An intelligent load reduction and drag reduction control method and a cross-medium vehicle using the same

Abstract

The present invention relates to the technical field of cross-media vehicles, and specifically discloses an intelligent load reduction and drag reduction control method and a cross-media vehicle using the method. During the specific implementation of the intelligent load reduction and drag reduction control method, when the vehicle enters and exits water and navigates at high speed in the water, the supercavitation intelligent controller primarily senses the flight medium in which the supercavitation cross-media vehicle is located, as well as the relationship between the cavitation and the cross-media vehicle structure. When the supercavitation intelligent controller senses that the cavitation and the vehicle are no longer matched, it actively controls the optimal strategy relationship between the cavitation and the cross-media vehicle under deep reinforcement learning, adjusts the ventilation volume, adjusts the vehicle attitude, and adjusts the relationship between the wing and the cross-media vehicle to achieve optimal matching between the cavitation and the vehicle, thereby achieving optimal load reduction and drag reduction. To achieve load reduction and drag reduction, all matrix operations involved in the controller algorithm use efficient quantum computing algorithms to improve calculation speed.

Description

Technical Field

[0001] The present invention relates to the technical field of cross-media vehicles, and in particular to an intelligent load reduction and drag reduction control method and a cross-media vehicle using the method. Background Art

[0002] Conventional trans-medium vehicles need to balance flight and underwater navigation, typically featuring fixed wings and two propulsion systems, one for air and one for underwater. This results in a large cross-sectional area, resulting in very low speeds for both entry and exit, and for underwater navigation. To achieve high-speed, stable entry and exit, and for underwater navigation, jet-flow venting cavities and solid-structure venting cavities are employed to assist in both entry and exit, achieving high-speed navigation. Given the structural characteristics of trans-medium vehicles, the typically formed venting cavities are difficult to completely enclose conventional trans-medium vehicles. Therefore, the matching of cavitation morphology with the trans-medium vehicle structure is crucial to truly achieve load and drag reduction for high-speed trans-medium entry and exit, and underwater navigation. The stability of the supercavitation morphology is dependent on the ventilation volume, speed, ambient pressure, and cavitation number. These parameters vary depending on the navigation state, making it difficult to maintain a constant cavitation morphology. Cavitation that is too large or too small is detrimental to the navigation of the vehicle, making the control of the supercavitation morphology crucial. Supercavitation morphology is typically adjusted by regulating ventilation. Controlling supercavitation requires first measuring and sensing the distance between the cavitation and the surface of the vehicle. Existing non-contact ranging technologies, such as laser and ultrasonic ranging, require receiving reflected signals. However, water generally has good permeability to light and sound waves, and the cavitation surface is uneven, which severely impacts measurement accuracy and increases response time, leading to delayed adjustments to ventilation and, consequently, the inability to adjust cavitation morphology. The bionic lateral line is a technology that mimics the way fish perceive underwater flow fields through bionic principles. It primarily uses piezoresistive, capacitive, or piezoelectric strain gauges as sensing elements. The connected probe is affected by the water flow, causing the sensing element to deform, generating electrical changes. The intensity of these changes is then used to measure and sense the flow velocity and eddies. Current bionic lateral lines are unable to sense cavitation morphology.

[0003] Currently, it is difficult to accurately and real-timely perceive, intelligently and efficiently control the morphology of cavitation bubbles during cross-medium entry and exit, and to match the cavitation with the vehicle structure. Consequently, it is impossible to achieve efficient, repetitive, high-speed, and stable entry and exit of cross-medium vehicles, as well as stable high-speed navigation in water. Therefore, a highly efficient and real-time intelligent control method for supercavitation morphology is proposed to address these issues. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent load reduction and drag reduction control method to achieve the needs of rapid matching of cavitation and aircraft to reduce load and drag.

[0005] In order to solve the above problems, the present invention adopts the following technical solutions:

[0006] In a first aspect, an embodiment of the present disclosure provides an intelligent load reduction and drag reduction control method, comprising the following steps:

[0007] A biomimetic side-line speed sensor array was used to collect voltage data in calibration tests, and a gradient descent deep learning method was used to train and evaluate the speed-voltage model.

