motor: ； ； The energy conversion efficiency of the motor (input electrical power → output mechanical power) is quantified, including loss calculation, to ensure that the torque distribution ratio optimized by the neural network meets the energy efficiency requirements; By coupling and constraining the above formulas, a mechanism model of a three-phase permanent magnet synchronous motor is constructed: the voltage equation and flux linkage equation clarify the electromagnetic constraints of the motor, the electromagnetic torque equation establishes the relationship between flux linkage and power output, the motion equation determines the dynamic response law of speed and torque, and the input and output power equation quantifies the energy conversion efficiency; finally, the output parameters are the motor torque, motor speed, power constraints, and dynamic response characteristics.

4. The optimization design method for hybrid engine drive switching control based on neural networks according to claim 3, characterized in that, Step three specifically involves: The generator converts the mechanical energy output by the turboshaft engine into electrical energy, which is then transmitted to the electric motor via a power bus. The generator is considered an auxiliary component of the engine, and the two together form a turbine power generation system. The generator's design speed is different from that of the engine, hence the following: Based on the engine speed, power, and temperature parameters output in step one, and the motor torque, speed, and power constraint parameters output in step two, engine-motor power coupling and coordinated control are performed to achieve the underlying execution and stable operation of hybrid engine drive switching. The generator converts the mechanical energy output by the turboshaft engine into electrical energy, which is then transmitted to the electric motor via a power bus. The generator serves as an auxiliary component of the engine, and the two together form a turbine power generation system. A generator speed-torque matching equation is introduced to achieve coordinated matching of speed and torque. This equation takes the engine dynamics characteristics from step one as input and the electromagnetic and dynamic constraints of the motor from step two as output boundaries, efficiently converting the engine's mechanical energy into usable electrical energy for the motor. This provides a power transmission basis for the coordinated operation of the engine and electric motor and the switching of hybrid drive.

5. The optimization design method for hybrid engine drive switching control based on neural networks according to claim 4, characterized in that, Step three specifically involves: (1). Generator speed-torque matching equation: ； ； In the formula: The speed at the output of the reducer. The speed at the input end of the reducer. The output torque of the reducer. The input torque of the reducer. The reduction ratio of the reducer. To optimize the mechanical efficiency of the reducer; the input end is connected to the engine power shaft, and the output end is connected to the generator working shaft; All-electric operating mode ； Charging mode ； (2). The cascaded PI controller is as follows; ； The cascaded PI controller is the core execution unit for real-time correction and smooth transition during HEPS mode switching. Its control parameters and target values ​​are dynamically optimized through a BP neural network and a GA genetic algorithm, forming a closed-loop collaborative control chain of GA+BP+PI. The specific interaction logic is as follows: Outer loop: The target speed of the gas turbine obtained by global optimization using the GA genetic algorithm is used as the outer loop setpoint; the feedback value of the outer loop is the real-time speed of the engine power turbine; the output result of the outer loop is used as the upper limit of the torque reference of the inner loop, ensuring that the speed is stable, which is a prerequisite for torque distribution; Inner ring: Composed of two parts based on the inner ring setting value; The torque distribution ratio output by the BP neural network; The outer ring compensates for the speed deviation; the inner ring feedback value is the real-time output torque of the motor, which is ultimately controlled by a PI regulator to output the motor torque control signal, thereby achieving precise torque tracking. The BP neural network takes the load demand power change rate, battery SOC, and exhaust temperature as input features, and outputs PI parameter correction coefficients adapted to the current operating conditions through a trained nonlinear mapping model, and updates the PI controller parameters in real time. Among them, the load demand power change rate is calculated by combining the motor motion equation and input / output power equation in the motor mechanism modeling in step two with the real-time flight condition load demand, reflecting the dynamic change of the motor side load. The battery SOC is calculated in real time by the hybrid power system energy storage battery module using the coulomb counting method, and is used to determine the battery's charging and discharging capacity and the conditions for switching drive modes. The exhaust temperature is calculated in real time by the combustion chamber outlet temperature equation in the turboshaft engine mechanism modeling step one, reflecting the engine thermal state and safe operating boundary. (3). The basic voltage equation of a three-phase motor in synchronous rotating coordinates along the dq axis: ； This equation transforms the motor model from a natural coordinate system to a synchronous rotating coordinate system, simplifying control calculations. Simultaneously, it improves the real-time performance of motor control, adapting to the fast response requirements of neural networks. (4). SOC calculation ； Upper limit of charging: SOC ≤ 0.95; Lower limit of discharging: SOC ≥ 0.

