A method and system for controlling a water-air multi-rotor aircraft

Abstract

The invention discloses a water-to-air multi-rotor aircraft control method and a system thereof. The method comprises: obtaining flight instructions and state information of the water-to-air multi-rotor aircraft; obtaining the target state and initial state of the water-to-air multi-rotor aircraft in the current task; using a deep reinforcement learner to obtain the action of the water-to-air multi-rotor aircraft at the current moment and sending the action at the current moment to the water-to-air multi-rotor aircraft; controlling the water-to-air multi-rotor aircraft to execute the action at the current moment; obtaining the state of the water-to-air multi-rotor aircraft at the next moment; controlling the water-to-air multi-rotor aircraft to cache the initial state, the action at the current moment, the state at the next moment and the reward function at the current moment as a data packet and feeding the data packet back to the deep reinforcement learner; using the deep reinforcement learner to judge whether the state at the next moment is the target state, and to judge whether the actual position and attitude are the target position and target attitude, so as to realize the control of the aircraft.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft control, and in particular to a water-to-air multi-rotor aircraft control method and system thereof. Background Art

[0002] With the rapid development of science and technology and the continuous progress of society, a single type of aircraft has been difficult to meet work needs in some specific situations. In particular, for some plane crashes in the wild environment (water, air, land), the search and positioning operations of a single type of aircraft are limited. Therefore, there is an urgent need for an aircraft that can adapt to multiple environments to solve the problems of poor flexibility, weak adaptability and limited multi-functional movement mode of a single type of aircraft in the search and positioning of water, land and air.

[0003] Water-air multi-rotor aircraft combines the advantages of aerial drones and underwater robots, and can complete tasks that cannot be completed by a single aircraft. It has many advantages such as strong environmental adaptability, wide range of activities, and three-dimensional spatial perception. Water-air multi-rotor aircraft generally refers to unmanned sea-air integrated aircraft that can survive in both air and water environments (surface). It has the advantages of high speed, high maneuverability and rapid deployment capabilities of drones, the rapid cruising capabilities of unmanned surface ships, or the high concealment of unmanned underwater vehicles.

[0004] The forces on the irregular shape of water-air multi-rotor aircraft in different media are highly complex, and they will be affected by different environments during the execution of flight missions. For example, the flight state of water-air multi-rotor aircraft will be affected by wind speed in the air; it will be affected by water flow velocity underwater; especially at the water-air boundary, it will be affected by wind, water waves and propeller thrusters of water-air multi-rotor aircraft, making its motion control very difficult. The water-air multi-rotor aircraft system has the characteristics of multiple inputs and multiple outputs, under-actuation, strong coupling and nonlinearity, as well as the complexity of the dynamic model, parameter uncertainty and simplification in the modeling process leading to inaccurate modeling, which have high requirements on the robustness and anti-interference of its controller. The traditional linear control method cannot achieve real-time tracking of its dynamic change process and has insufficient anti-interference ability. Summary of the invention

[0005] The object of the present invention is to provide a water-to-air multi-rotor aircraft control method and system thereof, so as to improve the real-time motion perception, motion control accuracy and anti-disturbance capability of the water-to-air multi-rotor aircraft.

[0006] The technical solution of the present invention to solve the above technical problems is as follows:

[0007] The present invention provides a method for controlling a water-to-air multi-rotor aircraft, the method comprising:

[0008] S1: Obtain flight instructions and status information of the water-air multi-rotor aircraft;

[0009] S2: Obtaining the target state and initial state of the water-air multi-rotor aircraft in the current mission according to the flight command and state information;

[0010] S3: According to the initial state, using a deep reinforcement learner, calculating the action of the water-air multi-rotor aircraft at the current moment and sending the action at the current moment to the water-air multi-rotor aircraft;

[0011] S4: controlling the water-air multi-rotor aircraft to perform the action at the current moment;

[0012] S5: using sensors to obtain the state of the water-air multi-rotor aircraft at the next moment;

[0013] S6: Control the water-air multi-rotor aircraft to cache the initial state, the action at the current moment, the state at the next moment, and the reward function at the current moment as a data packet and feed the data packet back to the deep reinforcement learner;

[0014] S7: Using the deep reinforcement learner to determine whether the state at the next moment is the target state, if so, proceed to step S8; otherwise, update the state at the next moment to the initial state and return to step S3;

[0015] S8: Acquire the actual position and attitude of the water-air multi-rotor aircraft in the target state;

[0016] S9: Determine whether the actual position and the attitude are the target position and the target attitude. If so, end the control program of the current task. Otherwise, perform self-anti-disturbance correction on the water-air multi-rotor aircraft.