[0008] A biomimetic whisker resistance sensor array was used to collect current data in calibration experiments, and a gradient descent deep learning method was used to train and evaluate the resistance-current model.

[0009] A supercavitation perception model is constructed based on the underwater resistance formula, velocity-voltage model and resistance-current model;

[0010] The cavitation morphology, ventilation volume, vehicle resistance, and acceleration data of the cross-medium vehicle were obtained during water entry and exit tests and underwater navigation tests. The supercavitation perception model was trained using reinforcement learning methods to obtain a supercavitation intelligent control model.

[0011] The real-time voltage and current data of the cross-medium vehicle are input into the supercavitation intelligent control model, and the cavitation morphology is sensed based on the real-time speed and resistance.

[0012] The cavitator ventilation volume and the attitude of the cross-medium vehicle are adjusted in real time according to the sensed cavitation morphology, so as to adjust the cavitation morphology and the attitude of the cross-medium vehicle to the optimal state.

[0013] Preferably, the relationship formula of the supercavitation perception model is:

[0014] Water resistance:

[0015] Wetting section length: L1 = L-L2;

[0016] From the above formula, we can get the distance between the supercavitation bubble and the boundary of the vehicle (length of the dry section):

[0017] in:

[0018] ρ D : density of water;

[0019] U: speed;

[0020] C D : drag coefficient of carbon fiber whisker rod;

[0021] The bionic tentacle drag sensor consists of a carbon fiber tentacle rod with a length of L and a diameter of H. It is located at the head of the trans-medium vehicle. The length L1 of the carbon fiber tentacle rod is outside the cavitation bubble and is the wetted section. The dry section inside the cavitation bubble is L2. L2 is the distance between the supercavitation boundary and the trans-medium vehicle surface.

[0022] Preferably, the real-time adjustment is based on active control through reinforcement learning with cavitation morphology characteristics as the environment and flow control as the action. By setting a reward function, the optimal cavitation morphology strategy is achieved during the entire process of the cross-medium vehicle entering and exiting water and sailing at high speed in water.

[0023] Preferably, the specific method of the reinforcement learning active control is:

[0024] 1) Give a penalty P to the action or decision system of the cross-medium vehicle t or reward R t measure;

[0025] 2) The ultimate goal of the cross-medium vehicle is to obtain the maximum cumulative reward G t , the cross-medium vehicle takes action a based on the environment at time t t , in action a t Complete the status S t Transition to new state S t+1 , and feedback reward R t+1 ;

[0026] 3) Finally, after continuous iteration and interaction with the environment, cumulative rewards are obtained

[0027] Among them, γ is the discount factor, and the larger γ is, the greater the emphasis on future rewards.

[0028] In a second aspect, an embodiment of the present disclosure provides a cross-medium vehicle that applies the intelligent load reduction and drag reduction control method, comprising: a fuselage, a front wing, a rear wing, a foldable tilt-rotor, a cross-medium propeller, a cavitator, and a supercavitation intelligent controller.

[0029] The body has a head support rod.

[0030] The front wing is foldably arranged on the fuselage.

[0031] The rear wing is foldably arranged on the fuselage.

[0032] The head support rod is provided with a bionic side linear speed sensor array, and the head of the fuselage is provided with a bionic tentacle resistance sensor array.

[0033] The foldable tilt rotor, the trans-medium thruster, the cavitator, the bionic side linear velocity sensor array and the bionic tentacle resistance sensor array are all electrically connected to the supercavitation intelligent controller.

[0034] Preferably, the supercavitation intelligent controller has an efficient quantum computing algorithm module.

[0035] Preferably, the supercavitation intelligent controller has a deep reinforcement learning module.

