Comprehensive electro-hydraulic drive control system for taking-off and landing of unmanned aerial vehicle

Abstract

The invention, which relates to the technical field of the unmanned aerial vehicle technology, discloses an unmanned aerial vehicle take-off and landing comprehensive electro-hydraulic drive control system comprising the following components: a data acquisition module, a deep learning model module, a drive control strategy generation module, and an electro-hydraulic drive execution module. Comprehensive environment information and historical take-off and landing data of the unmanned aerial vehicle are collected through the data acquisition module, a deep learning model is used for accurate prediction, an optimal driving control strategy is generated, in the take-off process, the system can accurately calculate the needed initial driving pressure and speed, dynamic adjustment is carried out according to different environment factors, and the driving control precision of the unmanned aerial vehicle is improved. According to the comprehensive electro-hydraulic driving and controlling system, the unmanned aerial vehicle can stably take off, when the unmanned aerial vehicle lands, a terrain perception and obstacle avoidance mechanism is introduced into the system, a three-dimensional topographic map is constructed in real time, obstacles are detected, a safe landing path and a driving and controlling strategy are generated, the landing accuracy and safety are improved, and the comprehensive electro-hydraulic driving and controlling system remarkably improves the taking-off and landing precision and stability of the unmanned aerial vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) technology, and in particular to an integrated electro-hydraulic drive control system for the take-off and landing of a UAV. Background Art

[0002] With the rapid development of drone technology, drones are being used more and more widely in various fields. However, the take-off and landing process of drones is greatly affected by environmental factors such as wind speed, wind direction, temperature, and humidity. Changes in these factors directly affect the take-off and landing stability of drones.

[0003] At present, most existing UAV drive and control systems use traditional control algorithms. These algorithms cannot fully consider the impact of complex and changeable environmental factors on UAV take-off and landing, and cannot automatically generate the optimal drive and control strategy according to different environmental conditions. When facing complex environments, the adaptability and reliability of traditional drive and control systems are poor, which limits the application of UAVs in a wider range of scenarios. Therefore, it is particularly important to develop a comprehensive electro-hydraulic drive and control system for UAV take-off and landing. Summary of the Invention

[0004] The purpose of this invention is to make up for the shortcomings of the existing technology and provide an integrated electro-hydraulic drive control system for UAV take-off and landing. It can collect comprehensive environmental information and historical take-off and landing data of the UAV through the data acquisition module, use the deep learning model to make accurate predictions, and generate the optimal drive control strategy. During the take-off process, the system can accurately calculate the required initial drive pressure and speed, and dynamically adjust according to different environmental factors to ensure stable take-off of the UAV.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: an integrated electro-hydraulic drive control system for UAV take-off and landing, the system comprising the following components: a data acquisition module, a deep learning model module, a drive control strategy generation module, and an electro-hydraulic drive execution module;

[0006] The data acquisition module is used to collect historical take-off and landing data and environmental information of the UAV, and transmit the collected data to the deep learning model module. Specifically, the historical take-off and landing data not only covers the take-off time, take-off position, take-off speed, landing time, landing position, landing speed, and UAV attitude changes, but also includes the operating parameters of the electro-hydraulic drive system during each take-off and landing, the change curve of the drive pressure, the fluctuation range of the drive speed, and the temperature change of the hydraulic oil. In addition to wind speed, wind direction, temperature, humidity, precipitation amount, precipitation type, and atmospheric pressure, the environmental information also includes light intensity, air density, and atmospheric turbulence intensity;

[0007] On the UAV body, a distributed, multi-dimensional sensor layout is adopted, and multiple wind speed sensors are installed at the nose, wings, and tail. Through the principle of triangulation, the wind speed vectors at different positions are accurately calculated to obtain more comprehensive airflow information. High-precision temperature sensors are arranged in different functional areas inside the fuselage to monitor the temperature changes of various parts in real time. Raindrop sensors are distributed in an array on the top, side, and bottom of the UAV. They can not only detect the amount and type of precipitation, but also judge the movement trajectory and angle of raindrops by the time difference and intensity change of raindrop impact. The humidity sensor adopts intelligent micro-electromechanical system technology and is integrated near the key circuit boards of electronic equipment to accurately measure the local humidity environment. The air pressure sensor is combined with the inertial measurement unit and installed at the center of gravity of the UAV to achieve high-precision measurement of atmospheric pressure and altitude. The data collected by these sensors are amplified, filtered, and converted into analog by the pre-processing circuit, and then transmitted to the deep learning model module using a specific communication protocol.

[0008] The deep learning model module: a deep intelligent model that integrates a built-in quantum-inspired neural network and adaptive dynamic programming. The algorithm is as follows:

[0009]

[0010] Among them, H(s,a) represents the comprehensive evaluation function of taking action a under state s, s is a high-dimensional state vector that integrates environmental information and the UAV's own state, a is the action set of the electro-hydraulic drive control system, including continuous variables such as driving pressure, driving speed, and driving frequency, as well as some discrete control mode selections, φ ij (s,a) is a set of basis functions based on quantum state encoding, which is used to capture the complex nonlinear characteristics of state-action pairs. n and m represent the two dimensions of the basis functions, ω ij is the same as φ ij The weights corresponding to (s, a) are first globally searched in the high-dimensional weight space using the quantum annealing algorithm to find a near-optimal solution, reducing the risk of falling into a local optimum. The obtained solution is then optimized and fine-tuned using a genetic algorithm, through selection, crossover, and mutation operations, so that the weights can better adapt to the training data.