[0017] Optionally, in step S1, the flight instruction includes the linear velocity and heading angular velocity of the water-air multi-rotor aircraft;

[0018] In step S2, the state information includes the attitude angle and water entry height of the water-air multi-rotor aircraft;

[0019] The step S2 comprises:

[0020] According to the flight instruction, a target linear velocity and a target heading angular velocity of the water-air multi-rotor aircraft in a current mission are obtained;

[0021] Obtaining a target state according to the target linear velocity, the target heading angular velocity, the target attitude angle and the target water entry height;

[0022] The initial state is obtained according to the initial water entry height of the water-air multi-role vehicle.

[0023] Optionally, the action includes a propeller rotation speed and a thruster tilt angle of the water-air multi-rotor aircraft.

[0024] Optionally, in step S3, the deep reinforcement learner includes:

[0025] An input layer, an output layer, and K fully connected layers between the input layer and the output layer. The input layer is used to input the initial state, and the output layer is used to output the action at the current moment. Each fully connected layer includes m0, m1, m2, ..., m K Neuron, for the kth fully connected layer:

[0026]

[0027] in, is the i-th neuron in the k-th layer; m k is the number of neurons in the kth layer; is the weight of the i-th row and j-th column of the k-th layer weight matrix; is the bias of the kth layer; W (k) is the weight matrix of the kth layer; Y (k-1) is the output vector of the k-1th layer and the input vector of the kth layer; Y (k) is the output vector of the kth layer; net (k) Each element in represents the weighted sum of the k-th layer input vector multiplied by the weight matrix and the bias vector; f (k) is the activation function of the k-th layer of neurons.

[0028] Optionally, in step S6, the reward function is:

[0029]

[0030] Among them, U t represents the linear velocity of the water-air multi-rotor aircraft at time t, Φ t represents the attitude angle of the water-air multi-rotor aircraft at time t, Ω t represents the angular velocity of the water-air multi-rotor aircraft at time t, U target , Φ target and Ω target They represent the target linear velocity, target attitude angle and target angular velocity respectively, and d represents the water entry depth.

[0031] Optionally, in step S9, a cascade ADRC active disturbance rejection control system is used to perform active disturbance rejection correction on the water-air multi-rotor aircraft.

[0032] Optionally, the cascade ADRC self-disturbance rejection control system includes an outer loop control subsystem and an inner loop control subsystem, wherein the outer loop control subsystem is used to estimate the internal and external disturbances of the actual position in real time and eliminate the disturbances;

[0033] The inner loop control subsystem is used to eliminate disturbances to the posture and perform real-time control.

[0034] Optionally, the outer loop control subsystem and the inner loop control subsystem both include a tracking differentiator, an extended state observer and a nonlinear error feedback control.

[0035] The tracking differentiator is used to track and differentiate the input signal of the system to obtain a stable system input signal; the extended state observer is used to estimate the real-time state and the total disturbance of the system; the nonlinear error feedback control is used to compensate the control rate according to the stable system input signal, the estimated real-time state and the total disturbance of the system, and generate the final control quantity of the position and attitude;

[0036] The input of the tracking differentiator is the actual position / attitude, and the output is the first-order tracking differential signal and the second-order tracking differential signal of the actual position / attitude;

[0037] The input of the extended state observer is the expected position information / expected attitude information of the water-air multi-rotor aircraft and the product of the expected position information / expected attitude information and a preset coefficient, and the output is the expected position signal / expected attitude signal, the expected position signal change rate / expected attitude signal change rate and the expected position signal observed total disturbance / expected attitude signal observed total disturbance;

[0038] The error feedback control input is the position signal error / attitude signal error and the change rate error, and the output is the error feedback control amount.

