CN-121983939-A - Aircraft energy system combination control method considering topology and parameter changes

CN121983939ACN 121983939 ACN121983939 ACN 121983939ACN-121983939-A

Abstract

The invention belongs to the technical field of direct current micro-grid energy system control, and discloses an aircraft energy system combination control method considering topology and parameter changes, which comprises the following steps of establishing a multi-bus near space aircraft energy system compact model; the distributed secondary controller is designed to reduce the influence of connecting wire resistance on current sharing precision, simultaneously express parameter uncertainty and reinforcement learning agent output action as polyhedral vertices, solve the controller to ensure system robustness, and design charge state balancing strategy based on multi-agent reinforcement learning to realize charge state balancing of all distributed power generation units. The invention adopts the aircraft energy system combination control method considering topology and parameter changes, and realizes voltage stabilization, current sharing and charge state equalization of the energy system while considering topology switching and parameter uncertainty.

Inventors

ZHANG ZHICHENG
Hu Linshen
LI PENG
ZUO ZHIQIANG
WANG YIJING

Assignees

天津大学

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (10)

1. An aircraft energy system combination control method taking topology and parameter changes into account is characterized by comprising the following steps: step S1, based on the actual light energy storage source system and the adjacent space environmental characteristics, establishing a compact model of the multi-bus adjacent space aircraft energy source system comprising primary control and secondary control inputs; Step S2, designing a distributed state feedback primary controller for realizing voltage tracking control of an energy system according to a robust control theory; S3, designing a distributed secondary controller, reducing the influence of the resistance of a connecting wire on the current sharing precision, simultaneously expressing the uncertainty of parameters and the output action of the reinforcement learning agent as polyhedral vertexes, and solving the controller to ensure the robustness of the system; and S4, designing a charge state balancing strategy based on multi-agent reinforcement learning, wherein each agent only observes state information of adjacent nodes and outputs a droop coefficient to adjust current, so that charge state balancing of all distributed generation units DGU is realized.
2. The method of combined control of an aircraft energy system taking into account topology and parameter variations according to claim 1, wherein in step S1 a compact model of a multi-bus near space aircraft energy system comprising primary and secondary control inputs is built as follows: step S11, a dynamic equation of the circuit is established by applying kirchhoff' S law, and the dynamic equation is as follows: ; Wherein, the Is that The output voltage of (a) is set; Is that Is a capacitor of (a); Is that Is provided; Is in combination with A set of electrically connected neighbor nodes; Is that Flow direction Is set to be a current of (a); is a constant power load; Is a resistive load; Is that Is set in the memory cell; step S12, loading constant power The model is at the voltage working point Linearization is performed as follows: ; Wherein, the Is equivalent current; Is an approximation error; step S13, in the energy system, droop control is adopted to realize current sharing and voltage stabilization, and the droop control is adopted to obtain Voltage tracking value of (2) The following is shown: ; Wherein, the Is a sagging coefficient; output to the secondary controller to realize current sharing and voltage regulation between DGUs; Step S14, order Wherein Representing a voltage tracking error integral term, and combining a circuit dynamic equation, a constant power load model and droop control to obtain a compact model of the energy system of the near space vehicle; Wherein the established compact model of the energy system of the spacecraft incorporates the primary controller output And a secondary controller output The following is shown: ; Wherein, the A derivative of the state variable with respect to time; Is a state variable; Is a disturbance; Is a system matrix; Is a primary control input matrix; is a secondary control input matrix; is an interference input matrix; Is an output matrix; is an output variable.
3. The method according to claim 1, wherein in step S2, a distributed state feedback primary controller is designed for implementing voltage tracking control of the energy system, and the specific process is as follows: step S21, the primary controller adopts a state feedback controller, and the state feedback controller is as follows: ; Wherein, the Representing the primary controller gain; s22, regarding the current on the connecting line between DGUs as disturbance, and solving the gain of the primary controller according to a robust control theory and combining a pole region allocation method; setting the output of the secondary controller 0, Gain Obtained by solving the following linear matrix inequality: ; ; ; Wherein, the Is a positive definite matrix; is a positive scalar variable corresponding to the interference suppression level; representing a system matrix of the closed-loop system; Is a unit matrix; Is a radius; values of the circle center abscissa; Representing a transpose operation; the inequality above minimizes disturbances while allocating the energy system closed loop pole to one Radius (0- ) A circular area as the center of a circle The following is shown: ; Wherein, the A circular region that is a complex plane; Is a closed loop system pole; Conjugate poles for closed loop systems; Is a complex domain.
4. The method according to claim 1, wherein in step S3, a distributed secondary controller is designed to reduce the influence of the connection line resistance on the current sharing precision, and simultaneously, the parameter uncertainty and the reinforcement learning agent output action are represented as polyhedral vertices, and the controller is solved to ensure the system robustness, wherein the method comprises the following steps: Step S31, the secondary controller adopts a distributed PI controller, and outputs control quantity according to average voltage and current errors so as to eliminate voltage deviation and realize accurate current sharing, wherein the control quantity is as follows: ; ; Wherein, the Output for the secondary controller; an integrating part for the secondary controller; is a voltage regulation term; is a current sharing item; Is a voltage regulating item Is a coefficient of integration of (a); To correspond to current sharing item Is a coefficient of integration of (a); Representing time; Is a voltage regulation item Is a proportional coefficient of (2); Is corresponding to the current sharing item Proportional coefficient of (2), voltage regulation term And a current sharing term The following is shown: ; Wherein, the Is that The output voltage of (a) is set; Is that A sag factor in (2); Is that Is provided; Is in combination with A neighbor node set with communication connection; elements of the adjacency matrix for the communication link graph; The degree of the communication link diagram; And A sequence number indicating DGU; And S32, fusing a closed loop system comprising the primary controller with the secondary controller, and constructing a new energy system model according to quasi-stable linear approximation of the power line.