[0011] ψ k (s) is a set of characteristic functions that only depend on state s and is used to extract the inherent characteristics of the state. p is the number of characteristic functions, and λ kare the corresponding weights, which are adjusted through an adaptive dynamic programming algorithm. During the training process, the weights are dynamically optimized based on the system feedback and state transition information to minimize the difference between the prediction error and the actual result. During the training process, a quantum-classical hybrid training strategy based on reinforcement learning is adopted. The agent interacts with the environment. The state s, action a, reward r, and next state s′ generated by each interaction form a four-tuple (s, a, r, s′), which is stored in a quantum-classical hybrid memory pool. A batch of four-tuples are randomly sampled from the memory pool, and the model parameters are updated by minimizing the following loss function:

[0012]

[0013] Where Θ is the set of all parameters of the model, M is the data distribution in the quantum-classical hybrid memory pool, is the target value, and γ is the discount factor. This deep learning model learns and analyzes a large amount of historical take-off and landing data and environmental information to establish a complex mapping relationship between environmental factors and the take-off and landing status of the UAV, and predict the impact of different environmental factors on the take-off and landing of the UAV.

[0014] The drive control strategy generation module automatically generates the optimal drive control strategy based on the prediction results, specifically by solving the following optimization problem:

[0015]

[0016] Where A is the action space, C(s,a) is the immediate cost function of taking action a in the current state s, β is the discount factor for future costs, T is the time span of the prediction, and E[H(s t ,a t )] is the state s at time t in the future t Next take action a t The expected comprehensive evaluation function value is used to predict the conditions that the drone may encounter during takeoff based on the current environmental information before takeoff. The electro-hydraulic drive parameters are adjusted based on the optimization results. The dynamic response characteristics of the electro-hydraulic drive system are also considered to dynamically adjust the parameters. During landing, the environment changes are analyzed in real time and the drive control strategy is dynamically adjusted, taking into account the terrain information and obstacles on the ground.

[0017] The electro-hydraulic drive execution module controls the actuator of the electro-hydraulic drive system according to the drive control strategy generated by the drive control strategy generation module to achieve takeoff and landing control of the UAV. This module adopts an intelligent electro-hydraulic servo drive architecture and integrates advanced micro-electromechanical system sensors and high-speed digital signal processors. The electro-hydraulic servo valve adopts a piezoelectric ceramic driven valve core structure. By precisely controlling the voltage of the piezoelectric ceramic, high-precision displacement control of the valve core is achieved, thereby accurately regulating the flow and pressure of the hydraulic oil. The driver adopts a permanent magnet synchronous motor, combined with a vector control algorithm and a direct torque control algorithm, to achieve rapid response and precise control of the motor.

[0018] During the control process, the electro-hydraulic drive execution module collects the pressure, flow, and temperature feedback information of the electro-hydraulic system, as well as the attitude and position status information of the UAV in real time, and transmits them to the DSP through a high-speed communication bus. Based on the received information and the instructions of the drive control strategy generation module, the DSP uses an adaptive sliding mode control algorithm to adjust the electro-hydraulic servo valve and driver in real time. At the same time, the module also has fault diagnosis and fault-tolerant control functions. Through real-time monitoring and analysis of sensor data and system operating parameters, it can detect system faults in a timely manner and adopt corresponding fault-tolerant control strategies.

[0019] Furthermore, the sensor data fusion mechanism in the data acquisition module uses a Bayesian network data fusion algorithm to fully utilize the data collected by different sensors, as the data have different temporal resolutions, spatial resolutions, and data formats. The data collected by each sensor is independently preprocessed and feature extracted, and the raw data is converted into a representative feature vector. A Bayesian network model is constructed. The model uses environmental factors and drone status as nodes and sensor data as edges. The dependency relationship between nodes is described by a conditional probability table. During the data fusion process, the posterior probability of each node is calculated according to the Bayesian formula:

[0020]

[0021] Among them, X i represents an environmental factor or drone state node, E represents the data set collected by all sensors, P(X i ) is node X i The prior probability, P(E|X i ) is at a given node X i The likelihood probability of observing data E in state;

[0022] Through this data fusion algorithm, data from multiple sensors can be organically integrated to improve data accuracy and reliability. When calculating the airflow state around the drone, the fusion of data from wind speed sensors, air pressure sensors, and inertial measurement units can more accurately assess the direction, speed, and turbulence intensity of the airflow, providing more precise input information for the deep learning model, thereby improving the model's prediction accuracy and the accuracy of the drive and control strategy.

[0023] Furthermore, the quantum-classical hybrid memory pool management mechanism in the deep learning model module consists of quantum storage units and classical storage units. The quantum storage unit uses quantum bit technology to store data in the form of quantum states, with high storage density and parallel processing capabilities. The classical storage unit uses traditional random access memory and flash memory technology to store regular data and intermediate calculation results.