[0036] The beneficial effects of the present invention are: aiming at the cavitation morphology perception and active control of the ventilated supercavitation cross-medium vehicle, realizing the demand of rapid matching of cavitation and vehicle for load and drag reduction, solving the deficiencies in the prior art, making the ventilated supercavitation cross-medium vehicle body not hinder the supercavitation generation, and being able to autonomously and effectively perceive and efficiently control the cavitation morphology during entry and exit of water and high-speed underwater navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 Flowchart of the intelligent load reduction and drag reduction control method in the embodiment.

[0039] Figure 2 Schematic diagram of the structure of the cross-media vehicle in the embodiment.

[0040] Figure 3 Schematic diagram of the local structure of the cross-media vehicle in the embodiment.

[0041] Figure 4 Schematic diagram of the bionic whisker resistance sensor array structure in the embodiment.

[0042] Figure 5 Schematic diagram of the cross-medium vehicle in the folded state.

[0043] Figure 6 This is the state diagram of the cross-medium vehicle entering the water.

[0044] Figure 7 This is a state diagram of the underwater navigation process of a cross-media vehicle.

[0045] Figure 8 Diagram of the reinforcement learning process in the experimental environment.

[0046] In the picture:

[0047] 10. Body; 11. Head support rod; 12. Bionic side linear velocity sensor array; 13. Bionic tentacle resistance sensor array; 131. Carbon fiber tentacle rod; 132. Flexible base; 133. Triboelectric nanogenerator unit;

[0048] 20. Front wing;

[0049] 30. rear wing;

[0050] 40. Foldable tilt rotor;

[0051] 50. Trans-medium thruster;

[0052] 60. Cavitation device. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments will be described clearly and completely below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments, rather than all the embodiments.

[0054] During the implementation of this intelligent load and drag reduction control method, when a vehicle enters and exits water or travels at high speed in the water, the supercavitating intelligent controller primarily senses the flight medium in which the supercavitating cross-medium vehicle resides, as well as the relationship between the cavitation and the cross-medium vehicle structure. When the supercavitating intelligent controller senses that the cavitation and the vehicle are no longer compatible, it actively controls the optimal strategy for matching the cavitation and the cross-medium vehicle using deep reinforcement learning. This strategy adjusts the ventilation volume, the vehicle's attitude, and the relationship between the wing and the cross-medium vehicle to achieve optimal matching, thereby reducing load and drag. Because cross-medium vehicles travel at extremely high speeds during water entry and exit and underwater navigation, traditional algorithms cannot guarantee rapid decision-making. Therefore, to achieve load and drag reduction, all matrix operations in the controller algorithm utilize efficient quantum computing algorithms to increase computational speed.

[0055] like Figure 1 As shown, this embodiment provides an intelligent load reduction and drag reduction control method, including the following steps:

[0056] 1) A biomimetic side-line speed sensor array was used to collect voltage data in a calibration test, and a gradient descent deep learning method was used to train and evaluate the speed-voltage model;

[0057] A biomimetic whisker resistance sensor array was used to collect current data in calibration experiments, and a gradient descent deep learning method was used to train and evaluate the resistance-current model.

[0058] 2) Construct a supercavitation perception model based on the underwater resistance formula, velocity-voltage model, and resistance-current model;

[0059] 3) Obtain cavitation morphology, ventilation volume, vehicle resistance, and acceleration data of the cross-medium vehicle during water entry and exit tests and underwater navigation tests. Use reinforcement learning to train a supercavitation perception model and obtain a supercavitation intelligent control model.

[0060] 4) Input the real-time voltage and current data of the cross-medium vehicle into the supercavitation intelligent control model to sense the cavitation morphology based on the real-time speed and resistance;

[0061] 5) According to the sensed cavitation morphology, the cavitator ventilation volume and the attitude of the cross-medium vehicle are adjusted in real time to adjust the cavitation morphology and the attitude of the cross-medium vehicle to the optimal state.

