A method and system for energy scheduling of a distributed electric propulsion aircraft

Abstract

The application discloses an energy scheduling method and system of a distributed electric propulsion aircraft, and relates to the field of aviation energy management. The method comprises the following steps: acquiring flight state parameters of each aircraft in a distributed electric propulsion aircraft system; constructing a multi-objective energy scheduling model, with the aim of minimizing the generator fuel consumption cost, fuel quality compensation loss and battery full life cycle equivalent attenuation cost; solving the model by using the flight state parameters through a flexible actor-critic algorithm to obtain energy scheduling instructions containing generator power distribution instructions and battery power distribution instructions; and adjusting each aircraft in the distributed electric propulsion aircraft system according to the energy scheduling instructions to realize energy scheduling. The application can effectively optimize energy distribution, reduce fuel consumption and battery attenuation cost while meeting the operating requirements of the aircraft, and improve energy utilization efficiency and overall system performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aviation energy management, and particularly relates to an energy scheduling method and system for a distributed electric propulsion aircraft. BACKGROUND

[0002] Aircraft onboard systems are facing the transformation pressure of more and full electrification, and the electrification of aircraft power systems has become a key breakthrough direction. Compared with the traditional engine propulsion system, the distributed electric propulsion system has become the core path of power innovation for the next generation of civil aircraft and unmanned aerial vehicles due to its advantages of propulsion efficiency and power-to-weight ratio. However, the distributed electric propulsion system needs to meet the flight envelope constraints (such as airspeed and altitude) and the dynamic operation constraints (such as source-load real-time balance) of the onboard micro-grid during operation, which significantly improves the complexity level of system control.

[0003] Specifically, the dynamic flight environment (such as sudden wind turbulence and flight stage switching) of the distributed aircraft electric propulsion system will cause random fluctuations of electric propulsion power; in addition, the time-varying power demand of non-propulsion loads such as avionics systems, environmental control devices and electric actuators will also bring challenges to the stable operation of the onboard power system. Although the existing rule-based method has millisecond-level dynamic response capability, it is difficult to achieve global energy optimal allocation due to the limitation of local optimization mechanism; the optimization algorithm in the related technology cannot meet the online real-time scheduling demand. Further, the global optimization performance of the energy management strategy of the onboard power system in the related research is insufficient.

[0004] Therefore, it is urgent to build an intelligent scheduling architecture with online adaptive capability to optimize the hybrid energy supply sequence of the fuel generator and the lithium battery, and to ensure the energy efficiency optimization and equipment life balance under the whole task profile. SUMMARY

[0005] The purpose of the present application is to provide an energy scheduling method and system for a distributed electric propulsion aircraft, which can fully consider factors such as generator fuel consumption, fuel quality compensation, and battery full life cycle attenuation, effectively improve energy utilization efficiency, reduce operating costs, and ensure stable and efficient operation of the distributed electric propulsion aircraft in complex environments.

[0006] To achieve the above purpose, the present application provides the following solutions.

[0007] In a first aspect, the present application provides an energy scheduling method for a distributed electric propulsion aircraft, comprising:

[0008] obtaining flight state parameters of each aircraft in a distributed electric propulsion aircraft system;

[0009] According to the flight state parameters, a multi-objective energy scheduling model is constructed, with the objective of minimizing the generator fuel consumption cost, fuel quality compensation loss, and battery full life cycle equivalent attenuation cost of all aircraft in the distributed electric propulsion aircraft system.

[0010] The flight state parameters are input into the trained multi-objective energy scheduling model to obtain energy scheduling instructions; the energy scheduling instructions include generator power distribution instructions and battery power distribution instructions for each aircraft in the distributed electric propulsion aircraft system; the trained multi-objective energy scheduling model is obtained by iteratively training network parameters of the multi-objective energy scheduling model based on a historical training data set through a flexible actor-critic reinforcement learning algorithm until the network parameters corresponding to the previous iteration number are configured to the multi-objective energy scheduling model after the current iteration number exceeds a preset iteration threshold; the historical training data set includes historical flight state parameters and corresponding real energy scheduling instructions.

[0011] Each aircraft in the distributed electric propulsion aircraft system is scheduled through the energy scheduling instructions.

[0012] In a second aspect, the present application provides an energy scheduling system for a distributed electric propulsion aircraft, comprising:

[0013] A flight state parameter acquisition module is configured to acquire flight state parameters of each aircraft in the distributed electric propulsion aircraft system.

[0014] A multi-objective energy scheduling model construction module is configured to construct a multi-objective energy scheduling model based on the flight state parameters, with the objective of minimizing the generator fuel consumption cost, fuel quality compensation loss, and battery full life cycle equivalent attenuation cost of all aircraft in the distributed electric propulsion aircraft system.

[0015] An energy scheduling instruction generation module is configured to input the flight state parameters into the trained multi-objective energy scheduling model to obtain energy scheduling instructions; the energy scheduling instructions include generator power distribution instructions and battery power distribution instructions for each aircraft in the distributed electric propulsion aircraft system; the trained multi-objective energy scheduling model is obtained by iteratively training network parameters of the multi-objective energy scheduling model based on a historical training data set through a flexible actor-critic reinforcement learning algorithm until the network parameters corresponding to the previous iteration number are configured to the multi-objective energy scheduling model after the current iteration number exceeds a preset iteration threshold; the historical training data set includes historical flight state parameters and corresponding real energy scheduling instructions.

[0016] An energy scheduling execution module is configured to schedule each aircraft in the distributed electric propulsion aircraft system through the energy scheduling instructions.

[0017] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the energy scheduling method of the distributed electric propulsion aircraft according to any one of the preceding aspects.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the energy scheduling method of the distributed electric propulsion aircraft according to any one of the preceding aspects.

[0019] In a fifth aspect, the present application provides a computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the steps of the energy scheduling method of the distributed electric propulsion aircraft according to any one of the preceding aspects.

