Unmanned aerial vehicle multi-system navigation decoy system based on intelligent algorithm optimization

Abstract

The invention provides an unmanned aerial vehicle multi-system navigation decoy system based on intelligent algorithm optimization, and relates to the technical field of unmanned aerial vehicle processing, a sensing module comprises a cognitive radar, a photoelectric / infrared sensor, a magnetic and acoustic sensor and an electromagnetic environment scanning device, and a digital twinning and countermeasure AI module comprises a digital twinning model module and a countermeasure AI engine; the cross-domain multi-source cheating module comprises a multi-system navigation signal synthesis module and a sensor-level cheating module; the reconfigurable intelligent antenna array comprises a metamaterial unit, a real-time beam controller and a biological heuristic intelligent algorithm module, and the system security hiding module comprises an encryption access control unit, a micro-power frequency expansion and hopping unit and a redundant backup and disaster recovery unit. The response time of the system from detection to interference can be remarkably shortened by performing high-fidelity simulation on the flight control logic of the unmanned aerial vehicle and the actual electromagnetic environment in the digital twin environment and combining the hybrid decoy strategy generated by the adversarial AI.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle processing, and in particular to an unmanned aerial vehicle multi-system navigation deception system based on intelligent algorithm optimization. Background Art

[0002] According to a multi-mode satellite navigation decoy anti-UAV system disclosed in Chinese patent publication number "CN109597100A", the system includes a control computer, a simulation computer, an intermediate frequency signal processor and a radio frequency module. The beneficial effects of the present invention are that a theoretical analysis is performed on the propagation environment, radio frequency interference, environmental noise, ionospheric error and tropospheric error of satellite signals, thereby realizing a true simulation of multi-mode satellite signals; time and space unification, direct digital frequency synthesis and multi-carrier signal synthesis technology are adopted to realize the synthesis of multi-mode satellite signals, realize navigation messages, i.e., observation data and other information, and thereby generate satellite navigation simulation signals of any specified position and speed; the present invention drives away, defends or forces illegal UAVs to land by transmitting low-power decoy signals.

[0003] The above patent documents and prior art have the following technical problems when used:

[0004] In addition to multi-mode satellite navigation, modern drones often use inertial measurement units (IMUs), vision / optical flow, laser ranging, magnetometers and other sensing methods for positioning. If the deception is limited to the satellite navigation channel, once the drone integrates multi-sensor data, the deception effect will be greatly reduced; Summary of the invention

[0005] Technical issues solved

[0006] In view of the deficiencies of the prior art, the present invention provides a UAV multi-system navigation deception system based on intelligent algorithm optimization to solve the problems of the prior art.

[0007] Technical Solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a multi-system navigation deception system for unmanned aerial vehicles based on intelligent algorithm optimization, the system comprising:

[0009] The perception module is used to detect and identify the UAVs in the target airspace through multi-source information. The perception module includes:

[0010] A cognitive radar is used to obtain the distance, speed and direction information of the target UAV; the cognitive radar is distributed in multiple base stations;

[0011] Photoelectric / infrared sensors are used to capture the shape, posture and visual features of the drone;

[0012] Magnetic and acoustic sensors to detect magnetic field fluctuations or noise generated by the operation of drone motors;

[0013] Electromagnetic environment scanning device, used to collect the distribution of local satellite navigation frequency bands, ground cellular networks and other radio frequency signals in real time;

[0014] Digital twin and adversarial AI module, used for model simulation based on UAV flight control and navigation fusion logic, adversarial simulation and optimal deception strategy generation. Digital twin and adversarial AI module includes:

[0015] Digital twin model module, used to build a simulation environment for the dynamics and flight control algorithms of the target UAV;

[0016] Adversarial AI engine, used to search and output multi-dimensional deception parameters in digital twin environments through adversarial attacks, reinforcement learning, or generative adversarial networks;

[0017] The cross-domain multi-source decoy module is used to generate and transmit composite decoy signals targeting drones. The cross-domain multi-source decoy module includes:

[0018] Multi-system navigation signal synthesis module, used to synthesize and synchronize the standard false navigation signals of GPS, GLONASS, Beidou, and Galileo at the baseband level, and combine the ground cellular network and satellite-based broadband communication signals;

[0019] The sensor-level deception module is used to simulate or inject the visual, geomagnetic, and air pressure sensor data of the drone to build multi-level hybrid deception;

[0020] The communication link jammer module is used to tamper with, delay, forge and interfere with the communication data between the drone and its ground station or satellite link;

[0021] Reconfigurable smart antenna array for flexible control of decoy signal pattern, polarization and phase, including:

[0022] Metamaterial units for secondary reflection, beamforming or phase modulation of electromagnetic waves;

[0023] A real-time beam controller that adjusts the operating mode of the antenna array or RIS panel based on the tracking data from the perception module and the output of the digital twin and adversarial AI modules;

[0024] A biologically inspired intelligent algorithm module, which is used to perform parallel search and optimization in high-dimensional, multi-objective deception parameter space, and assist in completing real-time adaptive deception strategies;

[0025] The system security concealment module is used to manage the command issuance, data exchange and concealed transmission of the entire system, including:

[0026] Encrypted access control unit, used to authenticate and encrypt the interactions between each submodule of the system to prevent unauthorized access;

[0027] Micro-power spread frequency hopping unit to reduce the risk of decoy signals being detected or located;

[0028] Redundant backup and disaster recovery units are used to maintain the core functions of the system when key nodes fail or are attacked.