2. (5) Switching between different flight phases Hybrid power system shared power supply mode; When the aircraft is in the high-load phase of taxiing, takeoff, and climb, and the power demand of the load is greater than the output power of the turbine generator system, the energy storage battery pack needs to discharge to compensate for the insufficient power of the engine and keep the engine running at its highest efficiency point. ； ； ； To meet the power requirements of the electric motor, This refers to the generator's output power. This refers to the battery's output power. When the aircraft is in low to medium load conditions such as cruise or descent, if the SOC is high and the energy storage battery pack has sufficient energy, the battery can provide power independently. All-electric operating mode ； ； ； Battery charging working mode Power required when the aircraft is in the cruise phase Power supplied by generator Approximately but smaller than the latter, the energy storage battery pack charges the battery via a bidirectional DC / DC converter; ； ； ； The generator speed-torque matching equation and the basic voltage equation of the three-phase motor in the synchronous rotating coordinate system of the dq axis mainly provide physical constraints; Through the mutual coupling and constraints of the above five parts, a complete engine-motor power coupling and coordinated control mechanism is formed: the generator speed-torque matching equation takes the engine modeling parameters from step one as input and the motor constraints from step two as boundaries, laying the foundation for power transmission; the three-phase motor dq-axis voltage equation is derived based on the motor mechanism from step two, limiting the physical limits of motor power output; the cascade PI controller uses the first two parts as a reference and basis to achieve real-time correction of speed and torque; SOC calculation provides energy safety boundaries and constrains the driving mode switching conditions; the mode switching of different flight stages combines operating conditions and SOC to determine the switching logic of engine drive, motor drive, and hybrid drive; Ultimately, it outputs stable, smooth, and safe hybrid engine drive switching control commands and torque distribution commands, realizing coordinated operation of the engine and electric motor and smooth switching between multiple drive modes.

6. The optimization design method for hybrid engine drive switching control based on neural networks according to claim 5, characterized in that, The generator speed-torque matching equation resolves the contradiction of the speed mismatch between the engine and the generator, and clarifies the transmission efficiency of the engine's mechanical energy to electrical energy. The basic voltage equation of a three-phase motor in synchronous rotating coordinates along the dq axis quantifies the electromagnetic relationship between voltage, current, flux linkage, and speed, and clarifies the torque output limit of the motor. The cascaded PI controller and SOC calculation precisely adjust power within constraints according to the instructions from the upper layer, while simultaneously managing battery power. The speed loop reference of the cascade PI controller comes from the generator speed-torque matching equation, and the current command must satisfy the dq axis voltage equation. When the trigger mode changes during flight phase switching, the PI controller will dynamically adjust the parameters to adapt to the torque requirements of the new operating condition. When calculating SOC, if SOC ≥ 95%, the all-electric operating mode is triggered; if SOC ≤ 20%, the engine power supply + battery charging mode is triggered. When the SOC is too low, the PI controller will limit the motor's assist torque to prioritize battery safety. The engine torque distribution ratio is increased, the generator speed is increased, and the charging current is increased; Switching between different flight phases: Observe flight conditions and SOC, determine the working mode, and issue instructions to the lower level; The switching between different flight phases means that the speed loop setpoint and torque distribution ratio of the PI controller will be adjusted synchronously in different modes; the power output in different modes must conform to the physical laws of the generator speed-torque matching equation and the dq axis voltage equation.