[0020] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0021] The present application provides an energy scheduling method and system of a distributed electric propulsion aircraft, which obtains flight state data of the distributed electric propulsion aircraft, thereby providing basic information for subsequent accurate modeling and scheduling, solves the problem of inaccurate scheduling due to lack of comprehensive state information, and realizes comprehensive control of the operating state of the aircraft; constructs a multi-objective energy scheduling model by taking minimizing the generator fuel consumption cost, fuel quality compensation loss, and battery full life cycle equivalent attenuation cost as the target according to the flight state data, thereby solving the problem that single-target optimization cannot meet the complex energy scheduling requirements, and realizing global optimization considering multiple cost factors; obtains energy scheduling instructions by solving the model using a flexible actor-critic algorithm according to the flight state data, and adjusts the output power of the generator and the battery, thereby solving the problems of insufficient online real-time scheduling and global optimization performance, and realizing scientific, efficient, and real-time energy allocation and scheduling. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0023] Figure 1 FIG. 1 is a schematic diagram of an application environment of an energy scheduling method of a distributed electric propulsion aircraft according to an embodiment of the present application.

[0024] Figure 2 FIG. 2 is a flowchart of an energy scheduling method of a distributed electric propulsion aircraft according to an embodiment of the present application.

[0025] Figure 3 A system architecture diagram of a distributed electric propulsion aircraft is provided for an embodiment of the present application.

[0026] Figure 4 A flight envelope diagram of a distributed electric propulsion aircraft is provided for an embodiment of the present application.

[0027] Figure 5 A flowchart of interactive training of a distributed electric propulsion aircraft environment using a SAC algorithm is provided for an embodiment of the present application.

[0028] Figure 6 A flowchart of implementing power distribution of a distributed electric propulsion aircraft fuel generator and lithium battery using a SAC algorithm is provided for an embodiment of the present application.

[0029] Figure 7 A functional module diagram of an energy scheduling system of a distributed electric propulsion aircraft is provided for an embodiment of the present application.

[0030] Figure 8 A structural diagram of a computer device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0031] First, some technical terms related to the present application are introduced.

[0032] The electrification transformation of the aircraft power system has become an inevitable development trend, and the multi-constraint collaborative power distribution strategy of the fuel generator and the lithium battery is of great significance to realize the energy efficiency optimization of the distributed electric propulsion aircraft in the whole task profile and the dynamic balance of the equipment life.

[0033] Although the optimization algorithm in the related art can construct a multi-time scale model, it cannot meet the online real-time scheduling demand due to the strong coupling characteristics of the system nonlinearity causing dimension disaster. Further, the related research relies on a deterministic static model, lacks robust representation and adaptive ability to nonlinear disturbance, and thus leads to insufficient global optimization performance of the energy management strategy of the airborne power system.

[0034] With the development of sequential decision-making technologies such as reinforcement learning and deep reinforcement learning and the accumulation of historical flight data, real-time energy scheduling methods based on deep reinforcement learning provide a new paradigm for dealing with the uncertainty of distributed electric propulsion aircraft systems. Compared with methods based on deterministic static models, deep reinforcement learning methods can autonomously mine the nonlinear mapping relationship between high-dimensional state space and action space through an end-to-end learning mechanism, and are particularly suitable for the mixed scheduling scenario of fuel generators and lithium batteries in distributed electric propulsion aircraft systems. However, the distributed electric propulsion aircraft energy scheduling scenario has the characteristics of high-dimensional continuous action space and strong coupling of multiple constraints, which poses a serious challenge to deep reinforcement learning algorithms. Existing methods generally face the dual challenges of insufficient strategy exploration efficiency and super parameter sensitivity when dealing with such complex optimization problems, resulting in insufficient convergence stability of the algorithm and difficulty in meeting the real-time optimization requirements under complex flight profiles.

[0035] The purpose of the present application is to address the shortcomings of the above-mentioned technology and provide a distributed electric propulsion aircraft energy scheduling method based on deep reinforcement learning. This method is directly driven by data, and by fusing the randomness model of electric propulsion loads and the time-varying disturbance characteristics of non-propulsion loads, it introduces a maximum entropy framework to enhance the strategy exploration entropy, significantly improving the exploration efficiency and strategy robustness of the agent under complex working conditions, and thus realizing the dynamic collaborative optimization of fuel generators and lithium batteries under the constraints of flight envelopes and the operating boundaries of onboard microgrids.

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0037] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0038] The energy scheduling method of the distributed electric propulsion aircraft provided in the embodiments of the present application can be applied to, for example Figure 1The application environment is shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required by the server 104 to process. The data storage system can be separately arranged, or integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the flight state parameters of each aircraft in the distributed electric propulsion aircraft system to the server 104. After the server 104 receives the flight state parameters of each aircraft in the distributed electric propulsion aircraft system, the server 104 inputs the flight state parameters into the trained multi-objective energy scheduling model according to the flight state parameters, constructs a multi-objective energy scheduling model with the goal of minimizing the generator fuel consumption cost, fuel quality compensation loss and battery full life cycle equivalent attenuation cost of all aircrafts in the distributed electric propulsion aircraft system, and obtains an energy scheduling instruction; The energy scheduling instruction includes the generator power distribution instruction and the battery power distribution instruction of each aircraft in the distributed electric propulsion aircraft system; The trained multi-objective energy scheduling model is obtained by iteratively training the network parameters of the multi-objective energy scheduling model based on the historical training data set through the flexible actor-critic reinforcement learning algorithm until the current iteration number is greater than the preset iteration threshold; The historical training data set includes historical flight state parameters and corresponding real energy scheduling instructions; Through the energy scheduling instruction, each aircraft in the distributed electric propulsion aircraft system is scheduled. The server 104 can feed back the obtained energy scheduling instruction to the terminal 102. In addition, in some embodiments, the energy scheduling method of the distributed electric propulsion aircraft can also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 can directly perform energy scheduling processing on the flight state parameters of each aircraft in the distributed electric propulsion aircraft system, or the server 104 can obtain the flight state parameters of each aircraft in the distributed electric propulsion aircraft system from the data storage system and perform energy scheduling processing on the flight state parameters of each aircraft in the distributed electric propulsion aircraft system.

[0039] The terminal 102 can be, but is not limited to, various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0040] In an exemplary embodiment, as Figure 2 shown, a distributed electric propulsion aircraft energy scheduling method is provided, which is executed by a computer device, specifically by a terminal or a server, etc. Computer device alone, or by a terminal and a server together, in the embodiment of the application, the method is applied toFigure 1 The server 104 in the system 100 is taken as an example for illustration, including the following steps 201 to 204. Among them:

[0041] Step 201, obtaining the flight state parameters of each aircraft in the distributed electric propulsion aircraft system.