[0029] Preferably, the digital twin model module in the digital twin and adversarial AI module further includes:

[0030] Environmental modeling unit, used to simulate the terrain, buildings, electromagnetic reflection and meteorological data of the actual deployment area;

[0031] The flight control logic and sensor fusion modeling unit is used to simulate the calculation process of UAV inertial navigation, satellite navigation, vision, air pressure and geomagnetic fusion.

[0032] Preferably, the adversarial AI engine utilizes a generative adversarial network and an adversarial attack algorithm to generate adversarial samples for the deep learning and data fusion models of the target UAV, which are used to introduce confusion interference into the visual and inertial sensing channels.

[0033] Preferably, the cross-domain multi-source deception module includes a geomagnetic field anomaly simulation module, which is composed of a number of controllable electromagnetic coil arrays and is used to apply local magnetic field distortion within the target airspace, causing the geomagnetic navigation of the drone to drift significantly.

[0034] Preferably, the reconfigurable smart antenna array uses a metamaterial unit and has an electrically tunable phase-shifting array, which can achieve millimeter-level to centimeter-level phase synchronization accuracy under the guidance of cognitive radar return data.

[0035] Preferably, the bio-inspired intelligent algorithm module adopts neuromorphic computing, and has global search and fast convergence effects when processing navigation signal power allocation, phase control, and communication interference strategy.

[0036] Preferably, the system also includes a hardware acceleration and disaster recovery platform, which includes GPU, FPGA, and ASIC nodes. The hardware acceleration and disaster recovery platform accelerates large-scale real-time simulation and AI training, and deploys hot backup servers in the cloud to achieve system recovery when a system failure occurs.

[0037] Preferably, the system comprises the following steps:

[0038] Sp1, detection and identification step: monitor the target airspace through the perception module to obtain the type, navigation mode, posture and position of the target UAV;

[0039] Sp2, digital twin modeling and adversarial deduction steps: Based on the target drone type and environmental information, the digital twin model is loaded, and the adversarial AI engine deduces the optimal hybrid deception strategy in a simulated environment;

[0040] Sp3, decoy signal generation step: the cross-domain multi-source decoy module synthesizes and transmits multi-standard false navigation signals according to the deception parameters output by the adversarial AI, and simulates or forges sensor channels such as vision, geomagnetism or air pressure;

[0041] Sp4, beam steering and power control step: Through the reconfigurable smart antenna array or RIS, the decoy signal is injected into the receiving end of the target UAV with a specific pattern and power, and the beam shape and phase are dynamically adjusted according to the real-time tracking information;

[0042] Sp5, closed-loop monitoring and strategy iteration step: After the target drone begins to deviate or fly abnormally, continue to monitor its trajectory and attitude changes, compare the system's expected results, and use quantum or biological heuristic algorithms to update deception parameters within milliseconds or seconds;

[0043] Sp6. Complete the intervention steps: When the target drone is induced to a safe area, forced to land or disabled, the system enters the standby or withdrawal mode, and stores and analyzes the process data for subsequent strategy optimization.

[0044] Preferably, the adversarial AI engine first identifies the importance of the inertial navigation characteristics and visual modules of the target UAV, and preferentially generates adversarial attack methods for the key perception channel. The reconfigurable intelligent antenna array uses the airspace situation information transmitted in real time by the cognitive radar during beamforming to dynamically avoid interference with non-target areas or legitimate aircraft. During the strategy iteration process, if it is detected that the target UAV triggers an autonomous fail-safe mode or an escape maneuver, it immediately switches to a preset emergency interference plan to continue to maintain the controlled or harmless state of the target UAV. After the Sp6 completes the intervention, the system automatically calls the security and concealment module to securely erase the deceptive signal transmission records, and saves the key operation logs on the intranet or blockchain for subsequent auditing and responsibility tracing.

[0045] Beneficial Effects

[0046] The present invention provides a multi-system navigation deception system for unmanned aerial vehicles based on intelligent algorithm optimization. It has the following beneficial effects:

[0047] 1. The present invention can significantly shorten the system's reaction time from detection to interference by performing high-fidelity simulation of the UAV flight control logic and the actual electromagnetic environment in a digital twin environment, combined with a hybrid deception strategy generated by adversarial AI, and achieve rapid switching of multi-source signal collaborative deception and interference schemes; at the same time, the dynamic beamforming of cognitive radar and reconfigurable antenna array can effectively focus on the area where the target UAV is located, making the interference energy utilization rate higher and the overall efficiency of interference and detection significantly improved.