[0024] During data storage, quantum storage units are used to store data with high uncertainty and complexity, such as sensor data collected in extreme environments or high-dimensional feature vectors during model training. Through quantum coding technology, the data is mapped into quantum state space, and the characteristics of quantum superposition and entanglement are used to achieve efficient storage and fast retrieval of data. For conventional state-action pair data and reward information, classical storage units are used for storage.

[0025] In the data sampling process, the quantum Monte Carlo method and the classical random sampling method are combined. For the data in the quantum storage unit, the quantum Monte Carlo method is used to sample, and representative data samples are obtained by measuring and statistics of the quantum state. For the data in the classical storage unit, the classical random sampling method is used for sampling. Through this hybrid sampling method, the advantages of quantum and classical storage can be fully utilized, the efficiency and quality of data sampling can be improved, thereby accelerating the model training process and enhancing the generalization ability of the model.

[0026] Furthermore, the optimization process of the adaptive dynamic programming algorithm in the deep learning model module introduces a multi-agent collaborative optimization strategy to regard the deep learning model as a collection of multiple agents. Each agent is responsible for processing a part of the state space and action space. During the training process, these agents jointly optimize the model parameters through information interaction and collaboration. Specifically, each agent calculates the local optimal strategy and value function based on its own local observations and experience, and exchanges their respective strategy and value function information through the communication network between agents. In the process of information interaction, a fusion algorithm with a consensus mechanism is adopted. Each agent updates its own strategy and value function based on the information received from other agents, so that the strategies and value functions of all agents gradually tend to be consistent. The algorithm formula is:

[0027]

[0028] Among them, θ i (t) represents the parameter vector of the i-th agent at time t, N is the total number of agents, a ij (t) is the communication weight between agents i and j, satisfying

[0029] Through this multi-agent collaborative optimization strategy, we can fully utilize the experience and knowledge of different agents in different state spaces and action spaces, accelerate the convergence speed of the adaptive dynamic programming algorithm, and improve the optimization effect of the model, so that the deep learning model can more accurately predict the impact of different environmental factors on the take-off and landing of drones.

[0030] Furthermore, the control strategy generation module takes into account the subdivision and refined control of the drone's takeoff phase when generating the takeoff control strategy. The takeoff process is divided into three stages: initial acceleration, attitude adjustment, and stable ascent.

[0031] During the initial acceleration phase, the deep learning model's prediction of the current environment, combined with the drone's payload and battery level information, accurately calculates the required initial driving pressure and speed. At high altitudes, due to the thin air, greater driving pressure is required to provide sufficient lift. In high-temperature environments, battery performance may degrade, requiring adjustment of driving parameters to ensure proper motor operation.

[0032] During the attitude adjustment phase, the UAV's attitude changes are monitored in real time using data from the inertial measurement unit and visual sensors. When attitude deviation is detected, the drive control strategy generation module adjusts the various actuators of the electro-hydraulic drive system based on the size and direction of the deviation to correct the attitude deviation and ensure that the UAV maintains a stable attitude during takeoff.

[0033] During the stable ascent phase, the drive frequency and power are dynamically adjusted according to the current wind speed, wind direction and atmospheric turbulence to achieve the optimal ascent speed and energy consumption. At the same time, the interaction between the UAV and the surrounding environment and the interference of airflow on the UAV are taken into account. By optimizing the drive control strategy, the UAV can maintain a stable ascent trajectory in complex environments and avoid altitude fluctuations and attitude loss of control caused by airflow disturbances.

[0034] Furthermore, the control strategy generation module introduces terrain perception and obstacle avoidance mechanisms when generating landing control strategies. The drone is equipped with a variety of terrain perception and obstacle avoidance equipment such as lidar, millimeter-wave radar and visual sensors.

[0035] During landing, the LiDAR scans the terrain below the drone in real time to build a three-dimensional topographic map. The millimeter-wave radar is used to detect surrounding obstacles, especially in low-visibility environments. The visual sensor uses image recognition algorithms to identify target areas and obstacles on the ground.

[0036] The control strategy generation module uses the information collected by these sensors and the prediction results of the deep learning model to generate a safe landing path and control strategy. When an obstacle is detected in the landing area, the control strategy generation module adjusts the drone's landing direction and altitude to bypass the obstacle. When approaching the ground, the control strategy generation module dynamically adjusts the drone's descent rate and attitude based on the undulations of the terrain to ensure a smooth landing.

[0037] At the same time, in order to improve the accuracy and safety of landing, a method based on model predictive control (MPC) is adopted. The MPC algorithm optimizes the control input in the future based on the current state and future prediction information, so that the UAV can land according to the predetermined trajectory. By continuously updating the prediction and adjusting the control strategy, it can effectively respond to environmental changes and uncertainties and improve the reliability of the landing process.