[0042] Step 202, constructing a multi-objective energy scheduling model according to the flight state parameters, with the goal of minimizing the generator fuel consumption cost, fuel quality compensation loss and battery full life cycle equivalent attenuation cost of all aircrafts in the distributed electric propulsion aircraft system.

[0043] Step 203, inputting the flight state parameters into the trained multi-objective energy scheduling model to obtain energy scheduling instructions; the energy scheduling instructions include generator power distribution instructions and battery power distribution instructions of each aircraft in the distributed electric propulsion aircraft system; the trained multi-objective energy scheduling model is obtained by iteratively training the network parameters of the multi-objective energy scheduling model based on the historical training data set through the flexible actor-critic reinforcement learning algorithm until the current iteration number is greater than the preset iteration threshold; the historical training data set includes historical flight state parameters and corresponding real energy scheduling instructions.

[0044] Step 204, scheduling each aircraft in the distributed electric propulsion aircraft system through the energy scheduling instructions.

[0045] By implementing the above steps 201 to 204, the present application can effectively solve the problems of insufficient global optimization performance in distributed electric propulsion aircraft energy scheduling, inability to meet online real-time scheduling requirements and the like, and realize scientific management and efficient utilization of energy, thereby providing strong support for the development of green aviation.

[0046] In another exemplary embodiment of the present application, the objective function of the multi-objective energy scheduling model in step 202 is:

[0047] .

[0048] Wherein, represents the objective function value at time t; represents the generator fuel cost in the generator fuel consumption cost at time t; represents the generator start-up cost in the generator fuel consumption cost at time t; represents the fuel quality compensation loss at time t; represents the battery full life cycle equivalent attenuation cost at time t.

[0049] In another example embodiment of the present application, the constraints of the multi-objective energy scheduling model include: a distributed electric propulsion aircraft system power balance constraint, a distributed electric propulsion aircraft system safety redundancy constraint, a generator operation constraint, a battery pack operation constraint, and a distributed electric propulsion aircraft flight constraint.

[0050] In another example embodiment of the present application, the training process of the multi-objective energy scheduling model specifically includes:

[0051] Based on the objective function and the constraints of the multi-objective energy scheduling model, a state space, an action space, and a reward function of a flexible actor-critic intelligent agent model are defined through a Markov decision process framework; the flexible actor-critic intelligent agent model includes a policy network, a value function network, a target value function network, a Q1 network, and a Q2 network; the state space is constructed based on historical flight state parameters; the action space is constructed based on real energy scheduling instructions corresponding to the historical flight state parameters; and the reward function is constructed based on the objective function and the constraints.

[0052] The flexible actor-critic intelligent agent model is initialized, and the initial value of the iteration number i is 1.

[0053] The state space of the current time and the i th iteration is input into the policy network to obtain the action space of the current time and the i th iteration.

[0054] The reward function of the current time and the i th iteration is calculated based on the action space of the current time and the i th iteration, and the state space of the current time and the i +1th iteration corresponding to the reward function of the current time and the i th iteration is calculated.

[0055] A transition sample data is formed by the state space of the current time and the i th iteration, the action space, the reward function, and the state space of the current time and the i +1th iteration, and is stored in an experience replay pool.

[0056] If the number of transition samples stored in the experience replay pool is greater than a preset data capacity of the experience replay pool, and the current time is a time point for updating network parameters, then the network parameters of the flexible actor-critic intelligent agent model are updated through an entropy regularization term in a maximum entropy framework.

[0057] If the number of transition samples stored in the experience replay pool is less than or equal to the preset data capacity of the experience replay pool, and the current time t is greater than or equal to a preset single-cycle termination threshold T, then the iteration numberi = i +1, and returning to the step of obtaining the action space of the state space input policy network of the current time and the current iteration number. i i

[0058] If the number of transition samples stored in the experience replay pool is less than or equal to the preset data capacity of the experience replay pool, and the current time t is less than the preset single-cycle termination threshold T, the next time is taken as the current time, and the step of obtaining the action space of the state space input policy network of the current time and the current iteration number is returned. i i

[0059] If the current iteration number is greater than the preset iteration threshold, the training is ended, and the network parameters of the trained flexible actor-critic intelligent agent model are obtained.

[0060] Based on the network parameters of the trained flexible actor-critic intelligent agent model, the multi-objective energy scheduling model is configured, and the trained multi-objective energy scheduling model is obtained.

[0061] In another exemplary embodiment of the present application, the network parameters of the flexible actor-critic intelligent agent model are updated by dynamically adjusting the policy and the policy gradient through the entropy regularization term in the maximum entropy framework, specifically including:

[0062] The network parameters of the flexible actor-critic intelligent agent model are updated by the following formula:

[0063] .

[0064] .

[0065] .

[0066] .

[0067] .

[0068] .

[0069] .

[0070] wherein, denotes the policy network parameter at t+1 time; denotes the policy network parameter at t time; denotes the policy network learning rate; denotes the gradient operator of the policy network parameter; denotes the objective function of the policy network.​​​​ The network parameters represent the values ​​at time t+1; The network parameters represent the values ​​at time t. The value function represents the network learning rate; The gradient operator represents the value function network parameters; The objective function of the value function network is represented by j = {1, 2}; j = 1 and j = 2 represent the Q1 network and the Q2 network, respectively. This represents the network parameters of the j-th Q-network at time t+1; This represents the network parameters of the j-th Q-network at time t; This represents the learning rates of the Q1 and Q2 networks; This represents the gradient operator for the network parameters of the j-th Q-network; Denotes the objective function of the j-th Q-network; The network parameters represent the objective value function at time t+1; The network parameters represent the objective value function at time t; This represents the network soft update parameters of the objective value function; This represents the mathematical expectation operation, and the states and actions used are the states s at time t sampled from the experience replay pool D. t And based on the policy network's state s at time t t Action a at time t during downsampling t D represents the experience replay pool; Indicates the action at time t; Indicates the state at time t; Represents the entropy temperature coefficient; Represents a policy network; This represents the parameterized j-th Q-network; Representation of value function networks; This represents the entropy regularization term; This represents the mathematical expectation operation, and the states and actions used are state-action pairs sampled from the experience replay pool D at time t; This represents the mathematical expectation operation, and the action used is the action sampled from the parameterized policy network at time t; This represents the mathematical expectation operation, and the state-action-reward-next-state quadruple used is a state-action-reward-next-state quadruple sampled from the experience replay pool D; Indicates the discount factor; Represents the objective value function network; This represents the state at time t+1.