[0048] With the help of bio-inspired intelligent algorithms and the iterative optimization capabilities of adversarial AI, the system can automatically evaluate target weaknesses and flexibly adjust interference schemes when facing different types of drones, different navigation methods or complex electromagnetic environments. When the target attempts to escape or applies multi-sensor anti-interference strategies, the system can quickly switch to a higher intensity or more targeted emergency interference mode through closed-loop monitoring and parameter updates with millimeter or millisecond cycles, ensuring continuous control of the target and reducing the risk of being avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a system workflow diagram of the present invention;

[0050] Figure 2 This is a structural diagram of the multi-system navigation deception system for unmanned aerial vehicles of the present invention;

[0051] Figure 3 The echo signal and the time limit diagram of the present invention;

[0052] Figure 4 This is a diagram of the echo signal power and detection results of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Specific embodiment one:

[0055] like Figure 1-4 As shown, a multi-system navigation deception system for unmanned aerial vehicles based on intelligent algorithm optimization includes a hardware acceleration and disaster recovery platform, which includes GPU, FPGA, and ASIC nodes. The hardware acceleration and disaster recovery platform accelerates large-scale real-time simulation and AI training, and deploys hot backup servers in the cloud to achieve system recovery when a system failure occurs. The system modules include:

[0056] The perception module is used to detect and identify UAVs in the target airspace through multiple sources. The perception module includes: cognitive radar, which is used to obtain the distance, speed and direction information of the target UAV; cognitive radar is distributed in multiple base stations; photoelectric / infrared sensors are used to capture the shape, posture and visual characteristics of the UAV; magnetic and acoustic sensors are used to detect magnetic field fluctuations or noise generated by the operation of the UAV motor; electromagnetic environment scanning device is used to collect the distribution of local satellite navigation frequency bands, ground cellular networks and other radio frequency signals in real time;

[0057] Digital twin and adversarial AI modules are used to simulate models based on the flight control and navigation fusion logic of drones, conduct adversarial simulations and generate optimal deception strategies. The digital twin and adversarial AI modules include: digital twin model modules, which are used to build a simulation environment for the dynamics and flight control algorithms of target drones; adversarial AI engines, which are used to search and output multi-dimensional deception parameters in a digital twin environment through adversarial attacks, reinforcement learning or generative adversarial networks;

[0058] The cross-domain multi-source deception module is used to generate and transmit composite deception signals for drones. The cross-domain multi-source deception module includes: a multi-system navigation signal synthesis module, which is used to synthesize and synchronize the standard false navigation signals of GPS, GLONASS, Beidou, and Galileo at the baseband level, and combine the ground cellular network and satellite-based broadband communication signals; a sensor-level deception module, which is used to simulate or inject the visual, geomagnetic, air pressure and other sensor data of the drone to build a multi-level hybrid deception; a communication link interference module, which is used to tamper with, delay, forge and interfere with the communication data between the drone and its ground station or satellite link;

[0059] Reconfigurable smart antenna array for flexible control of the directional pattern, polarization and phase of the decoy signal, including: metamaterial unit for secondary reflection, beamforming or phase modulation of electromagnetic waves; real-time beam controller for adjusting the working mode of the antenna array or RIS panel according to the tracking data of the perception module and the output of the digital twin and adversarial AI module; biologically inspired intelligent algorithm module for parallel search and optimization in the high-dimensional, multi-target decoy parameter space to assist in the completion of real-time adaptive deception strategy. The biologically inspired intelligent algorithm module adopts neuromorphic computing and has global search and fast convergence effects when processing navigation signal power allocation, phase control, and communication interference strategy;

[0060] The bio-inspired algorithm of the bio-inspired intelligent algorithm module has good global search and fast convergence capabilities in complex optimization problems with high dimensions, multiple constraints, and multiple objectives by simulating the discharge mechanism of biological neurons, group cooperative behavior, or adaptive evolutionary processes. In this system, the algorithm needs to conduct a comprehensive search and optimization of key parameters such as power, phase, frequency, and modulation mode of the navigation decoy signal in a very short time according to the real-time measured state of the target drone, the electromagnetic environment, and the confrontation intention, and work in coordination with other interference methods (such as visual deception, magnetic field distortion, communication link tampering, etc.) to achieve a continuous and accurate decoy effect. Neuromorphic computing can be understood as a spiking neural network (Spiking Neural Network, denoted as SNN) or a computing framework that simulates the discharge mechanism of brain neurons. The core idea is to use pulse sequences to represent information and use synaptic plasticity to achieve adaptive learning. Traditional deep networks are mainly based on real number weighted summation, while spiking neural networks emphasize information transmission in the time domain, which is very efficient for processing large-scale parallel, event-driven data. Adopting the LIF (Leaky Integrate-and-Fire) model:

[0061] In the LIF model, the membrane potential V(t) of each neuron is accumulated according to the external input current or synaptic input and decays over time. th When , the neuron generates a pulse discharge and resets the membrane potential. Its mathematical form is as follows:

[0062] V(t) represents the neuron membrane potential; C represents the neuron membrane capacitance; g L Indicates membrane conductivity; V rest represents the resting potential of the neuron;

[0063] SNN for deceiving parameter optimization I syn (t) represents the input current from the synapse, and its magnitude is related to the pulse firing of other neurons and the synaptic weight. When V(t)≥V th When , the neuron immediately fires a pulse and resets V(t) to V reset In the specific hardware implementation, dedicated neuromorphic chips or accelerated simulation on GPU / FPGA are often used to achieve the purpose of real-time response and parallel optimization. Each optimization variable is regarded as the plasticity weight of several neurons or a group of synapses, so that the optimization process becomes a process in which the network state gradually evolves and adaptively updates with the input (actual state of the target drone, expected decoy result). The decoy signal power vector is recorded as P = (P1, P2, ..., P n ), each P i Corresponding to a variable synaptic weight or the firing rate of a neuron group. The phase vector of the decoy signal is denoted as φ = (φ1, φ2, ..., φ n), used to describe the phase control amount of each carrier or beam channel. A set of objective functions can be defined (explained later), and the network performs a global search through pulse iterations and outputs the corresponding decoy parameter configuration when the optimal or suboptimal solution is reached.