[0038] Furthermore, the adaptive sliding mode control algorithm parameter adjustment mechanism in the electro-hydraulic drive execution module introduces a parameter adaptive adjustment method based on fuzzy logic. The core of sliding mode control is to design a sliding surface so that the system state can run stably on the sliding surface. In this system, the parameters of the sliding surface are adaptively adjusted according to the state and environmental information of the UAV. The fuzzy logic controller adjusts the parameters of the sliding mode control through a set of fuzzy rules based on the current system error and error change rate input variables. The design of fuzzy rules is based on in-depth analysis and experience summary of the UAV take-off and landing process. When the system error is large, the sliding mode gain is increased to speed up the system response speed. When the error change rate is small, the sliding mode gain is reduced to reduce the system chattering. The form of the fuzzy rules is as follows:

[0039]

[0040] Among them, R k represents the kth fuzzy rule, e is the system error, is the error rate of change, and is a fuzzy set, Δp i is the sliding mode control parameter that needs to be adjusted, is the corresponding fuzzy output set;

[0041] Through this fuzzy logic-based parameter adaptive adjustment method, the adaptive sliding mode control algorithm can better adapt to various uncertainties and interferences during the take-off and landing process of the UAV, improve the control accuracy and robustness of the electro-hydraulic drive execution module, and ensure that the take-off and landing actions of the UAV can be executed more accurately and smoothly.

[0042] Furthermore, the system has a multi-level fault diagnosis and fault-tolerant control architecture, covering four levels: sensors, deep learning models, drive control strategy generation modules, and electro-hydraulic drive execution modules;

[0043] Sensor fault diagnosis uses a data-driven approach, building a normal operating model based on historical data. When the data deviates from the model, statistical analysis and machine learning algorithms are used to determine the fault type and location. For example, a Gaussian mixture model is used for wind speed sensors, and an ARMA model is used for temperature sensors.

[0044] Deep learning model fault diagnosis is achieved by monitoring training and prediction results. When the prediction error exceeds the threshold, the parameters and structure are analyzed to determine whether it is overfitting, underfitting, or parameter abnormality. In the case of overfitting, the regularization strength or data augmentation is adjusted, and in the case of underfitting, the model complexity is increased.

[0045] The drive control strategy generation module fault diagnosis is based on the rationality and feasibility of the strategy, and the strategy is simulated and verified. If problems such as unstable posture and abnormal energy consumption occur, the prediction results, optimization algorithm and input data are analyzed to find the root cause of the fault. If necessary, the optimization strategy is adjusted or the data is repaired.

[0046] Fault diagnosis of the electro-hydraulic drive execution module is performed by monitoring the electro-hydraulic system parameters and motor status. When the pressure, flow, current, and voltage parameters are abnormal, the characteristics are analyzed to determine the fault type. For example, the pressure and flow curves are used to analyze electro-hydraulic servo valve faults, and the electromagnetic and mechanical characteristics are used to analyze motor faults.

[0047] In terms of fault-tolerant control, when a sensor fails, a backup sensor is switched or other sensor data is integrated. Deep learning model failures are adjusted according to type. When the drive control strategy generation module fails, a backup algorithm or a safety strategy is used. When the electro-hydraulic drive execution module fails, corresponding measures are taken according to the severity. When multiple faults occur simultaneously, they are handled according to priority, and strategies are adjusted in real time.

[0048] In addition, the system has a fault recording and reporting function, which records the fault time, location, and type information, and generates a report containing a fault overview, cause analysis, treatment measures, and performance impact assessment, to assist in troubleshooting and system optimization, improve system reliability and stability, and ensure the safe takeoff and landing of drones.

[0049] Compared with the existing technology, this UAV take-off and landing integrated electro-hydraulic drive control system has the following beneficial effects:

[0050] 1. This system collects comprehensive environmental information and historical takeoff and landing data of drones through a data acquisition module, uses a deep learning model to make accurate predictions, and generates the optimal drive control strategy. During takeoff, the system can accurately calculate the required initial drive pressure and speed, and dynamically adjust according to different environmental factors to ensure stable takeoff of the drone. During landing, the system introduces terrain perception and obstacle avoidance mechanisms, constructs a three-dimensional terrain map and detects obstacles in real time, generates a safe landing path and drive control strategy, and improves the accuracy and safety of landing. This integrated electro-hydraulic drive control system significantly improves the accuracy and stability of drone takeoff and landing.

[0051] 2. This system has a multi-level fault diagnosis and fault-tolerant control architecture, covering four levels: sensors, deep learning models, drive control strategy generation modules, and electro-hydraulic drive execution modules. Through data-driven methods, monitoring training and prediction results, strategy rationality and feasibility analysis, and electro-hydraulic system parameters and motor status monitoring, the system can detect and diagnose faults in a timely manner. In terms of fault-tolerant control, the system takes measures such as switching backup sensors, fusing other sensor data, adjusting deep learning models, switching backup algorithms, or using safety strategy measures to ensure that the UAV can still take off and land safely in the event of a fault. This powerful fault diagnosis and fault-tolerant capability improves the reliability and stability of the system, and provides a strong guarantee for the safe flight of the UAV.