[0071] In another example embodiment of the present application, the flight state data includes flight speed data of the distributed electric propulsion aircraft, flight thrust demand data of the distributed electric propulsion aircraft system, fuel mass data, and battery state of charge data.

[0072] In another example embodiment of the present application, in order to solve the problems of random disturbance of electric propulsion load and time-varying power fluctuation of non-electric propulsion load caused by dynamic flight environment, an energy scheduling method for distributed electric propulsion aircraft based on deep reinforcement learning is proposed based on the SAC (Soft Actor-Critic) algorithm framework, so as to improve the dynamic response capability and energy scheduling efficiency of the distributed electric propulsion system under complex working conditions. The method mainly includes the following steps:

[0073] 1) An energy system model of the distributed electric propulsion aircraft is constructed, which covers an electric propulsion power dynamic demand model of a typical flight mission profile, an operating model of a fuel generator and a lithium battery.

[0074] 2) A multi-objective energy scheduling model of the distributed electric propulsion aircraft is established, which takes minimizing the fuel consumption cost of the generator, the compensation loss of the fuel mass, and the equivalent attenuation cost of the battery throughout the life cycle as the comprehensive optimization objective.

[0075] 3) The multi-objective energy scheduling model is modeled by Markov decision process, and the SAC reinforcement learning algorithm is trained by using historical flight data.

[0076] 4) The agent energy scheduling model trained and converged is deployed to an electric energy scheduling controller, which generates optimal power allocation instructions for the fuel generator and the lithium battery in real time based on real-time flight parameter state input, and drives the dynamic update of the energy system state of the distributed electric propulsion aircraft.

[0077] Specifically, the energy scheduling method for the distributed electric propulsion aircraft based on deep reinforcement learning of the present embodiment includes the following steps:

[0078] Step S1, as shown in Figure 3 The distributed electric propulsion aircraft system is composed of a fuel generator, a lithium battery pack, a power transmission and distribution system, a propulsion motor, a non-propulsion load, and an electric energy scheduling controller. The fuel generator and the lithium battery pack serve as the core power supply system, which provides power support to the load system composed of the electric propulsion load, the environmental control system, the avionics equipment, and the electric actuator through the power transmission and distribution system. The electric energy scheduling controller dynamically generates optimal scheduling instructions for energy equipment based on real-time sensing and intelligent analysis of multi-source operating state data. The thrust of the key node can be calculated by the aircraft dynamics formula, and the corresponding instruction demand thrust of the flight profile can be given by fitting the curve.

[0079] The embodiment constructs a distributed electric propulsion aircraft energy system model according to the equipment objects contained in the distributed electric propulsion aircraft system, covering an electric propulsion power dynamic demand model of a typical flight mission profile, a fuel generator operation model and a lithium battery operation model:

[0080] Electric propulsion power dynamic model:

[0081] .

[0082] Fuel generator operation model:

[0083] .

[0084] Lithium battery operation model:

[0085] .

[0086] .

[0087] .

[0088] .

[0089] .

[0090] .

[0091] wherein, represents the electric propulsion power of the distributed electric propulsion aircraft system at time t; represents the thrust of the electric propulsion motor at time t; represents the flight speed of the electric propulsion aircraft at time t; represents the propulsion efficiency of the electric propulsion motor; represents the output power of the fuel generator at time t; represents the start-stop state of the fuel generator at time t; represents the output power load factor of the fuel generator at time t; represents the maximum power of the fuel generator ; represents the total number of fuel generators; represents the state of charge of the lithium battery pack at time t; represents the real-time capacity of the lithium battery pack at time t; represents the rated capacity of the lithium battery pack; represents the initial capacity of the lithium battery pack; represents the charging current of the lithium battery pack at time t; represents the open-circuit voltage of the lithium battery pack; is the resistance of the lithium battery pack; is the charging power of the lithium battery pack at time t, which is positive for discharging and negative for charging; is the charging efficiency of the lithium battery pack; is the discharging efficiency of the lithium battery pack; is the number of series-connected battery cells in the lithium battery pack; is the open-circuit voltage of the lithium battery cell; is the number of parallel-connected battery cells in the lithium battery pack; is the rated capacity of the lithium battery cell; is the resistance of the lithium battery cell.

[0092] Step S2, a multi-objective energy scheduling model of the distributed electric propulsion aircraft is established, with the minimum generator fuel consumption cost, fuel quality compensation loss, and battery full life cycle equivalent attenuation cost as the comprehensive optimization objectives. The generator operating cost includes fuel cost and start-up cost. The fuel quality compensation loss is represented by the equivalent fuel cost of the fuel consumption increment caused by the additional fuel weight compensated by the electric propulsion aircraft. The battery full life cycle equivalent attenuation cost is represented by the equipment wear and tear and depreciation cost caused by battery charging and discharging. The uncertainty quantification of the multi-objective energy scheduling model includes: 1) electric propulsion load uncertainty caused by dynamic flight environment, and 2) time-varying power disturbance of non-electric propulsion loads on the aircraft (such as environmental control systems, avionics, and electric actuation systems). The system operation constraints included in the multi-objective energy scheduling model mainly include: distributed electric propulsion aircraft system power balance constraint, safety redundancy constraint, fuel generator operation constraint, lithium battery operation constraint, and flight constraint of the distributed electric propulsion aircraft.

[0093] As an optional implementation, the calculation formula of the fuel cost of the fuel generator is as follows:

[0094] .

[0095] .

[0096] The calculation formula of the start-up cost of the fuel generator is as follows:

[0097] .

[0098] .

[0099] The calculation formula of the fuel quality compensation loss is as follows:

[0100] .

[0101] .