[0064] Multi-objective optimization model and objective function, decoy parameter optimization often needs to consider the following multiple objectives:

[0065] Maximize the effectiveness of deception: the greater the deviation of the drone's position, the confusion of its posture, or the failure of the opponent's anti-interference strategy, the better; minimize the system energy consumption: the transmission power and antenna resources should be as small as possible; concealment: avoid exposing the interference source of the party; real-time: obtain a usable solution in a very short time (such as milliseconds), summarize it into a multi-objective optimization problem, and synthesize it into a weighted or hierarchical objective function. The exemplary weighted form is as follows:

[0066] E deviate (P, φ) represents the quantitative index of the yaw / deviation or attitude confusion of the UAV after being disturbed; E power (P) represents the transmit power consumption or total energy usage of the decoy system, E exposure (P, φ) represents the probability or exposure of being discovered by the enemy's radar or detection system; w1, w2, and w3 are the weight coefficients of the corresponding targets, which can be set according to actual combat needs. In the implementation of the neuromorphic algorithm, the spiking neural network can adaptively adjust the synaptic weights and membrane potential based on error feedback (the difference between the observed drone trajectory deviation and the expected target), and iteratively converge to a set of better P, φ solutions.

[0067] Workflow of the biologically inspired algorithm:

[0068] 1. Initialization: Initialize the neuron membrane potential, synaptic weight, etc. of the spiking neural network or neuromorphic computing unit according to the current electromagnetic environment, drone status and preset weights w1, w2, w3, etc.;

[0069] 2. Input encoding: Encode external observation values ​​(drone position, speed, electromagnetic environment information, etc.) into pulse input.

[0070] 3. Pulse iteration: The network continuously updates the neuron membrane potential and calculates the pulse output in the time domain; if the decoy configuration corresponding to the output needs to be verified, the actual interference effect (feedback from the perception module or digital twin module) is required to form a closed loop.

[0071] 4. Error evaluation and weight update: Calculate the quality of the current solution according to the objective function, and update the network parameters under the synaptic plasticity mechanism to make the subsequent pulse state more inclined to a high fitness solution.

[0072] 5. Output the optimal configuration: After meeting the real-time or iteration constraints, output the current optimal deception parameters to the cross-domain multi-source deception module.

[0073] The system security concealment module is used to manage the command issuance, data exchange and concealed transmission of the entire system; the encrypted access control unit is used to authenticate and encrypt the interaction of each submodule of the system to prevent unauthorized access; the micro-power spread frequency hopping unit is used to reduce the risk of decoy signals being detected or located; the redundant backup and disaster recovery unit is used to maintain the core functions of the system when key nodes fail or are attacked. The digital twin model module in the digital twin and adversarial AI module further includes: an environmental modeling unit, which is used to simulate the terrain, buildings, electromagnetic reflection and meteorological data of the actual deployment area; a flight control logic and sensor fusion modeling unit, which is used to simulate the calculation process of drone inertial navigation, satellite navigation, vision and air pressure, and geomagnetic fusion; the digital twin model is a high-fidelity dynamic simulation system for the target drone and the flight environment, which can accurately reproduce the drone's flight dynamics, sensor characteristics and external environmental elements in the virtual space. Through the digital twin model, different decoy or interference strategies can be quickly deduced and evaluated, helping the system to significantly reduce the trial and error cost when it is actually deployed. For multi-rotor drones or fixed-wing drones, a six-degree-of-freedom motion equation can be established to describe the evolution of the drone's position, velocity, and attitude in three-dimensional space. The six-degree-of-freedom model is used for illustration:

[0074] 1. State vector:

[0075] Among them, p x ,p y ,p z represents the three-dimensional position of the UAV in the inertial coordinate system; v x ,v y ,v z represents the three-dimensional velocity component; α, β, γ represent the yaw angle, pitch angle, and roll angle respectively; ω x ,ω y ,ω z represents the angular velocity component.

[0076] 2. Control input:

[0077] Where T represents the total thrust or engine thrust, δ α ,δ β ,δ γ Indicates the attitude control rudder or motor differential distribution (for multi-rotor scenarios, it can also be represented by the individual thrust of four or more rotors).

[0078] 3. Equations of motion:

[0079] By introducing the Newton-Euler equation into the six-degree-of-freedom framework, the dynamic equation of the drone can be abstracted as x

[0080] Among them, w(t) represents process noise, which comes from external uncertainties such as wind field disturbances and air density changes. Function f(·) reflects mechanical relationships such as aerodynamics, gravity, thrust, and torque balance. 4. Output or observation equation:

[0081] Among them, y(t) represents the simulated measurement value in the digital twin environment, such as position, velocity, attitude, etc., and v(t) represents the measurement noise. Through the above dynamic equations, the digital twin model can simulate the evolution trajectory of the UAV over time given the control input (including human control instructions or autonomous flight control instructions) and environmental parameters, providing a high-fidelity virtual test platform for adversarial AI and biological heuristic algorithms.