[0062] Specifically, the relationship of the supercavitation perception model is:

[0063] Water resistance:

[0064] Wetting section length: L1 = L-L2;

[0065] From the above formula, we can get the distance between the supercavitation bubble and the boundary of the vehicle (length of the dry section):

[0066] in:

[0067] ρ D : density of water;

[0068] U: speed;

[0069] C D : drag coefficient of carbon fiber whisker rod;

[0070] The bionic tentacle drag sensor consists of a carbon fiber tentacle rod with a length of L and a diameter of H. It is located at the head of the trans-medium vehicle. The length L1 of the carbon fiber tentacle rod is outside the cavitation bubble and is the wetted section. The dry section inside the cavitation bubble is L2. L2 is the distance between the supercavitation boundary and the trans-medium vehicle surface.

[0071] The following will describe a cross-medium vehicle using the intelligent load reduction and drag reduction control method with reference to the accompanying drawings.

[0072] like Figure 2 As shown, this embodiment provides a trans-medium vehicle whose fuselage does not hinder supercavitation generation and can autonomously and effectively sense and efficiently control cavitation morphology during entry and exit of water and high-speed underwater navigation. The supercavitating trans-medium vehicle includes a fuselage 10, a front wing 20, a rear wing 30, a foldable tilt-rotor 40, a trans-medium propeller 50, a cavitator 60, and a supercavitation intelligent controller (not shown).

[0073] like Figure 3As shown, the fuselage 10 has a head support mast 11. Both the front wing 20 and the rear wing 30 are foldably mounted on the fuselage 10. A bionic side linear velocity sensor array 12 is mounted on the head support mast 11, and a bionic whisker drag sensor array 13 is mounted on the head of the fuselage 10. The foldable tilt-rotor 40, trans-medium thruster 50, cavitator 60, bionic side linear velocity sensor array 12, and bionic whisker drag sensor array 13 are all electrically connected to the supercavitation intelligent controller. Three foldable tilt-rotors 40 are provided.

[0074] Exemplarily, both the foldable tilt rotor 40 and the trans-medium thruster 50 are tilted by a steering gear.

[0075] like Figure 2 As shown, the supercavitation cross-medium aircraft is in vertical take-off and landing and multi-rotor flight state. At this time, the three foldable tilt rotors 40 simultaneously provide upward pulling force.

[0076] Exemplarily, the front two foldable tilt-rotors 40 are tilted forward, and the tail foldable tilt-rotor 40 is tilted backward, while providing forward pulling force, and the front wing 20 and the rear wing 30 generate upward lift, so that the supercavitation cross-medium aircraft can achieve high-speed fixed-wing flight in the air.

[0077] like Figure 5 The figure shows the cross-medium vehicle entering and exiting the water, and navigating underwater. The front and rear wings 20 and 30 are folded, and the foldable tilt-rotor 40 automatically folds. In this state, the cross-sectional area is consistent with the main body diameter. This effectively achieves a coordinated alignment between the axisymmetric cavitation and the vehicle structure, enabling high-speed entry and exit, as well as underwater navigation.

[0078] like Figure 3 As shown, the bionic side linear speed sensor array 12 is composed of a plurality of bionic side linear speed sensors arranged evenly and mounted on the aircraft head support rod 11. The bionic side linear speed sensor body is a cylinder with a length L3 and a diameter H1 made of an intelligent flexible material ion exchange polymer metal (IPMC), which is fixed to the aircraft head support rod 11 by a clamp (not shown). When the IPMC material bends under the action of an external force, the internal ion migration causes a potential difference between the upper and lower electrode surfaces, generating a voltage. The bionic side linear speed sensor is located at the front end of the aircraft, outside the supercavitation, and is therefore always in a fully wetted state. The water resistance causes the ion exchange polymer metal cylinder to bend and deform, thereby generating current. The faster the speed, the greater the water resistance, the greater the deformation, and the higher the voltage. According to this law, the relationship between speed and voltage is fitted based on the gradient descent deep learning method to perceive the real-time speed.