[0102] The formula for calculating the equivalent degradation cost over the entire battery lifecycle is as follows:

[0103] .

[0104] in, This represents the fuel cost of the fuel generator at time t. For fuel prices; fuel generator at time t Instantaneous fuel consumption rate; Indicates a preset time interval; fuel generator ; It is a fuel with a low calorific value; This represents the starting cost of the fuel generator at time t; fuel generator at time t The start flag variable; fuel generator Startup costs; This represents the fuel mass compensation loss of the fuel generator at time t; fuel generator Fuel compensation quality; This represents the equivalent degradation cost of the battery over its entire lifespan at time t. This is the fuel quality compensation coefficient; This is the depreciation factor for lithium battery packs.

[0105] As an optional implementation, the power balance constraint of the distributed electric propulsion aircraft system is:

[0106] .

[0107] The safety redundancy constraints for distributed electric propulsion aircraft systems are:

[0108] .

[0109] .

[0110] .

[0111] The operating constraints for fuel generators are:

[0112] .

[0113] .

[0114] .

[0115] The operating constraints of the lithium battery pack are:

[0116] .

[0117] .

[0118] .

[0119] The flight constraints for a distributed electric propulsion aircraft are:

[0120] .

[0121] .

[0122] .

[0123] .

[0124] in, Let be the electric propulsion power of the distributed electric propulsion aircraft system at time t; Let t be the non-electric propulsion power of the distributed electric propulsion aircraft system at time t; Let be the standby power of the fuel generator at time t; Let t be the reserve power of the lithium battery pack at time t; The minimum propulsion power required for a distributed electric propulsion aircraft system; The minimum non-propulsion power required for a distributed electric propulsion aircraft system; This represents the maximum discharge power of the lithium battery pack. fuel generator Minimum output power; fuel generator The rate of climb; Let be the fuel mass at time t; For emergency fuel reserves; The ramp-up rate for lithium battery packs; This represents the minimum state of charge (SOC) of the lithium battery pack. This represents the maximum state of charge (SOC) of the lithium battery pack. The minimum safe speed for a distributed electric propulsion aircraft; This is the maximum permissible speed for a distributed electric propulsion aircraft. Flight altitude; For the angle of attack of flight; Speed ​​is limited by the strength of the aerodynamic structure; Thermodynamically limited velocity; the flight envelope diagram of this embodiment is shown below. Figure 4 As shown. electric propulsion motor The lower limit of thrust, electric propulsion motor upper limit of the thrust of the electric propulsion motor, for the electric propulsion motor thrust, total thrust of the electric propulsion motor at time t, total number of electric propulsion motors.

[0125] Step S3, the multi-objective energy scheduling model of step S2 is modeled by Markov decision process, and the network parameter training of SAC algorithm is performed by using historical flight data. Based on the analysis of the time sequence of the flight state parameters in the historical flight data, the SAC algorithm is trained offline, and when the training iteration number reaches the preset threshold, the converged agent network parameters are stored.

[0126] Step S31, according to the flight state parameter information (including speed, height, battery SOC and load power demand, etc.), control variables and optimization objectives of the multi-objective energy scheduling model of step S2, Markov decision process modeling is performed, and the state space, action space and reward function of the SAC reinforcement learning agent are defined. The SAC algorithm of the present application dynamically adjusts the entropy of the policy distribution information through the entropy regularization term in the maximum entropy framework, so that the exploration efficiency of the agent is improved under the composite working condition of sudden wind disturbance and load fluctuation, and the double Q function of the Critic network is used to suppress the overestimation of the value function, and the stability of the agent training is improved through the network parameter soft update strategy. The specific agent model components of the embodiment include:

[0127] 1) State space S:

[0128] .

[0129] Wherein, represents the state at time t; is the flight speed of the distributed electric propulsion aircraft at time t; is the thrust of the distributed electric propulsion motor at time t; is the non-electric propulsion power of the distributed electric propulsion aircraft system at time t; is the fuel mass at time t; is the state of charge of the lithium battery at time t, is the current time value.

[0130] 2) Action space A:

[0131] .

[0132] Wherein, is the output power increment of the fuel generator at time t; is the output power of the lithium battery at time t; and is the action selected by the agent at time t, and the dimension of the action space is is the total number of fuel generators; is the maximum ramp rate of the fuel generator ; represents the output power increment of the lithium battery pack at time t; is the maximum ramp rate of the lithium battery pack.

[0133] 3) Instantaneous reward function r:

[0134] .

[0135] wherein, represents the reward value at time t; is the scaling coefficient of the reward function; is the penalty value for violating the operating constraints in steps S1 and S2 at time t.

[0136] Step S32, based on the historical flight data, the flight state parameter time sequence is analyzed, and the SAC algorithm is trained offline. When the training iteration number reaches the preset threshold, the converged agent network parameters are stored, and the training process of the SAC agent is as shown in Figure 5 , and the specific steps are as follows:

[0137] 1) The time sequence parameter set is parsed and extracted from the historical flight data, including the flight speed of the electric propulsion aircraft, the flight thrust demand of the electric propulsion system, the fuel quality , and the lithium battery state of charge , and the agent training set is constructed.

[0138] 2) Based on the energy equipment configuration of the electric propulsion aircraft, the dimension of the agent action space is defined as dim(A)=N DG +1.

[0139] 3) Initialize the policy network and the value function network , create the target value function network , and define two independent Q networks: Q1 network and Q2 network ; set the training hyperparameters of the agent, including the policy network learning rate α π , the value function network learning rate α V (the learning rates of the value function network and the target value function network are the same), and the Q network learning rate α ​Q (Q1 network and Q2 network learning rate gamma , discount factor β , entropy temperature coefficient tau , target value function network soft update parameter k , network parameter update interval M , preset iteration threshold T , monotonicity cycle termination time threshold D , experience replay pool W , preset data capacity B , batch size

[0140] 4) Set the current iteration number i = 1, if i ≤ M , start training from the initial time of the scheduling period t = t 0 environment state s t (namely, the flight state data of the distributed electric propulsion aircraft system at time t), otherwise jump to step 9).