[0082] In addition to the UAV's motion equations, the digital twin model also includes a description of the external environment, including but not limited to:

[0083] Terrain and obstacles: Use discrete grids or vector maps to model the terrain and buildings in the target area to simulate scenarios such as multipath reflection and occlusion.

[0084] Electromagnetic environment: Divide the RF frequency band, define the background noise energy, the distribution of existing communication base stations, satellite navigation ephemeris data, etc., so as to evaluate the propagation effect of decoy or jamming signals in the simulation.

[0085] Weather and meteorology: including wind speed, wind direction, temperature, humidity, etc., used to calculate aerodynamics and sensor errors. If high accuracy is required, turbulence models or local strong winds can also be considered.

[0086] Magnetic field distribution: Regional interpolation of the geomagnetic environment or calculation based on the international geomagnetic magnetic field model to provide support for simulating drone magnetometer readings, geomagnetic anomaly simulation, etc.

[0087] The adversarial AI engine uses generative adversarial networks and adversarial attack algorithms to generate adversarial samples for the target drone’s deep learning and data fusion models, which are used to introduce confusion interference into the visual and inertial sensor channels.

[0088] The data fusion model is a unified estimation and correction of multi-source perception information (including inertial measurement values, satellite navigation signals, visual detection results, barometers, magnetometers, etc.) to obtain the optimal state estimation of the UAV. This model is used to simulate the internal perception and decision-making logic of the UAV in the digital twin environment, and is also used to understand the behavior and anti-interference ability of the target UAV when the system is actually deployed. Multi-sensor fusion is usually based on a unified state vector to jointly estimate the measurement information given by each sensor. Using extended Kalman filtering:

[0089] State vector:

[0090] Here, the nine-dimensional state is taken, including position, velocity and attitude angle. In actual engineering, parameters such as gyroscope bias and wind field estimation can be added as needed.

[0091] 2. State prediction equation:

[0092] in, Represents the predicted state, is the prediction covariance matrix, F k -1 is for the state equation in The Jacobian matrix obtained by linearization at k-1 is the process noise covariance matrix.

[0093] 3. Measurement update equation: Define the measurement equation for each sensor (such as inertial measurement, satellite positioning, visual detection, etc.):

[0094] Among them, z k represents the sensor output, ν k is the measurement noise. The EKF performs the following in the update step:

[0095] Among them, H k is the Jacobian of the measurement equation, R k is the measurement noise covariance matrix, K k By periodically or asynchronously updating the measurements of different sensors for the Kalman gain, the fusion model can obtain a more accurate and robust estimate of the UAV state, thereby simulating the navigation accuracy and anti-interference level of the real flight control system in the digital twin environment.

[0096] Visual fusion: If the drone is equipped with a visual positioning system based on image features, the coordinates or relative displacements extracted by vision can be regarded as measurement signals and integrated into the EKF update step. The measurement function will contain a projection model determined by the attitude and the internal and external parameters of the camera, and the error reflects the randomness of visual feature matching and optical flow calculation.

[0097] Geomagnetic fusion: Magnetometers provide heading information but are susceptible to magnetic field distortion or interference. Fusion converts magnetometer readings into dead reckoning measurements, and estimates the ideal reading based on the magnetic field model in the digital twin environment, thereby calculating measurement residuals and updating state estimates.

[0098] Collaborative operation of digital twin model and data fusion model:

[0099] Offline stage: The drone dynamics, environmental factors and multi-sensor data fusion logic are fully loaded into the digital twin model; according to the needs of adversarial AI and neuromorphic algorithms, batch simulations are run to compare the impact of various interference strategies on the fusion model, so as to obtain vulnerability assessments and preliminary optimal interference combinations.

[0100] Online stage: When the system detects a real drone attack, the data fusion model is executed in the target drone body (or its execution is inferred), and the digital twin model is simulated synchronously in the local system; the adversarial AI module and the bio-inspired algorithm module compare the digital twin output with the real feedback detected by the sensor to generate real-time deception strategies or interference parameter updates; if a large deviation is observed between the actual flight state of the target drone and the twin simulation result, the digital twin parameters (such as environmental wind field, model characteristics) can be corrected in time to ensure that the simulation and reality maintain a high degree of consistency.

[0101] Digital Twin Model: Provides a high-fidelity simulation platform for drone dynamics, sensor-environment interaction; works with adversarial AI and bio-inspired algorithms, combining offline and online to quickly evaluate the feasibility and effectiveness of multiple interference strategies.

[0102] Data fusion model: simulate or reconstruct the target UAV’s internal multi-sensor fusion mechanism, including inertial navigation, satellite navigation, vision and geomagnetic sensing, etc.; help identify the opponent’s navigation anti-interference ideas, and provide a reference for the design and real-time tuning of decoy signals.