[0052] Other advantages, objects and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be learned from the practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0054] Figure 1 This is a process operation diagram of an integrated electro-hydraulic drive control system for UAV take-off and landing. DETAILED DESCRIPTION

[0055] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0056] Example 1

[0057] This example describes a mission located in a mountainous area with complex terrain, including valleys and peaks. The climate is changeable, with frequent strong winds, large temperature fluctuations, and potential precipitation in some areas. The mission involves using a drone to transport emergency medical supplies to a remote village within the mountainous area.

[0058] Before takeoff, multiple wind speed sensors mounted on the nose, wings, and tail of the drone use triangulation to accurately measure wind speed vectors at different locations, capturing complex airflow information within valleys, such as changes in wind direction and intensity. High-precision temperature sensors in different functional areas of the drone's fuselage monitor the operating temperature of its internal equipment in real time under varying ambient temperatures. For example, low battery temperatures in low-temperature environments may affect flight endurance. Arrayed raindrop sensors on the top, sides, and bottom detect sudden precipitation in mountainous areas, including the amount and type of precipitation, as well as the trajectory and angle of raindrops. They identify localized light rain with large inclination angles. A humidity sensor using intelligent micro-electromechanical system technology, integrated near key electronic circuit boards, measures the potential impact of high humidity in mountainous areas on electronic equipment. A barometric pressure sensor, combined with an inertial measurement unit (IMU), located at the drone's center of gravity, accurately measures atmospheric pressure and altitude changes, providing an altitude reference for mountain flight. The data collected by these sensors undergoes signal amplification, filtering, and analog-to-digital conversion in a preprocessing circuit before being transmitted to the deep learning model module using a specific communication protocol.

[0059] A deep intelligent model that integrates built-in quantum-inspired neural networks and adaptive dynamic programming, with a comprehensive evaluation function in the model Where s is a high-dimensional state vector that integrates mountain environment information (such as wind speed, wind direction, temperature, humidity, precipitation, and atmospheric pressure) and the UAV's own state (such as attitude and power), a is the action set of the electro-hydraulic drive control system (such as driving pressure, driving speed, and driving frequency), and φ ij (s,a) The basis function based on quantum state encoding captures the complex nonlinear characteristics of state-action pairs, ω ij is its corresponding weight, which is fine-tuned by global search of quantum annealing algorithm and optimization of genetic algorithm, ψ k (s) Extract state inherent features, λ k is its corresponding weight, which is adjusted by the adaptive dynamic programming algorithm. The training process adopts a quantum-classical hybrid training strategy based on reinforcement learning. The state s, action a, reward r and next state s′ generated by the interaction between the agent and the environment form a quadruple (s, a, r, s′) and are stored in the quantum-classical hybrid memory pool. A batch of quadruple is randomly sampled from the memory pool and the loss function L(Θ) = E is minimized. (s,a,r,s′)～M [(yH(s,a;θ)) 2] (where y = r + γ·maxH(s′, a′; θ), γ is a discount factor) updates the model parameters and establishes a complex mapping relationship between mountain environmental factors and the take-off and landing status of UAVs by learning and analyzing a large amount of historical take-off and landing data and environmental information. This can predict the impact of different factors on the take-off and landing of UAVs in mountainous environments, such as the possibility of encountering strong wind shear when flying in a valley.

[0060] Generate optimal drive control strategies based on deep learning model prediction results to solve optimization problems Where A is the action space, C(s,a) is the immediate cost function of taking action a in the current state s, β is the future cost discount factor, and T is the prediction time span. Before takeoff, considering the complex terrain and changeable climate in mountainous areas, the conditions that may be encountered during takeoff, such as strong winds and low temperatures, are predicted based on current environmental information. Combined with the drone's payload and battery power information, the initial drive pressure and speed are accurately calculated. In the initial acceleration phase, since the air density in the mountains may be uneven, the drive pressure is adjusted to ensure a smooth takeoff of the drone. In the attitude adjustment phase, the inertial measurement unit and visual sensor data are used to monitor the drone's attitude changes in real time. When encountering airflow between valleys causing attitude deviations, the electro-hydraulic drive system actuators are adjusted according to the size and direction of the deviation. In the stable ascent phase, the drive frequency and power are dynamically adjusted according to wind speed, wind direction, and atmospheric turbulence, and the drive control strategy is optimized to enable the drone to maintain a stable ascent trajectory in complex mountainous environments.

[0061] The electro-hydraulic drive system actuator is controlled according to the drive control strategy, adopting an intelligent electro-hydraulic servo drive architecture, integrating advanced micro-electromechanical system sensors and high-speed digital signal processors. The electro-hydraulic servo valve adopts a piezoelectric ceramic drive valve core structure, and achieves high-precision valve core displacement control by precisely controlling the piezoelectric ceramic voltage, and accurately adjusts the hydraulic oil flow and pressure. The driver adopts a permanent magnet synchronous motor, combined with vector control algorithm and direct torque control algorithm to achieve fast response and precise control. During the control process, the electro-hydraulic system pressure, flow, temperature feedback information and UAV attitude and position status information are collected in real time and transmitted to the DSP through a high-speed communication bus. The DSP generates module instructions based on the received information and the drive control strategy. An adaptive sliding mode control algorithm is used to adjust the electro-hydraulic servo valve and driver in real time. The sliding surface parameters are adaptively adjusted according to the UAV status and mountain environment information. The fuzzy logic controller inputs variables based on the system error and error change rate, and adjusts the sliding mode control parameters through fuzzy rules. For example, when the UAV encounters strong wind interference during the ascent, resulting in a large attitude deviation, the fuzzy logic controller increases the sliding mode gain to speed up the system response speed, allowing the UAV to quickly restore a stable attitude. At the same time, the module has fault diagnosis and fault-tolerant control functions. By monitoring sensor data and system operating parameters, it can detect system faults in a timely manner and adopt corresponding fault-tolerant control strategies to ensure the safe and stable operation of the UAV in mountain transportation missions.