[0141] 5) The agent interacts with the distributed electric propulsion aircraft energy system model environment, and the agent generates continuous actions s t after receiving state information a t = π Φ ( s t ) through the policy network, and the distributed electric propulsion aircraft energy system model executes the action a t , and feeds back the reward function r t and the next state s t+1 to the agent, and stores the four-tuple data s t , a t , r t , s t+1 in the experience replay pool D .

[0142] 6) When the number of transition samples stored in the experience replay pool D satisfies D |> W and t % kWhen == 0, the agent network parameters are updated using the following formula; otherwise, proceed to step 7).

[0143] .

[0144] .

[0145] .

[0146] .

[0147] .

[0148] .

[0149] .

[0150] in, Represents the policy network parameters at time t+1; Represents the policy network parameters at time t; Indicates the network learning rate of the policy; The gradient operator represents the parameters of the policy network; Represents the objective function of the policy network; The network parameters represent the values ​​at time t+1; The network parameters represent the values ​​at time t. The value function represents the network learning rate; The gradient operator represents the value function network parameters; The objective function of the value function network is represented by j = {1, 2}; j = 1 and j = 2 represent the Q1 network and the Q2 network, respectively. This represents the network parameters of the j-th Q-network at time t+1; This represents the network parameters of the j-th Q-network at time t; This represents the learning rates of the Q1 and Q2 networks; This represents the gradient operator for the network parameters of the j-th Q-network; Denotes the objective function of the j-th Q-network; The network parameters represent the objective value function at time t+1; The network parameters represent the objective value function at time t; This represents the network soft update parameters of the objective value function; This represents the mathematical expectation operation, and the states and actions used are the states s at time t sampled from the experience replay pool D. t And based on the policy network's state s at time t t Action a at time t during downsampling tD represents an experience replay pool; represents an action at time t; represents a state at time t; represents an entropy temperature coefficient; represents a policy network; represents a j-th parameterized Q network; represents a value function network; represents an entropy regularization term; represents a mathematical expectation operation, and the state and action used are a state-action pair at time t sampled from the experience replay pool D; represents a mathematical expectation operation, and the action used is an action at time t sampled from the parameterized policy network; represents a mathematical expectation operation, and the state-action-reward-next state four-tuple used is a state-action-reward-next state four-tuple sampled from the experience replay pool D; represents a discount factor; represents a target value function network; represents a state at time t+1.

[0151] The target of the policy network is to maximize the reward r t and the entropy (- ), which is equivalent to minimizing the policy network objective function . The entropy regularization term is introduced in the policy network objective function formula to reflect the influence of the entropy term on the policy parameter update, and the policy network parameters are updated by gradient based on the minimization of the policy network objective function formula (the first formula of this step ), so as to dynamically adjust the information entropy of the policy distribution and make the exploration of the agent more diversified; in order to reduce the overestimation of the value function, two independent Critic Q networks are introduced in this step, and the minimum value is taken when calculating the Q value, which can effectively reduce the deviation of Q value overestimation and significantly improve the stability of agent training.

[0152] 7) If the current time t reaches the single-cycle termination threshold T , terminate the current scheduling period training, update the iteration number i = i +1, and jump to step 4), otherwise jump to step 8).

[0153] 8) update the state s t = s t+1 and time t=t+1, jump to step 5) to continue the agent network parameter training in the current scheduling period.

[0154] 9) The training of the agent is completed, and the network parameters of the converged agent are stored, including the policy network , the value function network , the target value function network , and the two sets of Q networks , .

[0155] Step S4, as shown in Figure 6 , the agent energy scheduling model trained to converge is deployed to the electric energy scheduling controller, and based on the real-time flight parameter state input, the optimal power distribution instructions of the fuel generator and the lithium battery are generated in real time, and the distributed electric propulsion aircraft energy system state is dynamically updated, and the deployment process includes:

[0156] 1) Load the network model parameters of the trained and converged agent in the electric energy scheduling controller, including the policy network parameters , the value function network parameters , and , the Q network parameters , and initialize the starting time t = t0 of the current period.

[0157] 2) Real-time acquisition of flight environment state information s t at the current time t is input to the agent policy network , and the power distribution instructions of the electric propulsion aircraft fuel generator and lithium battery and are generated in real time r t and the next state s t+1 after executing the action instruction.

[0158] 3) If the current time t satisfies the scheduling period termination condition t-t0≥ T , then jump to step 4), otherwise update the state s t = s t+1 and the time t = t+1, jump to step 2).

[0159] 4) Terminate the current scheduling period, and output the cumulative single-step reward value .

[0160] Step S5, determine whether the current time reaches the scheduling period termination threshold of the flight mission profile, if the condition is met, output the action instruction sequence and terminate the current period, otherwise jump to S4 to execute the next time scheduling.

[0161] The application also provides an application scenario of the energy scheduling method of the distributed electric propulsion aircraft. Specifically, the energy scheduling method of the distributed electric propulsion aircraft provided in this embodiment can be applied in a scenario of performing long-distance cross-island material transportation and monitoring tasks by the distributed electric propulsion aircraft. The scenario includes a pre-task preparation and route planning link, material transportation, island monitoring and real-time dynamic energy scheduling in the flight process, and a data arrangement and equipment inspection and maintenance link after the task is completed. The aircraft enters a task execution state from the pre-task preparation and route planning link, obtains information such as completion of material transportation, island monitoring data and energy use optimization record through material transportation, island monitoring and real-time dynamic energy scheduling in the flight process, and enters the data arrangement and equipment inspection and maintenance link after the task is completed.

[0162] The energy scheduling method of the distributed electric propulsion aircraft provided in this embodiment belongs to a core link in the material transportation, island monitoring and real-time dynamic energy scheduling in the flight process. Specifically, in the process of performing long-distance cross-island material transportation and monitoring tasks by the aircraft, in the face of complex weather conditions, various task requirements and dynamically changing energy requirements, the method constructs a multi-objective energy scheduling model by real-time acquisition of flight state data such as flight speed, height, remaining power and material weight. By using a deep reinforcement learning algorithm, the agent can autonomously mine the nonlinear mapping relationship between the high-dimensional state space and the action space, innovatively introduce a maximum entropy framework to enhance the strategy exploration entropy, and significantly improve the exploration efficiency and strategy robustness under complex working conditions. For example, when encountering strong sea winds that increase flight resistance, the algorithm can quickly adjust the energy supply ratio of the fuel generator and the lithium battery, preferentially guarantee the energy supply of the propulsion system, and ensure the stable flight of the aircraft; when monitoring islands, the energy is reasonably allocated to meet the normal operation requirements of the monitoring equipment. In this way, dynamic collaborative optimization of the fuel generator and the lithium battery under the flight envelope constraints and the operating boundary of the onboard microgrid is realized, the task is ensured to be completed smoothly, and efficient use of energy and balanced equipment life are realized.