[0103] It is the perfection and coordination of these two parts that enables the system to maintain high precision, high concealment and rapid self-adaptation in multi-channel, comprehensive interference and deception against drones. Through digital twins and data fusion, the system builds a bridge between the virtual environment and the actual environment, providing reliable simulation data support and comparative evaluation mechanism for the decision-making process of adversarial AI and neuromorphic algorithms, thereby achieving stable and accurate navigation deception effects on target drones.

[0104] The adversarial AI engine uses the interference of the generative adversarial network in the visual / inertial channel to simulate the working process of drone flight control and sensor fusion (vision, inertial navigation, barometer, magnetometer, etc.) in the digital twin model environment. Through GAN (generative adversarial network) and adversarial attack algorithm, it generates perturbation samples for the deep learning model or data fusion module of the target drone, causing its decision to deviate from the normal trajectory or judgment:

[0105] GAN consists of a generator G and a discriminator D. The generator aims to generate "fake" data from random noise or initial samples, and the discriminator is responsible for distinguishing whether the input is "real data" or "fake data". The interactive optimization process of the two can be formalized as the following Minimax problem:

[0106] p data (x) represents the probability distribution of real data (such as visual images or inertial measurements normally received by the drone); p z (z) represents the distribution of noise input received by the generator; G(z) generates false images or false inertial data; D(·) is the discriminator's scoring of the authenticity of the input. In this system, "real data" is equivalent to the effective field quantity or sensor value read from the vision / IMU by the drone without interference, and "pseudo-data" is the adversarial sample synthesized by the system for deception. Through continuous adversarial training, the generator can learn how to generate "fake" visual or inertial signals, so as to inject them into the perception / fusion module of the drone during actual deception.

[0107] Adversarial attack algorithm: Given a target model (such as drone visual recognition or attitude estimation algorithm) f(·), and input data x, the adversarial perturbation is denoted as δ, and the goal is to find a minimum perturbation that makes the model decision result wrong. The formula is:

[0108] In more complex scenarios, specific categories or postures can also be set to prevent the drone from misidentifying, which can be written as

[0109] Where c is the expected error output (e.g., assuming that the drone believes it is at a certain height or direction). In the digital twin environment, the disturbance is first searched offline and fine-tuned online, and finally injected into the drone sensor data or communication link in a targeted manner in actual combat.

[0110] The collaboration between adversarial AI and bio-inspired algorithms mainly generates attack strategies for the drone's own control algorithm or sensor data fusion, and outputs interference samples. The bio-inspired intelligent algorithm mainly searches for the optimal launch / interference configuration in real time in the time-space-frequency multiple dimensions of the decoy signal.

[0111] The two work together in the closed loop of digital twins and the actual environment. Adversarial AI provides key vulnerability assessments or interference priorities for bio-inspired algorithms; and bio-inspired algorithms can ensure overall interference efficiency and concealment at the system level.

[0112] The cross-domain multi-source deception module includes a geomagnetic field anomaly simulation module, which is composed of several controllable electromagnetic coil arrays and is used to apply local magnetic field distortion within the target airspace, causing the geomagnetic navigation of the drone to drift significantly. The reconfigurable smart antenna array uses metamaterial units and has an electrically tunable phase-shifting array, which achieves millimeter-level to centimeter-level phase synchronization accuracy under the guidance of cognitive radar return data.

[0113] Perception module hardware:

[0114] Multi-base station cognitive radar stations: distributed in different locations to scan the airspace from multiple angles;

[0115] Photoelectric / infrared sensors: high-resolution cameras, infrared imaging devices, and matching video acquisition cards;

[0116] Magnetic / acoustic sensor array: deployed on a fixed or movable platform with a signal processing unit;

[0117] Electromagnetic environment scanning device: multi-channel RF front end and real-time spectrum analysis software.

[0118] Digital Twin and Adversarial AI Module Hardware / Software:

[0119] Servers or industrial computer clusters: Simulation software and adversarial AI engines that carry digital twin models;

[0120] Computing acceleration nodes: GPU, FPGA or ASIC, used to support large-scale parallel simulation and AI training / inference;

[0121] Dedicated data bus or high-speed network: connects sensor data to the simulation center.

[0122] Cross-domain multi-source deception module hardware / software:

[0123] Multi-system navigation signal synthesis unit: software defined radio (SDR) platform, high-precision local oscillator, and RF power amplifier;

[0124] Sensor-level deception simulators: devices that can forge visual images, geomagnetic signals, or air pressure data;

[0125] Communication link jammers: low-power or medium-power jammers with directional transmit antennas or reconfigurable antennas.

[0126] Reconfigurable Smart Antenna Array or RIS Hardware / Software:

[0127] Metamaterial unit or RIS panel: several electrically tunable phase shifters, phase shift control chips;

[0128] Real-time beam controller: Based on an embedded processor or FPGA, it executes the beamforming algorithm.

[0129] Biologically inspired intelligent algorithm module software:

[0130] Neuromorphic computing platforms or GPU-accelerated spiking neural network programs for high-dimensional, multi-target decoy parameter searches;

[0131] It can share some hardware acceleration resources with the adversarial AI engine.