[0039] Optionally, the tracking differentiator is:

[0040]

[0041]

[0042] Where r, P0 is the parameter to be adjusted; h is the operation step; k is the number of sampling moments; P1 is the tracking input signal of the input signal P0 at time k, P2(k) is the first-order differential signal of P1(k); sign(.) is the sign function; fhan(.) is the fastest control comprehensive function, a is the intermediate transition parameter, a0 is the boundary layer thickness, xyz is the actual position signal, h0 is another adjustable parameter independent of the time step, x1 and x2 are different signal input quantities, y1, s y , d1 and s a These are all intermediate parameters. is a first-order differential form, is the first-order differential form of the fastest control comprehensive function;

[0043] The extended state observer is:

[0044]

[0045] Where k is the number of sampling moments; z1(k) is the measured injection pressure at moment k; z2(k) is the observed injection pressure change rate at moment k; z3(k) is the total disturbance estimate of the internal and external disturbances of the injector at moment k; h() is the step size function, fal() is the saturation function, a1, a2, a3, β 01 ,β 02 and β 03 are adjustment parameters; δ is the length of the linear segment, and e is the gain;

[0046] The error feedback control amount of the error feedback control is processed as follows:

[0047]

[0048] Where u0 is the error feedback control quantity; β1 and β2 are both adjustable parameters; e1 is the position signal error / attitude signal error, e2 is the rate of change error, P1 and P2 are the first-order tracking differential signal and the second-order tracking differential signal of the actual position, a1 and a2 are the adjustment parameters, and δ is the area length of the linear segment.

[0049] The present invention also provides a control system based on the above-mentioned water-to-air multi-rotor aircraft control method, the control system comprising:

[0050] A data acquisition module, the data acquisition module is used to obtain flight instructions and status information of the water-air multi-rotor aircraft; use sensors to obtain the state of the water-air multi-rotor aircraft at the next moment and obtain the actual position and posture of the water-air multi-rotor aircraft in the target state;

[0051] A data processing module, the data processing module is used to obtain the target state and initial state of the water-air multi-rotor aircraft in the current task according to the flight instruction and state information; according to the initial state, using a deep reinforcement learner, calculate the action of the water-air multi-rotor aircraft at the current moment;

[0052] A data transmission module, the data transmission module is used to send the action at the current moment to the water-air multi-rotor aircraft;

[0053] An aircraft control module, the aircraft control module is used to control the water-air multi-rotor aircraft to perform the action at the current moment; control the water-air multi-rotor aircraft to cache the initial state, the action at the current moment, the state at the next moment, and the reward function at the current moment as a data packet and feed it back to the deep reinforcement learner together with the data packet;

[0054] A judgment module is used to use the deep reinforcement learner to judge whether the state at the next moment is the target state and to judge whether the actual position and the posture are the target position and the target posture.

[0055] The present invention has the following beneficial effects:

[0056] (1) The present invention can determine the current state of the water-air multi-rotor aircraft in real time, and recalibrate the signal according to the target state and the current state, that is, calibrate the linear speed, attitude angle and angular velocity of the water-air multi-rotor aircraft, and combine with the cascaded anti-disturbance control system to enable the water-air multi-rotor aircraft to reach the target state in the fastest and most stable way;

[0057] (2) The cascaded ADRC anti-disturbance control system provided by the present invention consists of three parts: a tracking differentiator, an extended state observer and a nonlinear error feedback control. The three parts cooperate with each other and work together to overcome the disturbance of the flight process to a certain extent and realize the control of the position and attitude of the water-air multi-rotor aircraft;

[0058] (3) During the process of water-to-air multi-rotor aircraft working in water, air, and water-to-air transition, the real-time motion perception, motion control accuracy, and anti-interference ability of the water-to-air multi-rotor aircraft are improved, and the motion state of the water-to-air multi-rotor aircraft is efficiently controlled to achieve the goals of strong adaptability to the working environment and a large range of activities of the water-to-air multi-rotor aircraft. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of the water-air multi-rotor aircraft control method of the present invention;

[0060] Figure 2 The figure is a schematic diagram of the structure of the six-rotor water-air vehicle of the present invention (1-vehicle carrier; 2-control console; 3-thruster; 4-rotor);

[0061] Figure 3 This is a schematic diagram of the deep neural network structure of the present invention;

[0062] Figure 4 It is a schematic diagram of the structure of the cascade active disturbance rejection control system of the present invention. DETAILED DESCRIPTION

[0063] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0064] The present invention provides a method for controlling a water-air multi-rotor aircraft, referring to Figure 1 As shown, the water-to-air multi-rotor aircraft control method includes:

[0065] S1: Obtain flight instructions and status information of the water-air multi-rotor aircraft;

[0066] The takeoff control method of the water-air multi-rotor aircraft adopts manual terminal control, that is, remote control. The experimenter issues the linear velocity instructions of the water-air multi-rotor aircraft in the x, y, and z directions and the heading angular velocity instructions in the r direction through the remote control.