[0062] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. An integrated electro-hydraulic control system for take-off and landing of unmanned aerial vehicles, characterized in that: The system includes the following components: data acquisition module, deep learning model module, drive control strategy generation module, and electro-hydraulic drive execution module; The data acquisition module is used to collect historical take-off and landing data and environmental information of the UAV, and transmit the collected data to the deep learning model module. Specifically, the historical take-off and landing data not only covers the take-off time, take-off position, take-off speed, landing time, landing position, landing speed, and UAV attitude changes, but also includes the operating parameters of the electro-hydraulic drive system during each take-off and landing, the change curve of the drive pressure, the fluctuation range of the drive speed, and the temperature change of the hydraulic oil. In addition to wind speed, wind direction, temperature, humidity, precipitation amount, precipitation type, and atmospheric pressure, the environmental information also includes light intensity, air density, and atmospheric turbulence intensity; On the UAV body, a distributed, multi-dimensional sensor layout is adopted, and multiple wind speed sensors are installed at the nose, wings, and tail. Through the principle of triangulation, the wind speed vectors at different positions are accurately calculated to obtain more comprehensive airflow information. High-precision temperature sensors are arranged in different functional areas inside the fuselage to monitor the temperature changes of various parts in real time. Raindrop sensors are distributed in an array on the top, side, and bottom of the UAV. They can not only detect the amount and type of precipitation, but also judge the movement trajectory and angle of raindrops by the time difference and intensity change of raindrop impact. The humidity sensor adopts intelligent micro-electromechanical system technology and is integrated near the key circuit boards of electronic equipment to accurately measure the local humidity environment. The air pressure sensor is combined with the inertial measurement unit and installed at the center of gravity of the UAV to achieve high-precision measurement of atmospheric pressure and altitude. The data collected by these sensors are amplified, filtered, and converted into analog by the pre-processing circuit, and then transmitted to the deep learning model module using a specific communication protocol. The deep learning model module: a deep intelligent model that integrates a built-in quantum-inspired neural network and adaptive dynamic programming. The algorithm is as follows: Among them, H(s,a) represents the comprehensive evaluation function of taking action a under state s, s is a high-dimensional state vector that integrates environmental information and the UAV's own state, a is the action set of the electro-hydraulic drive control system, including continuous variables such as driving pressure, driving speed, and driving frequency, as well as some discrete control mode selections, φ ij (s,a) is a set of basis functions based on quantum state encoding, which is used to capture the complex nonlinear characteristics of state-action pairs. n and m represent the two dimensions of the basis functions, ω ij is the same as φ ij The weights corresponding to (s, a) are first globally searched in the high-dimensional weight space using the quantum annealing algorithm to find a near-optimal solution, reducing the risk of falling into a local optimum. The obtained solution is then optimized and fine-tuned using a genetic algorithm, through selection, crossover, and mutation operations, so that the weights can better adapt to the training data. ψ k (s) is a set of characteristic functions that only depend on state s and is used to extract the inherent characteristics of the state. p is the number of characteristic functions, and λ k are the corresponding weights, which are adjusted through an adaptive dynamic programming algorithm. During the training process, the weights are dynamically optimized according to the system feedback and state transition information to minimize the difference between the prediction error and the actual result. During the training process, a quantum-classical hybrid training strategy based on reinforcement learning is adopted. The agent interacts with the environment, and each interaction generates a state s, an action a, a reward r, and a next state s. ′ Form a four-tuple (s,a,r,s ′ ), stored in a quantum-classical hybrid memory pool, a batch of quads are randomly sampled from the memory pool, and the model parameters are updated by minimizing the following loss function: Where Θ is the set of all parameters of the model, M is the data distribution in the quantum-classical hybrid memory pool, is the target value, and γ is the discount factor. This deep learning model learns and analyzes a large amount of historical take-off and landing data and environmental information to establish a complex mapping relationship between environmental factors and the take-off and landing status of the UAV, and predict the impact of different environmental factors on the take-off and landing of the UAV. The drive control strategy generation module automatically generates the optimal drive control strategy based on the prediction results, specifically by solving the following optimization problem: Where A is the action space, C(s,a) is the immediate cost function of taking action a in the current state s, β is the discount factor for future costs, T is the time span of the prediction, and E[H(s t ,a t )] is the state s at time t in the future t Next take action a t The expected comprehensive evaluation function value is used to predict the conditions that the drone may encounter during takeoff based on the current environmental information before takeoff. The electro-hydraulic drive parameters are adjusted based on the optimization results. The dynamic response characteristics of the electro-hydraulic drive system are also considered to dynamically adjust the parameters. During landing, the environment changes are analyzed in real time and the drive control strategy is dynamically adjusted, taking into account the terrain information and obstacles on the ground. The electro-hydraulic drive execution module controls the actuator of the electro-hydraulic drive system according to the drive control strategy generated by the drive control strategy generation module to achieve takeoff and landing control of the UAV. This module adopts an intelligent electro-hydraulic servo drive architecture and integrates advanced micro-electromechanical system sensors and high-speed digital signal processors. The electro-hydraulic servo valve adopts a piezoelectric ceramic driven valve core structure. By precisely controlling the voltage of the piezoelectric ceramic, high-precision displacement control of the valve core is achieved, thereby accurately regulating the flow and pressure of the hydraulic oil. The driver adopts a permanent magnet synchronous motor, combined with a vector control algorithm and a direct torque control algorithm, to achieve rapid response and precise control of the motor. During the control process, the electro-hydraulic drive execution module collects the pressure, flow, and temperature feedback information of the electro-hydraulic system, as well as the attitude and position status information of the UAV in real time, and transmits them to the DSP through a high-speed communication bus. Based on the received information and the instructions of the drive control strategy generation module, the DSP uses an adaptive sliding mode control algorithm to adjust the electro-hydraulic servo valve and driver in real time. At the same time, the module also has fault diagnosis and fault-tolerant control functions. Through real-time monitoring and analysis of sensor data and system operating parameters, it can detect system faults in a timely manner and adopt corresponding fault-tolerant control strategies.