[0163] The application has the following advantages:

[0164] 1) Compared with the traditional deterministic static model method, the application can realize end-to-end dynamic control from real-time flight parameter state data to hybrid energy equipment, solve the influence of random uncertainty of electric propulsion loads and time-varying power disturbance of non-electric propulsion loads on energy scheduling optimization under dynamic flight conditions, and guarantee the collaborative power distribution of the fuel generator and the lithium battery under the flight envelope constraints and the operating boundary of the onboard microgrid.

[0165] 2) Unlike traditional deep reinforcement learning methods, the maximum entropy framework constructed in the application realizes the adaptive adjustment of the information entropy level of the strategy distribution under the combined working conditions of sudden disturbance and load fluctuation by introducing a strategy entropy regularization term, significantly improving the exploration efficiency of the agent. At the same time, the dual Q function estimator collaborative optimization architecture based on the Critic network effectively improves the convergence stability of the strategy training by systematically suppressing the value function overestimation phenomenon combined with the soft update mechanism of the network parameters.

[0166] Based on the same inventive concept, the embodiments of the application also provide an energy scheduling system of a distributed electric propulsion aircraft for implementing the energy scheduling method of the distributed electric propulsion aircraft as described above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme described in the above method, so the specific limitations in one or more energy scheduling system embodiments of the distributed electric propulsion aircraft provided below can refer to the limitations of the energy scheduling method of the distributed electric propulsion aircraft described above, which will not be repeated here.

[0167] In one exemplary embodiment, as shown in Figure 7 An energy scheduling system of a distributed electric propulsion aircraft is provided, including:

[0168] A flight state data acquisition module 301 is configured to acquire flight state data of the distributed electric propulsion aircraft.

[0169] A multi-objective energy scheduling model construction module 302 is configured to construct a multi-objective energy scheduling model based on the flight state data, with the objective of minimizing the generator fuel consumption cost, the fuel mass compensation loss, and the battery full life cycle equivalent attenuation cost.

[0170] An energy scheduling instruction generation module 303 is configured to input the flight state parameters into the trained multi-objective energy scheduling model to obtain an energy scheduling instruction; the energy scheduling instruction includes a generator power distribution instruction and a battery power distribution instruction for each aircraft in the distributed electric propulsion aircraft system; the trained multi-objective energy scheduling model is obtained by iteratively training the network parameters of the multi-objective energy scheduling model based on a historical training data set through a flexible actor-critic reinforcement learning algorithm until the network parameters corresponding to the previous iteration number are configured to the multi-objective energy scheduling model after the current iteration number reaches a preset iteration number threshold; the historical training data set includes historical flight state parameters and corresponding real energy scheduling instructions.

[0171] An energy scheduling execution module 304 is configured to adjust the generator output power and the battery output power in the distributed electric propulsion system through the energy scheduling instruction to schedule the energy of the distributed electric propulsion aircraft.

[0172] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 8 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores flight status information and energy scheduling instructions for the distributed electric propulsion aircraft. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an energy scheduling method for a distributed electric propulsion aircraft.

[0173] Those skilled in the art will understand that Figure 8 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0174] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0175] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0176] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0177] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magneto resistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0178] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0179] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0180] The principles and implementation manners of the present application are described herein by using specific examples, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will have changes. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method of energy scheduling for a distributed electrically propelled aircraft, characterized in that, The energy scheduling method of the distributed electric propulsion aircraft comprises: acquiring flight state parameters of each aircraft in a distributed electric propulsion aircraft system; constructing a multi-objective energy scheduling model aiming to minimize the generator fuel consumption cost, the fuel mass penalty loss and the battery full life cycle equivalent attenuation cost of all aircrafts in the distributed electric propulsion aircraft system according to the flight state parameters; inputting the flight state parameters into the trained multi-objective energy scheduling model to obtain energy scheduling instructions; the energy scheduling instructions comprise generator power distribution instructions and battery power distribution instructions of each aircraft in the distributed electric propulsion aircraft system; the trained multi-objective energy scheduling model is obtained by iteratively training network parameters of the multi-objective energy scheduling model based on a historical training data set through a flexible actor-critic reinforcement learning algorithm until the network parameters corresponding to a previous iteration number are configured to the multi-objective energy scheduling model after the current iteration number is greater than a preset iteration threshold; the historical training data set comprises historical flight state parameters and corresponding real energy scheduling instructions; scheduling each aircraft in the distributed electric propulsion aircraft system through the energy scheduling instructions.

2. The energy scheduling method of a distributed electrically propelled aircraft according to claim 1, characterized in that, The objective function of the multi-objective energy scheduling model is: ； wherein, represents the target function value at time t; represents the generator fuel cost in the generator fuel consumption cost at time t; represents the generator start-up cost in the generator fuel consumption cost at time t; represents the fuel quality penalty loss at time t; represents the battery full life cycle equivalent degradation cost at time t.

3. The energy scheduling method of a distributed electrically propelled aircraft according to claim 1, characterized in that, The constraint conditions of the multi-objective energy scheduling model comprise power balance constraints and safety redundancy constraints of the distributed electric propulsion aircraft system, generator operation constraints, battery pack operation constraints and flight constraints of each aircraft in the distributed electric propulsion aircraft system.