7. The optimization design method for hybrid engine drive switching control based on neural networks according to claim 6, characterized in that, Step four specifically involves: The BP neural network model consists of an input layer, hidden layers, and an output layer, with a purely linear transfer function used between the hidden layers and the output layer. (1) Prepare a dataset for training and testing the neural network; by extracting the component-level mechanism model of the turboshaft engine established in step one and the mechanism model of the electric motor established in step two, and combining the engine-electric motor power coupling relationship, construct the HEPS hybrid engine drive switching dynamic model, extract full working condition state data based on the above model, and form the training set and test set required for the BP neural network; The input parameters, namely the input layer of the BP neural network, include the turbine speed of the turboshaft engine, the power turbine speed, the power distribution ratio of oil and electricity, the load demand power, the battery SOC, and the exhaust temperature. The output parameters, namely the output layer of the BP neural network, include engine operating mode commands, motor operating mode commands, and torque distribution ratios, which are used to control the mode switching and power output of the engine and motor. (2) Determine the network architecture. The BP neural network includes an input layer, a hidden layer and an output layer. The input layer has 6 parameters. These are the turbine speed of the turboshaft engine, the power turbine speed, the power distribution ratio between oil and electricity, the power demand under load, the battery SOC, and the exhaust temperature. After comparing the network prediction results with different numbers of nodes in the hidden layer, it was finally determined that the black box function was best fitted when there were 11 hidden layer nodes. Therefore, 11 hidden layer nodes were selected. The output layer outputs three parameters, which correspond to the precise control of the engine and motor's working mode switching and torque distribution. (3) Initialize weights and thresholds. In the programming software, use the rand function to randomly initialize the weight matrix V between the input layer and the hidden layer, the weight matrix W between the hidden layer and the output layer, as well as the hidden layer threshold b1 and the output layer threshold b2. The initialization range is [0,1]. (4) Select the activation function. Choose the ReLU function as the activation function of the neural network. Its function form is: ； in The linear input values ​​for the hidden layer neurons are, i.e. (Input parameter * corresponding weight) + threshold (b1); (5) The input data is transmitted to the network, and the output of each neuron is calculated through the neurons of each layer. This is the propagation process of the BP neural network. The input layer is denoted by L1, the output layer by L2, the weights between L1 and the hidden layer are denoted by V, and the matrix size is 11*6; the weights between L2 and the hidden layer are denoted by W, and the matrix size is 3*11. ； Where X is the input parameter vector (6×1) and H is the hidden layer output vector (11×1). ； Where Y is the output parameter vector (3×1); (6) The parameters are updated by the gradient of the weights and biases through the weighted multi-objective loss function, with the weight coefficients set based on priority; backpropagation uses the gradient descent algorithm or its variants to adjust the parameters to minimize the loss; backpropagation enables the neural network to learn the parameter mapping relationship under multi-objective constraints, providing a high-quality initial solution for the global optimization of the subsequent GA genetic algorithm; ； Torque fluctuation loss; This is due to fuel consumption loss; This is for emissions losses.