[0067] S2: Obtaining the target state and initial state of the water-air multi-rotor aircraft in the current mission according to the flight command and state information;

[0068] Specifically include:

[0069] According to the flight instruction, a target linear velocity and a target heading angular velocity of the water-air multi-rotor aircraft in a current mission are obtained;

[0070] Obtaining a target state according to the target linear velocity, the target heading angular velocity, the target attitude angle and the target water entry height;

[0071] The initial state is obtained according to the initial water entry height of the water-air multi-role vehicle.

[0072] That is, by analyzing the target command, the target state S of the water-air multi-rotor aircraft in this mission is obtained. t The state information includes the attitude angle and water entry height of the water-to-air multi-rotor aircraft; the initial state S0 is that the water-to-air multi-rotor aircraft is at a certain point in space and is in a balanced state S0 = [0, 0, 0, 0, 0, 0, 0, d0], d0 is the water entry depth, measured by the depth sensor.

[0073] The defined states are:

[0074] s t =[U t ,Φ t ,Ω t ,d]

[0075] Among them, U t is the linear velocity, Φ t is the attitude angle, Ω t is the angular velocity, d is the height. Attitude angle information Only the roll angle and pitch angle of the water-air multi-rotor aircraft are considered, and the heading angle of the water-air multi-rotor aircraft is not considered.

[0076] Furthermore, let Ω={p,q,r} T Indicates the angular velocity of the water-air multi-rotor aircraft in the rolling, pitching and yaw directions; Represents the attitude angle of the water-air multi-rotor aircraft; U = {u, v, w} T P is the linear velocity of the water-air multi-rotor aircraft in all directions; W ={x,y,z} T It is the position of the water-air multi-rotor aircraft.

[0077] S3: According to the initial state, using a deep reinforcement learner, calculating the action of the water-air multi-rotor aircraft at the current moment and sending the action at the current moment to the water-air multi-rotor aircraft;

[0078] The action includes the propeller speed and propeller tilt angle of the water-air multi-rotor aircraft, such as Figure 2 As shown in the figure, the propeller speeds [ω1, ω2, ω3, ω4, ω5, ω6] of the six rotor units of the water-air multi-rotor aircraft and the tilt angles [δ1, δ2] of the two ducted propeller tilt servos are selected as actions. Since it is difficult for the propeller speed and tilt angle of the tilt servos of the water-air multi-rotor aircraft to implement large changes in action in a short period of time, the action change threshold Δω is defined th and Δδ th , then the next action should be ω t+1 ∈[ω-Δω th ,ω+Δω th ] and δ t+1 ∈[δ-Δδ th ,δ+Δδ th ].

[0079] Here, reference Figure 3 As shown, the deep reinforcement learner includes:

[0080] An input layer, an output layer, and K fully connected layers between the input layer and the output layer. The input layer is used to input the initial state, and the output layer is used to output the action at the current moment. Each fully connected layer includes m0, m1, m2, ..., m K Neuron, for the kth fully connected layer:

[0081]

[0082] in, is the i-th neuron in the k-th layer; m k is the number of neurons in the kth layer; is the weight of the i-th row and j-th column of the k-th layer weight matrix; is the bias of the kth layer; W (k) is the weight matrix of the kth layer; Y (k-1) is the output vector of the k-1th layer and the input vector of the kth layer; Y (k) is the output vector of the kth layer; net (k) Each element in represents the weighted sum of the k-th layer input vector multiplied by the weight matrix and the bias vector; f (k) is the activation function of the k-th layer of neurons.