[0132] System security concealment module hardware / software:

[0133] Encrypted access control unit: with a secure chip (SecureElement) and a built-in encryption library for verifying instruction signatures;

[0134] Micro-power frequency-hopping unit: RF front-end with adaptive spectrum hopping function;

[0135] Redundant backup and disaster recovery unit: primary / backup server, network storage (NAS) or blockchain node for log storage and failover. Specific embodiment 2:

[0137] like Figure 1-2 As shown, based on the technical solution of the specific embodiment 1, the system further includes the following operating steps:

[0138] Sp1: Detection and identification steps:

[0139] Airspace monitoring: Multi-base station cognitive radar covers the target airspace in a segmented scanning or holographic scanning mode; the system software automatically identifies the echo characteristics of small / micro drones based on machine learning algorithms.

[0140] The optoelectronic / infrared sensor cooperates with the radar echo to lock the target direction, achieve high-resolution video / infrared imaging, and identify the shape and model of the drone.

[0141] Magnetic / acoustic sensors enhance the detection capability of low radar cross-section drones by capturing the magnetic field disturbances or acoustic characteristics generated by the drone's power system.

[0142] Signal characteristics and navigation method identification: The electromagnetic environment scanning device monitors the satellite navigation frequency band (GPS / GLONASS / Beidou / Galileo) and ground communication link (which may be a cellular network or other frequency band) used by the drone, and identifies its communication / navigation protocol characteristics.

[0143] The back-end software integrates the above multi-source sensor data and outputs core information such as the type, navigation system, flight altitude, speed, and attitude of the target UAV.

[0144] Data management: The recognition results are stored in the database and sent to the digital twin and adversarial AI modules for subsequent simulation and strategy generation.

[0145] If multiple drones are detected, unique tracking IDs are assigned and stored separately.

[0146] Cognitive radar station, optoelectronic camera, sensor array front-end and high-speed data interface, back-end supporting multi-threaded / distributed data processing software. Target tracking algorithm (extended Kalman filter / multi-target tracking algorithm), feature recognition algorithm (infrared recognition based on deep learning).

[0147] Sp2: Digital twin modeling and adversarial deduction steps:

[0148] Loading the digital twin model: On the server / computer cluster, load the kinematic and dynamic characteristics of the corresponding UAV model through a software platform (e.g., a high-fidelity dynamics simulation tool).

[0149] At the same time, the local environment model (terrain, meteorological data, electromagnetic environment) and the initial state currently recognized by the drone (position, speed, attitude, etc.) are imported.

[0150] Adversarial AI deduction: The adversarial AI engine reads the flight control and data fusion logic in the digital twin model and tests possible interference methods (such as GPS false signals, geomagnetic distortion, communication link tampering, etc.) in a virtual environment:

[0151] Generate corresponding false signal data (GAN / adversarial perturbation) to simulate attacks on the drone perception module;

[0152] Record the deviation of the drone’s attitude or trajectory, the triggering of the anti-interference mechanism, etc.

[0153] Through multiple rounds of iterations, the engine gradually finds the "optimal" or "locally optimal" combination of hybrid deception strategies (including signal power, phase, time planning, etc.) in the simulation environment and outputs it for subsequent use.

[0154] Output the optimal hybrid deception strategy: The adversarial AI ultimately generates a set of multi-dimensional deception parameters (including navigation false signal characteristics, geomagnetic interference intensity, visual false marker configuration, etc.) and priority ranking.

[0155] It is synchronously transmitted to the bio-inspired intelligent algorithm module so that the latter can perform more detailed real-time optimization in the actual environment.

[0156] High-performance servers / cloud computing platforms, hardware accelerators (GPU / FPGA / ASIC) for large-scale simulation operations. Digital twin platforms (supporting dynamics and environmental simulation), adversarial AI algorithms (GAN, reinforcement learning or adversarial attacks), simulated drone flight control / sensor fusion logic.

[0157] Sp3: Decoy signal generation steps:

[0158] Cross-domain multi-source signal synthesis: The multi-standard navigation signal synthesis module synchronously generates false GPS / GLONASS / Beidou / Galileo signals through the software radio (SDR) platform according to the pseudo code sequence, frequency offset, transmission power and other parameters given by the adversarial AI;

[0159] If the drone is compatible with ground cellular or satellite-based broadband, fake base station positioning references or fake satellite-based communication data can be mixed in accordingly.

[0160] Sensor-level deception: Visual deception: Where possible, fake landmarks are created through LEDs, lasers, or projections, or fake visual image data is directly injected into the communication link (if it is accessible to the drone's internal algorithms);

[0161] Geomagnetic deception: deploying controllable electromagnetic coils to generate local magnetic field distortion;

[0162] Barometric pressure spoofing: Creating altitude reading drift by simulating or modifying barometer communication data (if the drone barometer input channel is obtained).

[0163] Communication link interference: The communication link interference module turns on the directional interference or tampering function of the corresponding frequency band, such as delaying or erroneously retransmitting the remote control data, making it impossible for the UAV ground station to correct the navigation in time;

[0164] Adversarial AI can instruct the jammer to insert false instructions at critical moments, further increasing the drone’s confusion.

[0165] SDR platform and multi-channel RF power amplifier, sensor-level jammer (LED / laser device, controllable electromagnetic coil), communication jammer (supporting directional or low-power point-to-point jamming). Multi-channel signal synthesis software, pseudo-code generation and phase control program, visual deception (GAN generation or optical projection) management program.