5. The method according to claim 4, wherein in step S32, a new energy system model is constructed, and the specific procedure is as follows: step S321, according to quasi-stationary line approximation of the power line, Flow direction Is (1) the current of the (a) The following is shown: ; Wherein, the Is that And (3) with A line resistance therebetween; Is an approximation error; step S322, define new state variables The following is shown: ; from the compact model of the spacecraft energy system, the primary controller, the secondary controller, and the quasi-stationary line approximations, one can obtain: ; Wherein, the Is the derivative of the state variable; Is a state variable; Is a control input; is a disturbance input; Is a system matrix; for controlling the input matrix; is an interference input matrix; Is an output matrix; Is an output variable; Step S323, consider sagging coefficient Inductance And a capacitor Uncertainty of parameters, the parameter uncertainty is expressed as a polyhedral vertex using polyhedral theory, as follows: ; ; ; the energy system has eight vertices, each of which is composed of Representation of wherein The following is shown: ; ; ; ; step S324, for each vertex System matrix 、、 And Re-labeling from polyhedral vertices 、 And Obtaining secondary controller parameters by solving the following linear matrix inequality The following is shown: ; ; ; Wherein, the Is a positive definite matrix; is a positive scalar variable; To obtain voltage regulation term And a current sharing term Coefficient matrix of (a); the inequality ensures that the pole of the closed loop system is at radius The center of the circle is ) Is within the center region of the circle.
6. The method for controlling the combination of the aircraft energy systems according to claim 1, wherein in the step S4, a charge state balancing strategy for multi-agent reinforcement learning is designed to realize the charge state balancing of all DGUs, and the charge state of each DGU is estimated by an ampere-hour integration method as follows: ; Wherein, the And Respectively representing the current value and the initial value of the DGU charge state; is battery capacity; When sag control realizes current sharing, there are By dynamically adjusting sagging coefficients Regulating the output current Thereby realizing charge state balance control; all local loads are equivalent to one total current for agent training, as follows: ; Wherein, the Representing the total number of DGUs; thus, the ampere-hour integration method is restated as: ; the charge state balance problem is translated into the following constrained optimization problem, as follows: ; Wherein, the ; And The upper and lower limits of the sag coefficient are respectively consistent with the vertex value of the polyhedron, and var (·) represents variance.
7. The method for controlling the combination of the energy systems of the aircraft, which takes the topology and parameter changes into consideration, according to claim 6, is characterized by providing a distributed multi-agent reinforcement learning charge state balance strategy to realize the optimization of charge state balance, and specifically comprises the following steps: Observation space of each agent The following is shown: ; Wherein, the Representing The observation space of the intelligent agent only comprises the charge state of the DGU And the average charge state of its neighboring nodes 。
8. An aircraft energy system combination control method accounting for topology and parameter variations as recited in claim 7, wherein the bonus function Awards balanced by charge state And a calm prize The composition is as follows: ; Wherein, the ; ; To reach a reference value for the agent action when the state of charge is balanced; To adjust And Weight coefficient of (2); And To amplify the index of the state of charge imbalance.
9. The method for combined control of an aircraft energy system according to claim 8, wherein based on a charge state balancing strategy of distributed multi-agent DDPG, an Actor-Critic dual-network structure is adopted, so that each agent observes state information only according to a distributed communication link in a training process and acts to explore an environment, and distributed training is realized through continuous network parameter updating, so that collaborative balancing of charge states is achieved; Wherein, the action space of the intelligent body The following is shown: ; Each Critic network is constructed by using a three-layer fully-connected neural network, and the activation function of the hidden layer is ReLU; critic online network adopts mean square error loss function and updates its parameters by gradient back propagation The loss function is as follows: ; Wherein, the Representing a loss function; Observing for t time steps; Actions that are t time steps; representing a desired operation; Representing the Q value generated by the Critic on-line network, the target Q value of the Critic network The following is shown: ; Wherein, the Is the target Q value; Rewarding the intelligent agent at the time step t; a discount factor for rewarding; Generating a Q value for a target network by Critic; Observation for time step t+1; a deterministic map defined by the Actor target network; critic target network parameters Soft update is performed after Critic online network iteration as follows: ; Wherein, the Is a soft update factor used to adjust the magnitude of the Critic target network update.
10. The method for combined control of an aircraft energy system taking into account topology and parameter variations according to claim 9, wherein the Actor network has the same structure as the Critic network, the only difference being that the activation function of the output layer uses a Tanh function and adopts linear transformation to satisfy the constraint condition of the optimization problem; actor online network updates its parameters by maximizing Q value The following is shown: ; Wherein, the Is an optimization target; Is an Actor online network parameter; representing the Q value generated by the Critic online network; Is that Is a distribution of (3); A deterministic map defined for an online network by an Actor; the Actor online network is updated by using a gradient rising method, and the strategy gradient is as follows: ; Wherein, the To optimise the gradient of the target, i.e For a pair of Is a gradient of (2); Actor target network parameters Soft updates are performed as follows: ; at each training step Ith agent Is to obtain local observations through a communication network And generate actions 。