[0083] S4: controlling the water-air multi-rotor aircraft to perform the action at the current moment;

[0084] S5: using sensors to obtain the state of the water-air multi-rotor aircraft at the next moment;

[0085] S6: Control the water-air multi-rotor aircraft to cache the initial state, the action at the current moment, the state at the next moment, and the reward function at the current moment as a data packet and feed the data packet back to the deep reinforcement learner;

[0086] The reward function is:

[0087]

[0088] Among them, U t represents the linear velocity of the water-air multi-rotor aircraft at time t, Φ t represents the attitude angle of the water-air multi-rotor aircraft at time t, Ω t represents the angular velocity of the water-air multi-rotor aircraft at time t, U target , Φ target and Ω target They represent the target linear velocity, target attitude angle and target angular velocity respectively, and d represents the water entry depth.

[0089] S7: Using the deep reinforcement learner to determine whether the state at the next moment is the target state, if so, proceed to step S8; otherwise, update the state at the next moment to the initial state and return to step S3;

[0090] S8: Acquire the actual position and attitude of the water-air multi-rotor aircraft in the target state;

[0091] S9: Determine whether the actual position and the attitude are the target position and the target attitude. If so, end the control program of the current task. Otherwise, perform self-anti-disturbance correction on the water-air multi-rotor aircraft.

[0092] The present invention utilizes a cascaded ADRC self-disturbance rejection control system to perform self-disturbance rejection correction on the water-air multi-rotor aircraft.

[0093] Specifically, the cascade ADRC self-disturbance rejection control system includes an outer loop control subsystem and an inner loop control subsystem, wherein the outer loop control subsystem is used to estimate the internal and external disturbances of the actual position in real time and eliminate the disturbances;

[0094] The inner loop control subsystem is used to eliminate disturbances to the posture and perform real-time control.

[0095] The outer ring of the present invention refers to the position of the aircraft, and the inner ring refers to the status of the aircraft.

[0096] Optionally, refer to Figure 4 As shown, the outer loop control subsystem and the inner loop control subsystem both include a tracking differentiator, an extended state observer and a nonlinear error feedback control.

[0097] The tracking differentiator is used to track and differentiate the input signal of the system to obtain a stable system input signal; the extended state observer is used to estimate the real-time state and the total disturbance of the system; the nonlinear error feedback control is used to compensate the control rate according to the stable system input signal, the estimated real-time state and the total disturbance of the system, and generate the final control quantity of position and attitude.

[0098] The input of the tracking differentiator is the actual position / attitude, and the output is the first-order tracking differential signal and the second-order tracking differential signal of the actual position / attitude;

[0099] The input of the extended state observer is the expected position information / expected attitude information of the water-air multi-rotor aircraft and the product of the expected position information / expected attitude information and a preset coefficient, and the output is the expected position signal / expected attitude signal, the expected position signal change rate / expected attitude signal change rate and the expected position signal observed total disturbance / expected attitude signal observed total disturbance;

[0100] Taking the outer loop control subsystem as an example, the final control amount of the expected position signal U1 is:

[0101] U1(k)=u0(k)-z3(k) / b0

[0102] Where k is the number of sampling moments; U1(k) is the final control output of the ADRC subsystem at moment k; z3(k) is the total disturbance observed at moment k; b0 is the compensation factor that determines the strength of compensation, and u0 is the error feedback control amount at moment k.

[0103] The error feedback control input is the position signal error / attitude signal error and the change rate error, and the output is the error feedback control amount.

[0104] The tracking differentiator is:

[0105]

[0106]

[0107] Where r, P0 is the parameter to be adjusted; h is the operation step; k is the number of sampling moments; P1 is the tracking input signal of the input signal P0 at time k, P2(k) is the first-order differential signal of P1(k); sign(.) is the sign function; fhan(.) is the fastest control comprehensive function, a is the intermediate transition parameter, a0 is the boundary layer thickness, xyz is the actual position signal, h0 is another adjustable parameter independent of the time step, x1 and x2 are different signal input quantities, y1, s y , d1 and s a These are all intermediate parameters. is a first-order differential form, is the first-order differential form of the fastest control comprehensive function;

[0108] The extended state observer is:

[0109]

[0110] Where k is the number of sampling moments; z1(k) is the measured injection pressure at moment k; z2(k) is the observed injection pressure change rate at moment k; z3(k) is the total disturbance estimate of the internal and external disturbances of the injector at moment k; h() is the step size function, fal() is the saturation function, a1, a2, a3, β 01 ,β 02 and β 03 are adjustment parameters; δ is the length of the linear segment, and e is the gain;

[0111] The error feedback control amount of the error feedback control is processed as follows:

[0112]

[0113] Where u0 is the error feedback control quantity; β1 and β2 are both adjustable parameters; e1 is the position signal error / attitude signal error, e2 is the rate of change error, P1 and P2 are the first-order tracking differential signal and the second-order tracking differential signal of the actual position, a1 and a2 are the adjustment parameters, and δ is the area length of the linear segment.