2. The integrated electro-hydraulic control system for take-off and landing of a UAV according to claim 1, characterized in that: The sensor data fusion mechanism in the data acquisition module uses a Bayesian network data fusion algorithm to fully utilize the data collected by different sensors, as the data have different temporal resolutions, spatial resolutions, and data formats. The data collected by each sensor is independently preprocessed and feature extracted, and the raw data is converted into a representative feature vector. A Bayesian network model is constructed. The model uses environmental factors and drone status as nodes and sensor data as edges. The dependency relationship between nodes is described by a conditional probability table. During the data fusion process, the posterior probability of each node is calculated according to the Bayesian formula: Among them, X i represents an environmental factor or drone state node, E represents the data set collected by all sensors, P(X i ) is node X i The prior probability, P(E|X i ) is at a given node X i The likelihood probability of observing data E in state .

3. The integrated electro-hydraulic control system for take-off and landing of a UAV according to claim 1, characterized in that: The quantum-classical hybrid memory pool management mechanism in the deep learning model module consists of quantum storage units and classical storage units. The quantum storage units use quantum bit technology to store data in the form of quantum states, with high storage density and parallel processing capabilities. The classical storage units use traditional random access memory and flash memory technology to store regular data and intermediate calculation results. In the data storage process, for data with high uncertainty and complexity, quantum coding technology is used to map the data into quantum state space, and the characteristics of quantum superposition and entanglement are used to store conventional state-action data and reward information in classical storage units. In the data sampling process, the quantum Monte Carlo method and the classical random sampling method are combined. For the data in the quantum storage unit, the quantum Monte Carlo method is used to sample. By measuring and statistics the quantum state, representative data samples are obtained. For the data in the classical storage unit, the classical random sampling method is used for sampling.

4. The integrated electro-hydraulic control system for take-off and landing of a UAV according to claim 1, characterized in that: The optimization process of the adaptive dynamic programming algorithm in the deep learning model module introduces a multi-agent collaborative optimization strategy that regards the deep learning model as a collection of multiple agents. Each agent is responsible for processing a part of the state space and action space. During the training process, these agents jointly optimize the model parameters through information interaction and collaboration. Specifically, each agent calculates the local optimal strategy and value function based on its own local observations and experience, and exchanges their respective strategy and value function information through the communication network between agents. During the information interaction process, a fusion algorithm with a consensus mechanism is adopted. Each agent updates its own strategy and value function based on the information received from other agents, so that the strategies and value functions of all agents gradually converge to the same. The algorithm formula is: Among them, θ i (t) represents the parameter vector of the i-th agent at time t, N is the total number of agents, a ij (t) is the communication weight between agents i and j, satisfying 5. The integrated electro-hydraulic control system for take-off and landing of a UAV according to claim 1, characterized in that: When generating the takeoff control strategy, the control strategy generation module considers the subdivision and refined control of the drone's takeoff phase. The takeoff process is divided into three stages: initial acceleration, attitude adjustment, and stable ascent. During the initial acceleration phase, the deep learning model's prediction of the current environment, combined with the drone's payload and battery level information, accurately calculates the required initial driving pressure and speed. At high altitudes, due to the thin air, greater driving pressure is required to provide sufficient lift. In high-temperature environments, battery performance may degrade, requiring adjustment of driving parameters to ensure proper motor operation. During the attitude adjustment phase, the UAV's attitude changes are monitored in real time using data from the inertial measurement unit and visual sensors. When attitude deviation is detected, the drive control strategy generation module adjusts the various actuators of the electro-hydraulic drive system based on the size and direction of the deviation. During the stable ascent phase, the driving frequency and power are dynamically adjusted according to the current wind speed, wind direction and atmospheric turbulence. At the same time, the interaction between the UAV and the surrounding environment and the interference of airflow on the UAV are taken into account. By optimizing the driving control strategy, the UAV can maintain a stable ascent trajectory in complex environments and avoid altitude fluctuations and attitude loss of control caused by airflow disturbances.