[0079] like Figure 4As shown, the bionic whisker resistance sensor array 13 consists of multiple bionic whisker resistance sensors evenly arranged and mounted on the head of the aircraft. The bionic whisker resistance sensor includes a carbon fiber whisker rod 131, a flexible base 132, and a triboelectric nanogenerator unit 133. The carbon fiber whisker rod 131 is L in length and H in diameter and is fixed to the flexible base 132. The flexible base 132 has a neutral restoring force and is fixed to the head of the fuselage 10. The triboelectric nanogenerator unit 133 is arranged around the flexible base 132. The triboelectric nanogenerator unit 133 is composed of a cast polypropylene film that can shield static electricity, a highly electronegative fluorinated ethylene propylene film, a highly positive conductive ink, and a highly watertight silicone rubber. When the triboelectric nanogenerator unit 133 deforms, the internal dielectric layer materials, due to their different electronegativity, generate a displacement current upon contact and separation. The contact of the dielectric materials within the bent sensor unit causes charge transfer, generating an electrical signal. The bionic tentacle drag sensor is located at the head of the fuselage 10. The length L1 of the carbon fiber tentacle stem is outside the cavitation bubble, representing the wetted section. The dry section L2 within the cavitation bubble represents the distance between the supercavitation boundary and the vehicle surface, a key parameter of supercavitation morphology. The water resistance in the wetted section drives the fiber tentacle stem, which in turn causes the triboelectric nanogenerator unit 133 to bend and deform, generating an electric current. The faster the speed, the greater the water resistance, the greater the deformation, and the higher the current. Based on this principle, a gradient descent deep learning method is used to fit the relationship between drag and current, thereby sensing real-time drag.

[0080] like Figure 3 As shown, the cavitator 60 is located on the head support rod 11 and behind the bionic side linear velocity sensor array 12 .

[0081] Figure 6 This is a state diagram of the trans-media vehicle entering water. As the trans-media vehicle enters water at high speed from air, the bionic side linear velocity sensor array 12 first contacts the water surface. Because water is over 800 times denser than air, the bionic side linear velocity sensor generates a sudden surge in voltage, which serves as the water entry ventilation signal for the cavitator 60. The cavitator 60, located behind the bionic side linear velocity sensor array 12, begins to eject air at maximum flow, forming supercavitation bubbles that envelop the surface of the submerged vehicle, reducing its load.

[0082] Figure 7 This diagram shows the state of a cross-medium vehicle during underwater navigation. When the vehicle is fully submerged, the bionic sideline velocity sensor array 12 is located outside the cavitation bubble, indicating a fully wetted state, and is used to sense the vehicle's speed. The bionic tentacle drag sensor array 13 is partially located outside the cavitation bubble, sensing water resistance. The vehicle's internal supercavitation sensor detects cavitation morphology based on real-time speed and drag. The cavitator 60 autonomously adjusts ventilation based on this cavitation morphology, adjusting the cavitation bubble to its optimal shape in real time.

[0083] To further illustrate the principles of the present disclosure, the optimal strategy for actively controlling the matching of cavitation and cross-medium vehicles by the deep reinforcement learning module under deep reinforcement learning will be introduced below.

[0084] Taking the load analysis of a jet-assisted water entry through a small cavity as an example, a cross-medium vehicle model with a spherical head and a jet as the control volume are considered. In this control volume, the airflow along the duct inside the cross-medium vehicle model has a stable inflow and outflow, that is, the outflow momentum is equal to the inflow momentum. Specifically, before the impact moment, the mass of the ejected air jet in the control volume is m j , the change in gas mass after impact with the small bubble is δm j According to the law of conservation of momentum, MU i +m j (U j +U i )=(M+m)U+(m j +δm j )(U j +U), where M is the mass of the missile model, U i is the water inlet and outlet speed, U j is the outflow velocity of the gas jet, U is the instantaneous velocity of the projectile model after impact, and m is the additional mass of the fluid during impact. Solving the above equations, we can obtain the vertical impact force of the gas jet assisting the entry into the water (assuming that the volume force of the gas is small enough, m j / M and δm j / M can be ignored), that is:

[0085]

[0086] Where R is the spherical radius of the head, B = s / R, s is the immersion depth of the free surface, and the first term on the right (dδm j / dt) is the mass flow rate of the gas jet during the impact. From the above formula, we can see that the impact force has the thrust of the gas jet [U j +U i ]dδm j The term m / M / dt (assuming m / M is small and negligible) has a first-order dependence on the rate of change of the added mass, dm / db. Given that the impact force is controlled by both the jet thrust and dm / db, this means that the reduction in dm / db may be greater than the jet thrust to a certain extent, resulting in a reduction in the impact force. This theoretically explains the load-reducing mechanism of jet-assisted water entry.