[0114] The present invention also provides a control system based on the above-mentioned water-to-air multi-rotor aircraft control method, the control system comprising:

[0115] A data acquisition module, the data acquisition module is used to obtain flight instructions and status information of the water-air multi-rotor aircraft; use sensors to obtain the state of the water-air multi-rotor aircraft at the next moment and obtain the actual position and posture of the water-air multi-rotor aircraft in the target state;

[0116] A data processing module, the data processing module is used to obtain the target state and initial state of the water-air multi-rotor aircraft in the current task according to the flight instruction and state information; according to the initial state, using a deep reinforcement learner, calculate the action of the water-air multi-rotor aircraft at the current moment;

[0117] A data transmission module, the data transmission module is used to send the action at the current moment to the water-air multi-rotor aircraft;

[0118] An aircraft control module, the aircraft control module is used to control the water-air multi-rotor aircraft to perform the action at the current moment; control the water-air multi-rotor aircraft to cache the initial state, the action at the current moment, the state at the next moment, and the reward function at the current moment as a data packet and feed it back to the deep reinforcement learner together with the data packet;

[0119] A judgment module is used to use the deep reinforcement learner to judge whether the state at the next moment is the target state and to judge whether the actual position and the posture are the target position and the target posture.

[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for controlling a water-air multi-rotor aircraft, characterized in that: The water-air multi-rotor aircraft control method comprises: S1: Acquire flight instructions and status information of the water-air multi-rotor aircraft, wherein the flight instructions include the linear velocity and heading angular velocity of the water-air multi-rotor aircraft; S2: According to the flight instruction and the state information, the target state and the initial state of the water-air multi-rotor aircraft in the current mission are obtained, wherein the state information includes the attitude angle and the water entry height of the water-air multi-rotor aircraft, specifically including the following sub-steps: According to the flight instruction, a target linear velocity and a target heading angular velocity of the water-air multi-rotor aircraft in a current mission are obtained; Obtaining a target state according to the target linear velocity, the target heading angular velocity, the target attitude angle and the target water entry height; According to the initial water entry height of the water-air multi-role vehicle, the initial state is obtained; S3: According to the initial state, using a deep reinforcement learner, calculating the action of the water-air multi-rotor aircraft at the current moment and sending the action at the current moment to the water-air multi-rotor aircraft; S4: controlling the water-air multi-rotor aircraft to perform the action at the current moment; S5: using sensors to obtain the state of the water-air multi-rotor aircraft at the next moment; S6: Control the water-air multi-rotor aircraft to cache the initial state, the action at the current moment, the state at the next moment, and the reward function at the current moment as a data packet and feed the data packet back to the deep reinforcement learner; S7: Using the deep reinforcement learner to determine whether the state at the next moment is the target state, if so, proceed to step S8; otherwise, update the state at the next moment to the initial state and return to step S3; S8: Acquire the actual position and attitude of the water-air multi-rotor aircraft in the target state; S9: Determine whether the actual position and the attitude are the target position and the target attitude, if so, end the control program of the current task, otherwise, perform anti-disturbance correction on the water-air multi-rotor aircraft; In the step S9, the water-air multi-rotor aircraft is subjected to an auto-disturbance rejection correction by using a cascaded ADRC auto-disturbance rejection control system; The cascade ADRC self-disturbance rejection control system includes an outer loop control subsystem and an inner loop control subsystem, wherein the outer loop control subsystem is used to estimate the internal and external disturbances of the actual position in real time and eliminate the disturbances; The inner loop control subsystem is used to eliminate disturbances to the posture and perform real-time control; The outer loop control subsystem and the inner loop control subsystem both include a tracking differentiator, an extended state observer and a nonlinear error feedback control. The input of the tracking differentiator is the actual position / attitude, and the output is the first-order tracking differential signal and the second-order tracking differential signal of the actual position / attitude; The input of the extended state observer is the expected position information / expected attitude information of the water-air multi-rotor aircraft and the product of the expected position information / expected attitude information and a preset coefficient, and the output is the expected position signal / expected attitude signal, the expected position signal change rate / expected attitude signal change rate and the expected position signal observed total disturbance / expected attitude signal observed total disturbance; The error feedback control input is the position signal error / attitude signal error, the rate of change error, and the output is the error feedback control amount; The tracking differentiator is: in, Parameters to be adjusted; is the operation step length; is the number of sampling moments; for Input signal at the moment The tracking input signal, for The first-order differential signal of is a symbolic function; is the fastest control synthesis function, a is the intermediate transition parameter, a 0 is the boundary layer thickness, xyz is the actual position signal, h 0 is another adjustable parameter independent of the time step, x 1 and x 2 is the different signal input quantities, , s y , d 1 and s a All are intermediate parameters. is a first-order differential form, is the first-order differential form of the fastest control comprehensive function; The extended state observer is: In the formula, is the number of sampling moments; for Measure the injection pressure at the time; for The observed injection pressure change rate at time ; for The total disturbance estimate of the internal and external disturbances of the injector at time t; is the step size function, is the saturation function, and All are adjustment parameters; is the length of the linear segment region, e is the error; the error feedback control amount of the error feedback control is processed as follows: In the formula, is the error feedback control quantity; and All are adjustable parameters; is the position signal error / attitude signal error, is the rate of change error, and are the first-order tracking differential signal and the second-order tracking differential signal of the actual position respectively, and are the adjustment parameters, is the length of the linear region.