Description

Aircraft energy system combination control method considering topology and parameter changes Technical Field The invention relates to the technical field of control of direct-current micro-grid energy systems, in particular to an aircraft energy system combination control method considering topology and parameter changes. Background The near space aircraft is an aircraft running in an airspace with an altitude of 20-100KM, and the running orbit of the near space aircraft is between that of a traditional aircraft and that of a satellite, so that the near space aircraft has great economic and military values. Researches show that the adjacent aircraft can play an important role in various fields, such as communication relay, investigation and monitoring, mapping, weather prediction and the like, as the high altitude pseudolite. With the increasing strategic value of the near space, research into near aircraft has been rapidly conducted in recent years. The energy system of the near space vehicle supplies power for the near space vehicle, and ensures long-endurance operation and reliability of the near space vehicle. To ensure sufficient redundancy, the spacecraft energy systems typically include a plurality of distributed power generation units (distributed generation unit, DGU). This configuration enhances the fault tolerance of the system to DGU faults in harsh environments, improving the safety and reliability of the system. The key challenges of the energy system of the near space aircraft are to solve the problem of topology change caused by DGU faults, and to cope with the uncertainty of parameters of electrical elements and working condition changes which occur along with the severe geographic environment of the near space, and to realize voltage stability, current sharing and charge state balance. In the existing control technology of the direct current energy system, the control method is mostly carried out under different time scales, and the bottom layer dynamics is not considered when developing a controller of a higher layer. This simplification essentially limits the tamper resistance and global stability of the upper layer controller. Voltage stabilization and accurate current sharing are the basis for energy scheduling, and conventional droop control has a contradiction between voltage drop and low current distribution accuracy, which motivates the application of distributed secondary control. Charge state balance plays a vital role in extending the cruising ability of a near space vehicle. An unbalanced charge state distribution will cause the DGU with a lower charge level to deplete energy more quickly, eventually turning into parasitic loads, disrupting the power balance of the system. This imbalance forces the remaining DGU to operate in overload conditions, severely compromising system stability. In summary, there is a need for a combined control method for energy systems of near space vehicles that accounts for topology switching and parameter uncertainty to meet the actual engineering needs thereof. Disclosure of Invention The invention aims to provide an aircraft energy system combination control method considering topology and parameter changes, which is used for establishing a multi-control input adjacent aircraft energy system model, a primary controller and a secondary controller based on robust control, a charge state balance strategy based on distributed multi-agent reinforcement learning and the like, and realizing voltage stability, current sharing and charge state balance under the condition of considering system topology changes and parameter uncertainties. In order to achieve the above object, the present invention provides a method for controlling an aircraft energy system in combination, which takes into account topology and parameter changes, comprising the steps of: step S1, based on the actual light energy storage source system and the adjacent space environmental characteristics, establishing a compact model of the multi-bus adjacent space aircraft energy source system comprising primary control and secondary control inputs; Step S2, designing a distributed state feedback primary controller for realizing voltage tracking control of an energy system according to a robust control theory; S3, designing a distributed secondary controller, reducing the influence of the resistance of a connecting wire on the current sharing precision, simultaneously expressing the uncertainty of parameters and the output action of the reinforcement learning agent as polyhedral vertexes, and solving the controller to ensure the robustness of the system; and S4, designing a charge state balancing strategy based on multi-agent reinforcement learning, wherein each agent only observes state information of adjacent nodes and outputs a droop coefficient to adjust current, so that charge state balancing of all distributed generation units DGU is realized. Preferably, in step S1, a compact model of t