8. The optimization design method for hybrid engine drive switching control based on neural networks according to claim 7, characterized in that, Step five specifically involves: The GA genetic algorithm, based on a multi-objective optimization function, obtains the optimal core control parameters, including: (1). Target speed of the gas turbine; (2) Power distribution ratio between oil and electricity; (3) Mode switching trigger threshold; Based on the GA genetic algorithm, evolutionary operators such as selection, crossover, and mutation are used to generate new combinations of design parameters and evaluate their performance to guide the search for the next generation; (1) In the process of HEPS multi-objective collaborative optimization, the smoothness of the power mode switching of the hybrid aero-engine is required to meet the preset threshold of aviation flight safety, while minimizing the aviation fuel consumption rate and NOx exhaust emission concentration, and optimizing the core adjustable control parameters of HEPS so that the comprehensive operating efficiency of the hybrid aero-engine under all flight conditions reaches the optimal level. In the construction of the fitness function, conditional statements are used to pre-filter the optimization output results. First, the pre-established mathematical model of HEPS power switching dynamics and energy consumption emission theory is imported for calculation, and the power switching smoothness corresponding to the parameter combination to be evaluated is verified first to meet the preset safety threshold requirements. When the smoothness index meets the preset requirements, the fuel consumption rate, exhaust emission concentration and comprehensive operating efficiency corresponding to the parameter combination are further calculated, and the comprehensive operating efficiency value that integrates fuel consumption and emissions is output. In subsequent iterations, the core control input parameter corresponding to the maximum comprehensive operating efficiency is selected as the optimal solution. If the smoothness index does not meet the preset requirements, the fitness value of the parameter combination is directly set to a minimum invalid value and eliminated. During the iterative process of finding the optimal solution, the convergence and computational accuracy of the genetic algorithm are ensured by limiting the output range of the fitness function to negative numbers. (2) The chromosome is a column vector that stores the three core adjustable control parameters of HEPS. The three core control parameters are: the target speed of the turboshaft engine gas turbine, the power distribution ratio of oil and electricity, and the trigger threshold for engine / motor mode switching. Each element of the column vector corresponds to a specific value of a control parameter, and the value range conforms to the safety operation boundary constraints of HEPS. Construct a zero matrix of appropriate dimensions to store the parent chromosome data during the iteration process of the genetic algorithm; The selection operator uses the roulette wheel selection method, the crossover operator uses the single-point crossover method, and the mutation operator uses the basic bit mutation method. New combinations of design parameters are generated through evolutionary operators such as selection, crossover, and mutation, and their performance is evaluated to guide the search of the next generation. The parent population is continuously updated in multiple iterations to finally obtain the optimal combination of control parameters. (3) Based on the global search capability of the GA genetic algorithm, the parent population is updated iteratively through selection, crossover, and mutation. On the basis of the "key state parameter-control parameter" mapping model constructed by the BP neural network, the optimal solution for multiple objectives is screened. The multi-objective optimal solution is physically represented as a set of control parameters that enable the hybrid propulsion system to achieve the highest overall physical operating efficiency under the current flight load and battery SOC state. This optimal solution directly affects the control system through the following three core physical variables: the target speed of the turboshaft engine gas turbine, the fuel-electric power distribution ratio, and the engine / motor mode switching trigger threshold, which correspond to the chromosome coding of the three core control parameters.

9. The optimized design method for hybrid engine drive switching control based on neural networks according to claim 8, characterized in that, The specific implementation is as follows: (1). Selection operation: The roulette wheel selection method is adopted. The parent chromosome (target speed of turboshaft engine, power distribution ratio of oil and electricity, mode switching trigger threshold) is input and substituted into the fitness function to calculate the comprehensive operating efficiency. The selection probability of a single parent is the comprehensive operating efficiency of the parent / the sum of the comprehensive operating efficiencies of all parents. The dominant parent is randomly selected according to the probability to ensure the inheritance of excellent genes. (2). Crossover operation: The rand function generates 1-2 crossover points, and the two sets of parent chromosomes are truncated at the crossover points and the parameter fragments are exchanged to achieve gene recombination and expand the search range; (3). Mutation operation: The parameters are perturbed by a normal distribution. After perturbation, the parameters must satisfy the HEPS safety boundary. If they exceed the boundary, they are corrected to the nearest boundary value to avoid invalid solutions. (4). Iteration termination: When the change in the optimal fitness value over 10 consecutive generations is ≤10⁻ 4 Alternatively, the iteration can stop after 100 iterations, and the optimal combination of control parameters can be output.

Images