[0087] Since the dynamic load of the cross-media vehicle is controlled to a certain extent by the cavitation morphology, maintaining a stable jet cavitation requires flow control. In order to further improve the stability and intelligence of water entry, active control technology based on reinforcement learning is introduced to allow the cross-media vehicle to continuously interact with the environment to obtain the optimal control strategy. Reinforcement learning is an important branch of machine learning. The overall idea is to allow the intelligent agent to continuously learn to accumulate experience to obtain the correct or optimal strategy. In the basic concept, the intelligent agent continuously interacts with the environment and responds to the action of the intelligent agent. t Or the decision π system gives a penalty P t or reward R t Measures. The ultimate goal of the agent is to obtain the maximum cumulative reward G t , whose basic mathematical framework is to solve the Markov decision process.

[0088] like Figure 8 As shown. Usually, the agent takes action a based on the environment at time t. t , in action a t Complete the status S t Transition to new state S t+1 , and feedback reward R t+1 Finally, after continuous iteration and interaction with the environment, the cumulative reward is obtained: γ is the discount factor. The larger γ is, the greater the emphasis on future rewards.

[0089] Based on reinforcement learning active control technology using cavitation morphology as the environment and flow control as the action, a specific reward function is set to achieve the optimal cavitation morphology strategy throughout the entire process of entering and exiting the water and sailing at high speed in water. In addition, a control strategy based on flow control and attitude stabilization is proposed in combination with the cross-medium vehicle attitude to achieve a low-load, high-efficiency high-speed water entry and exit and sailing at high speed in water.

[0090] To further illustrate the principles of the present disclosure, an efficient quantum computing algorithm module will be introduced below.

[0091] Because high-speed water entry and exit times for trans-media vehicles are extremely short, traditional algorithms are completely inadequate for achieving effective cavitation morphology control and matching between the cavitation and the vehicle. To achieve efficient controller computation, the control algorithm of this invention utilizes the quantum computing HHL algorithm for linear algebraic equations. Because the HHL algorithm exhibits exponential acceleration compared to classical algorithms under specific conditions, it exponentially improves computational and perception efficiency, enabling intelligent perception and control for high-speed water entry and exit, as well as for high-speed underwater navigation.

[0092] Classical computing is a logic circuit based on bits and logic gates for information processing. In classical computing, a bit is the smallest unit of information and can only take on one of two fixed values: 0 or 1. Logic gates are used to perform Boolean operations on bits of information, and various operations can be performed by combining universal logic gates. In contrast, quantum computing also uses logic circuits to process information using bits and logic gates, but the properties of these bits and logic gates are fundamentally different from those in classical computing. A quantum bit (also called a quantum bit or qubit) can be either 0 or 1, or simultaneously in a superposition of these two states. Furthermore, quantum logic gates have a wider range of capabilities, including the ability to implement unconventional quantum parallelism and quantum entanglement, operations that are impossible with classical computing. Therefore, while quantum computing and classical computing share certain similarities in their basic form, the fundamental differences between quantum bits and logic gates mean that quantum computing has the potential and capabilities to far surpass classical computing.

[0093] The HHL algorithm is a quantum algorithm for solving linear equations. The linear equation problem can be defined as: given a matrix A and a vector turn up satisfy

[0094] The HHL algorithm mainly includes the following three steps and requires the use of three registers: the right-hand side bit, the storage bit, and the auxiliary bit:

[0095] 1. The first step is to construct the right-hand quantum state, perform phase estimation of the left-hand matrix containing parameters for the storage bits and the right-hand bits, and transfer all integer eigenvalues ​​of the left-hand matrix to the basis vectors of the storage bits.