4. The energy scheduling method of a distributed electrically propelled aircraft according to claim 1, characterized in that, The training process of the multi-objective energy scheduling model specifically comprises: defining a state space, an action space and a reward function of a flexible actor-critic intelligent agent model through a Markov decision process framework based on the objective function and the constraint conditions of the multi-objective energy scheduling model; the flexible actor-critic intelligent agent model comprises a policy network, a value function network, a target value function network, a Q1 network and a Q2 network; the state space is constructed based on historical flight state parameters; the action space is constructed based on real energy scheduling instructions corresponding to the historical flight state parameters; the reward function is constructed based on the objective function and the constraint conditions; initializing a flexible actor-critic agent model, and setting an initial value of an iteration number i to 1; obtain the state space input policy network of the first iteration at the current moment i obtain the action space of the first iteration at the current moment i ​ Based on the state space of the current time and the action space of the current time i Based on the state space of the current time and the action space of the current time i Based on the state space of the current time and the action space of the current time i Based on the state space of the current time and the action space of the current time i Based on the state space of the current time and the action space of the current time The state space of the current time t i , the action space, the reward function, and the state space of the next iteration t i +1 constitute a transition sample data, which is stored in the experience replay pool. if the number of transition samples stored in the experience replay pool is greater than a preset data capacity of the experience replay pool and the current time is a time point for updating network parameters, then the network parameters of the flexible actor-critic intelligent agent model are updated through an entropy regularization term in a maximum entropy framework to dynamically adjust a policy and a policy gradient; If the number of transition samples stored in the experience replay pool is less than or equal to the preset data capacity of the experience replay pool, and the current time t is greater than or equal to the preset single-cycle termination threshold T, the iteration number is set to i = i + 1, and the step "obtaining the state space input policy network of the current time t and the iteration number n" is returned. i i " is returned.​ If the number of transition samples stored in the experience replay pool is less than or equal to the preset data capacity of the experience replay pool, and the current time t is less than the preset single-cycle termination threshold T, the next time is taken as the current time, and the step "obtaining the state space input policy network of the current time t and the first iteration" is returned. i "obtaining the action space of the current time t and the first iteration" is performed. i "obtaining the action space of the current time t and the first iteration" is performed. if the current iteration number is greater than the preset iteration threshold, then the training is ended and the network parameters of the trained flexible actor-critic intelligent agent model are obtained; the multi-objective energy scheduling model is configured based on the network parameters of the trained flexible actor-critic intelligent agent model to obtain the trained multi-objective energy scheduling model.

5. The energy scheduling method of a distributed electrically propelled aircraft according to claim 4, characterized in that, The network parameters of the flexible actor-critic intelligent agent model are updated through an entropy regularization term in a maximum entropy framework to dynamically adjust a policy and a policy gradient, specifically comprising: the network parameters of the flexible actor-critic intelligent agent model are updated through the following formula: ； ； ； ； ； ； ； in, Represents the policy network parameters at time t+1; Represents the policy network parameters at time t; Indicates the network learning rate of the policy; The gradient operator represents the parameters of the policy network; Represents the objective function of the policy network; The network parameters represent the values ​​at time t+1; The network parameters represent the values ​​at time t. The value function represents the network learning rate; The gradient operator represents the value function network parameters; The objective function of the value function network is represented by j = {1, 2}; j = 1 and j = 2 represent the Q1 network and the Q2 network, respectively. This represents the network parameters of the j-th Q-network at time t+1; This represents the network parameters of the j-th Q-network at time t; This represents the learning rates of the Q1 and Q2 networks; This represents the gradient operator for the network parameters of the j-th Q-network; Denotes the objective function of the j-th Q-network; The network parameters represent the objective value function at time t+1; The network parameters represent the objective value function at time t; This represents the network soft update parameters of the objective value function; This represents the mathematical expectation operation, and the states and actions used are the states s at time t sampled from the experience replay pool D. t And based on the policy network's state s at time t t Action a at time t during downsampling t D represents the experience replay pool; Indicates the action at time t; Indicates the state at time t; Represents the entropy temperature coefficient; Represents a policy network; This represents the parameterized j-th Q-network; Representation of value function networks; This represents the entropy regularization term; This represents the mathematical expectation operation, and the states and actions used are state-action pairs sampled from the experience replay pool D at time t; This represents the mathematical expectation operation, and the action used is the action sampled from the parameterized policy network at time t; This represents the mathematical expectation operation, and the state-action-reward-next-state quadruple used is a state-action-reward-next-state quadruple sampled from the experience replay pool D; denotes a discount factor; denotes a target value function network; denotes the state at time t+1.

6. The energy scheduling method of a distributed electrically propelled aircraft according to claim 1, characterized in that, The flight state parameters include flight speed, flight thrust demand, fuel mass and battery state of charge of each electric propulsion aircraft in the distributed electric propulsion aircraft system.

7. An energy scheduling system for a distributed electric propulsion aircraft, characterized in that, The energy scheduling system of the distributed electric propulsion aircraft applies the energy scheduling method of the distributed electric propulsion aircraft according to any one of claims 1-6, and the energy scheduling system of the distributed electric propulsion aircraft comprises: a flight state parameter acquisition module configured to acquire flight state parameters of each aircraft in the distributed electric propulsion aircraft system; a multi-objective energy scheduling model construction module configured to construct a multi-objective energy scheduling model based on the flight state parameters, with the objective of minimizing generator fuel consumption cost, fuel mass compensation loss and battery full life cycle equivalent attenuation cost of all aircraft in the distributed electric propulsion aircraft system; an energy scheduling instruction generation module configured to input the flight state parameters into the trained multi-objective energy scheduling model to obtain an energy scheduling instruction; the energy scheduling instruction includes generator power distribution instructions and battery power distribution instructions of each aircraft in the distributed electric propulsion aircraft system; the trained multi-objective energy scheduling model is obtained by iteratively training network parameters of the multi-objective energy scheduling model based on a historical training data set through a flexible actor-critic reinforcement learning algorithm until the current iteration number reaches a preset iteration number threshold, and then configuring the network parameters corresponding to the previous iteration number to the multi-objective energy scheduling model; the historical training data set includes historical flight state parameters and corresponding real energy scheduling instructions; an energy scheduling execution module configured to schedule each aircraft in the distributed electric propulsion aircraft system through the energy scheduling instruction.

8. A computer device comprising: A memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the energy scheduling method of the distributed electric propulsion aircraft according to any one of claims 1-6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the energy scheduling method of the distributed electric propulsion aircraft according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the energy scheduling method of the distributed electric propulsion aircraft according to any one of claims 1-6.

Images