2. The method for controlling a water-air multi-rotor aircraft according to claim 1, characterized in that: In step S3, the action includes the propeller speed and propeller tilt angle of the water-air multi-rotor aircraft.

3. The method for controlling a water-to-air multi-rotor aircraft according to any one of claims 1 to 2, characterized in that: In step S3, the deep reinforcement learner includes: An input layer, an output layer, and K fully connected layers between the input layer and the output layer, wherein the input layer is used to input the initial state, and the output layer is used to output the action at the current moment. Each fully connected layer includes a Neuron, for the kth fully connected layer: in, For the Layer neurons; For the The number of neurons in the layer; For the The layer weight matrix OK Column weights; For the Bias of the layer; For the The weight matrix of the layer; For the The output vector of the layer or The input vector of the layer; For the The output vector of the layer; For the The bias matrix of the layer, Each element in represents the The weighted sum of the layer input vector multiplied by the weight matrix and the bias vector; For the The activation function of the neurons in the layer.

4. The method for controlling a water-to-air multi-rotor aircraft according to claim 1, characterized in that: In step S6, the reward function for: in, express t The linear speed of the water-air multi-rotor aircraft at time, express t The attitude angle of the water-air multi-rotor aircraft at any moment, express t The angular velocity of the water-air multi-rotor aircraft at time, , and They represent the target linear velocity, target attitude angle and target angular velocity respectively. d Indicates the water depth.

5. A control system based on the water-air multi-rotor aircraft control method according to any one of claims 1 to 4, characterized in that: The control system comprises: A data acquisition module, the data acquisition module is used to obtain flight instructions and status information of the water-air multi-rotor aircraft; use sensors to obtain the state of the water-air multi-rotor aircraft at the next moment and obtain the actual position and posture of the water-air multi-rotor aircraft in the target state; A data processing module, the data processing module is used to obtain the target state and initial state of the water-air multi-rotor aircraft in the current task according to the flight instruction and state information; according to the initial state, using a deep reinforcement learner, calculate the action of the water-air multi-rotor aircraft at the current moment; A data transmission module, the data transmission module is used to send the action at the current moment to the water-air multi-rotor aircraft; An aircraft control module, the aircraft control module is used to control the water-air multi-rotor aircraft to perform the action at the current moment; control the water-air multi-rotor aircraft to cache the initial state, the action at the current moment, the state at the next moment, and the reward function at the current moment as a data packet and feed it back to the deep reinforcement learner together with the data packet; A judgment module is used to use the deep reinforcement learner to judge whether the state at the next moment is the target state and to judge whether the actual position and the posture are the target position and the target posture.

Images