[0096] 2. The second step is to perform a series of controlled rotations of parameters containing eigenvalues, filter out all eigenvalue-related quantum states, and transfer the eigenvalues ​​from the basis vectors of the storage bits to the amplitudes;

[0097] 3. The third step is to perform inverse phase estimation on the characteristic storage bit and the right-end item bit, and merge the characteristic value of the storage bit amplitude into the right-end item bit. When the auxiliary bit measurement obtains a specific state, the quantum state of the solution can be obtained on the right-end item bit.

[0098] Initial state preparation of HHL algorithm

[0099] The first step in implementing the HHL algorithm is to encode the problem into a quantized language. and After normalization, they are mapped to the quantum states |b> and |x> by encoding them on the amplitude. The original linear system equation problem is transformed into a quantum computing problem:

[0100] A|x>=|b>;

[0101] At this time, the matrix A in the above formula is a Hermitian matrix or a constructed Hermitian matrix, so the matrix A can be spectrally decomposed.

[0102] The spectral decomposition of matrix A is:

[0103]

[0104] where λ j and u j is the eigenpair of matrix A (where λ j is the eigenvalue of matrix A, u j is the eigenvector of matrix A).

[0105] Expand |b> with the eigenvector basis and we get:

[0106]

[0107] So the solution of the linear system equation can be expressed as

[0108]

[0109] It is not difficult to deduce from the above derivation that the basic idea of ​​the algorithm is to construct the solution quantum state |x> starting from the quantum state |b> on the right side. The problem of solving the linear equations is transformed into the problem of solving the eigenvalue information of the matrix A.

[0110] HHL Algorithm Phase Estimation (QPE)

[0111] In order to extract the eigenvalues ​​of the matrix to the amplitude of the solution quantum state, it is necessary to complete the eigenvalue extraction first. QPE quantum circuit can be used for eigenvalue extraction. Quantum phase estimation (QPE) can calculate the phase of the eigenvalue of a given unitary operator U, that is, solve Here |φ> is the eigenvector of U.

[0112] Quantum phase estimation involves the following steps:

[0113] 1. Transfer the eigenvalue phase decomposition of U to the amplitude of the auxiliary quantum bit through a series of special rotation quantum gate operations;

[0114] 2. Perform IQFT on the auxiliary qubit to transfer the eigenvalue phase on the amplitude to the basis vector;

[0115] 3. After measuring the basis vectors of the auxiliary quantum bits separately, the phase information of the eigenvalue can be obtained.

[0116] right Perform a phase estimation operation and get

[0117]

[0118] in is the corresponding eigenvalue λ j Through the process of quantum phase estimation (QPE), the characteristic information of matrix A is stored in the basis vector middle.

[0119] HHL algorithm controlled rotation

[0120] The purpose of the controlled rotation operation in the HHL algorithm is to Convert to In this step we will introduce an auxiliary qubit, according to The value in the eigenvalue register, when the introduced auxiliary quantum bit passes through the controlled gate, the state of the system will change to:

[0121]

[0122] The controlled rotation operation can realize the transfer of effective quantum information from the register to the quantum state amplitude. By adding quantum bits, the inverse of the eigenvalue is extracted proportionally to the probability radius of the corresponding ground state.

[0123] HHL algorithm for inverse phase estimation

[0124] Theoretically, the quantum state after controlled rotation can be measured to obtain the solution quantum state. j Same but Different quantum states that need to be merged The inverse QPE operation should be chosen to obtain the form The resulting quantum state.

[0125] Perform inverse QPE on the rotation result, and we have:

[0126]

[0127] Measure on the first qubit, when the measurement is |1>, we can get Cλ j b j |u j >, if it is |0>, recalculate. Because the HHL algorithm calculates the expected value of the operator associated with the solution of the linear system equation, rather than the solution itself, while theoretically reducing the time complexity of conventional algorithms, the HHL algorithm can only return an approximate solution, not an exact one. Simply measuring the result when the additional bit is |1> will yield an expression proportional to the solution to the linear system.