6. The integrated electro-hydraulic control system for take-off and landing of a UAV according to claim 1, characterized in that: The control strategy generation module introduces terrain perception and obstacle avoidance mechanisms when generating landing control strategies. The drone is equipped with a variety of terrain perception and obstacle avoidance equipment such as lidar, millimeter-wave radar, and visual sensors. During landing, the lidar scans the terrain below the drone in real time to build a three-dimensional topographic map. The millimeter-wave radar is used to detect surrounding obstacles, especially in low-visibility environments. The visual sensor uses image recognition algorithms to identify target areas and obstacles on the ground. The control strategy generation module uses the information collected by these sensors and the prediction results of the deep learning model to generate a safe landing path and control strategy. When an obstacle is detected in the landing area, the control strategy generation module adjusts the drone's landing direction and altitude to bypass the obstacle. When approaching the ground, the control strategy generation module dynamically adjusts the drone's descent rate and attitude based on the undulations of the terrain to ensure a smooth landing. At the same time, in order to improve the accuracy and safety of landing, a method based on model predictive control (MPC) is adopted. The MPC algorithm optimizes the control input in the future based on the current state and future prediction information, so that the UAV can land according to the predetermined trajectory.

7. The integrated electro-hydraulic control system for take-off and landing of a UAV according to claim 1, characterized in that: The adaptive sliding mode control algorithm parameter adjustment mechanism in the electro-hydraulic drive execution module introduces a parameter adaptive adjustment method based on fuzzy logic. The core of sliding mode control is to design a sliding surface so that the system state can run stably on the sliding surface. In this system, the parameters of the sliding surface are adaptively adjusted according to the state and environmental information of the UAV. The fuzzy logic controller adjusts the parameters of the sliding mode control through a set of fuzzy rules based on the current system error and error change rate input variables. The design of fuzzy rules is based on in-depth analysis and experience summary of the UAV take-off and landing process. When the system error is large, the sliding mode gain is increased to speed up the system response speed. When the error change rate is small, the sliding mode gain is reduced to reduce the system chattering. The form of the fuzzy rules is as follows: Among them, R k represents the kth fuzzy rule, e is the system error, is the error rate of change, and is a fuzzy set, Δp i is the sliding mode control parameter that needs to be adjusted, is the corresponding fuzzy output set.

8. The integrated electro-hydraulic control system for take-off and landing of a UAV according to claim 1, characterized in that: The system has a multi-level fault diagnosis and fault-tolerant control architecture, covering four levels: sensors, deep learning models, drive control strategy generation modules, and electro-hydraulic drive execution modules; Sensor fault diagnosis uses a data-driven approach, building a normal operating model based on historical data. When the data deviates from the model, statistical analysis and machine learning algorithms are used to determine the fault type and location. For example, a Gaussian mixture model is used for wind speed sensors, and an ARMA model is used for temperature sensors. Deep learning model fault diagnosis is achieved by monitoring training and prediction results. When the prediction error exceeds the threshold, the parameters and structure are analyzed to determine whether it is overfitting, underfitting, or parameter abnormality. In the case of overfitting, the regularization strength or data augmentation is adjusted, and in the case of underfitting, the model complexity is increased. The drive control strategy generation module fault diagnosis is based on the rationality and feasibility of the strategy, and the strategy is simulated and verified. If problems such as unstable posture and abnormal energy consumption occur, the prediction results, optimization algorithm and input data are analyzed to find the root cause of the fault. If necessary, the optimization strategy is adjusted or the data is repaired. Fault diagnosis of the electro-hydraulic drive execution module is performed by monitoring the electro-hydraulic system parameters and motor status. When the pressure, flow, current, and voltage parameters are abnormal, the characteristics are analyzed to determine the fault type. For example, the pressure and flow curves are used to analyze electro-hydraulic servo valve faults, and the electromagnetic and mechanical characteristics are used to analyze motor faults. In terms of fault-tolerant control, when a sensor fails, a backup sensor is switched or other sensor data is integrated. Deep learning model failures are adjusted according to type. When the drive control strategy generation module fails, a backup algorithm or a safety strategy is used. When the electro-hydraulic drive execution module fails, corresponding measures are taken according to the severity. When multiple faults occur simultaneously, they are handled according to priority, and strategies are adjusted in real time. In addition, the system has a fault recording and reporting function, which records the fault time, location, and type information, and generates a report containing a fault overview, cause analysis, treatment measures, and performance impact assessment to assist in troubleshooting and system optimization.

Images