[0128] Although the embodiments of the present application have been shown and described above, the scope of protection of the present invention is not limited thereto, and any changes or substitutions that are not conceivable through creative work should be included in the scope of protection of the present invention; unless expressly stated, any elements, actions or instructions used in this document should not be interpreted as critical or necessary.

Claims

1. An intelligent load reduction and drag reduction control method, characterized in that: The following steps are involved: A biomimetic side-line speed sensor array was used to collect voltage data in calibration tests, and a gradient descent deep learning method was used to train and evaluate the speed-voltage model. A biomimetic whisker resistance sensor array was used to collect current data in calibration experiments, and a gradient descent deep learning method was used to train and evaluate the resistance-current model. A supercavitation perception model is constructed based on the underwater resistance formula, velocity-voltage model and resistance-current model; The cavitation morphology, ventilation volume, vehicle resistance, and acceleration data of the cross-medium vehicle were obtained during water entry and exit tests and underwater navigation tests. The supercavitation perception model was trained using reinforcement learning methods to obtain the optimal strategy model for supercavitation intelligent control. The real-time voltage and current data of the cross-medium vehicle are input into the supercavitation intelligent control model, and the cavitation morphology is sensed based on the real-time speed and resistance. The cavitator ventilation volume and the attitude of the cross-medium vehicle are adjusted in real time according to the sensed cavitation morphology, so as to adjust the cavitation morphology and the attitude of the cross-medium vehicle to the optimal state.

2. The intelligent load reduction and drag reduction control method according to claim 1, characterized in that: The relationship of the supercavitation perception model is: Water resistance: Wetting section length: L1 = L-L2; From the above formula, we can get the distance between the supercavitation bubble and the boundary of the vehicle, that is, the length of the dry section: in: ρ D : density of water; U: speed; C D : drag coefficient of carbon fiber whisker rod; The bionic tentacle resistance sensor has a carbon fiber tentacle rod with a length of L and a diameter of H. The bionic tentacle resistance sensor is located at the head of the cross-medium vehicle. The portion of the carbon fiber tentacle rod outside the cavitation is the wetted section L1, and the dry section inside the cavitation is L. 2， L2 is the distance between the supercavitation boundary and the surface of the trans-medium vehicle.

3. The intelligent load reduction and drag reduction control method according to claim 2, characterized in that: The real-time adjustment is to introduce active control based on reinforcement learning, so that the cross-medium vehicle can continuously interact with the environment to obtain the optimal control strategy.

4. The intelligent load reduction and drag reduction control method according to claim 3, characterized in that: The specific method of the reinforcement learning active control is: 1) Give a penalty P to the action or decision system of the cross-medium vehicle t or reward R t measure; 2) The ultimate goal of the cross-medium vehicle is to obtain the maximum cumulative reward G t , the cross-medium vehicle takes action a based on the environment at time t t , in action a t Complete the status S t Transition to new state S t+1 , and feedback reward R t+1 ; 3) Finally, after continuous iteration and interaction with the environment, cumulative rewards are obtained: Among them, γ is the discount factor, and the larger γ is, the greater the emphasis on future rewards.

5. A cross-medium vehicle using the intelligent load reduction and drag reduction control method according to any one of claims 1 to 4, characterized in that: include: a fuselage having a head support bar; a front wing foldably disposed on the fuselage; a rear wing foldably disposed on the fuselage; foldable tilt-rotor; Cross-medium thrusters; cavitator; as well as Supercavitation intelligent controller; Wherein, the head support rod is provided with a bionic side linear velocity sensor array, and the fuselage head is provided with a bionic tentacle resistance sensor array; The foldable tilt rotor, the trans-medium thruster, the cavitator, the bionic side linear velocity sensor array and the bionic tentacle resistance sensor array are all electrically connected to the supercavitation intelligent controller.

6. The cross-media vehicle according to claim 5, characterized in that: The supercavitation intelligent controller has an efficient quantum computing algorithm module.

7. The cross-media vehicle according to claim 6, characterized in that: The supercavitation intelligent controller has a deep reinforcement learning module.

Images