[0166] Sp4: Beam steering and power control steps:

[0167] Reconfigurable Smart Antenna Array (or RIS) operation: The real-time beam controller obtains the target UAV’s three-dimensional coordinates and motion trend information from the perception module;

[0168] According to the beam pointing / phase allocation scheme given by the bio-inspired algorithm, the phase shift and gain of each unit are dynamically tuned through the metamaterial unit or RIS panel to generate a high-gain beam pointing to the target UAV in the shortest time.

[0169] Power Adaptation: The system adjusts the transmit power in real time according to distance attenuation, environmental noise, and drone reception threshold, so that the interference signal is sufficient to cover the target drone while reducing leakage to other areas.

[0170] If the target drone moves quickly in the airspace, the beam controller will update the phased array parameters in milliseconds or shorter cycles to form a continuous tracking beam.

[0171] Avoid interference with irrelevant aircraft: The cognitive radar and perception module continuously scan the airspace. If other legitimate aircraft are detected, the system automatically isolates the beam airspace or attenuates the power to avoid side effects.

[0172] Precise "Null Steering" can be formed through RIS panels or metamaterial units, making the transmission power in a specific direction close to zero.

[0173] Reconfigurable antenna array or RIS panel, built-in electronically controlled phase shifter or tunable element; real-time controller (FPGA / high-end MCU can be used). Beamforming algorithm (phased array principle / optimal beamforming), airspace avoidance or airspace isolation algorithm.

[0174] Sp5: Closed-loop monitoring and strategy iteration steps:

[0175] Real-time monitoring of drone response: The perception module (radar, optoelectronics, magnetic / acoustic, etc.) continuously detects whether the target drone’s route, attitude, and speed are yawed or abnormal;

[0176] If the drone has some kind of autonomous failure mode or anti-interference strategy, observe its performance (entering hover mode, emergency return, etc.).

[0177] Quantum or bio-inspired algorithms are updated in real time: Once it is found that the drone is not deviating as expected or begins to try to avoid interference, the adversarial AI and bio-inspired algorithms will recalculate the optimal parameters of the decoy signal based on the new observations;

[0178] The neuromorphic computing platform (or quantum-inspired algorithm) iterates in cycles ranging from tens of milliseconds to seconds, searching for new combinations of transmission power, phase, and interference, and passes them to the decoy module and antenna array for adjustment.

[0179] Priority of attack on key perception channels: During the strategy iteration process, the adversarial AI engine will dynamically evaluate the most critical perception channels: inertial navigation channel, visual SLAM channel, geomagnetic channel, etc., based on the changes in the drone's track and internal fusion logic analysis;

[0180] If it is identified that the target UAV is highly dependent on a certain channel, it will prioritize maintaining the interference intensity of the channel to disable the UAV as quickly as possible or keep it under control.

[0181] Emergency jamming plan: If the target drone is detected to trigger an autonomous fail-safe mode (such as automatic return) or perform a high-acceleration escape maneuver, the system immediately switches to a preset emergency jamming mode (increasing the transmission power or expanding the jamming bandwidth) to keep the target under control or completely disable it.

[0182] The perception module continuously provides data information, the AI ​​engine cluster performs fast calculations, and the deception module and antenna array have the hardware capabilities to quickly execute new strategy commands. The reinforcement learning / adversarial attack algorithm is iterated in the loop, and the real-time data bus or low-latency network ensures high-frequency data exchange.

[0183] Sp6: Complete the intervention steps:

[0184] Guiding to a safe area / forced landing / disabling: When the target drone is guided to a predetermined safe area or is found to be out of power and unable to continue normal flight, the system determines that the interference purpose has been achieved;

[0185] The system can automatically reduce the interference power or stop interference to avoid unnecessary electromagnetic radiation.

[0186] Standby or withdrawal mode: After confirming that the target drone threat has been eliminated, the decoy system switches to standby mode, retaining only low-power monitoring;

[0187] If the scenario needs to be run repeatedly, the core module can be kept on standby at all times; if the event is over, the relevant equipment can be shut down or withdrawn.

[0188] Process data storage and auditing: The system calls the security and concealment module to encrypt and securely erase the signal transmission records of this interference to prevent sensitive information from being intercepted by outsiders;

[0189] Key operation logs are written into the intranet database or blockchain node for encrypted storage, including interference parameter change curves, target drone trajectory records, etc., for subsequent audits and responsibility tracing.

[0190] Hardware encryption chips, blockchain or intranet storage devices in security modules; automated script management to deceive power convergence, logging and auditing systems (with timestamps and digital signatures), and security policy processes for failure or emergency scenarios.

[0191] The adversarial AI engine first identifies the importance of the target UAV's inertial navigation characteristics and visual modules, and prioritizes the generation of adversarial attack methods for this key perception channel. The reconfigurable intelligent antenna array uses the airspace situation information transmitted in real time by the cognitive radar during beamforming to dynamically avoid interference with non-target areas or legitimate aircraft. During the strategy iteration process, if the target UAV is detected to trigger an autonomous fail-safe mode or escape maneuver, it will immediately switch to the preset emergency interference plan to continue to maintain the target UAV in a controlled or harmless state. After Sp6 completes the intervention, the system automatically calls the security and concealment module to securely erase the deceptive signal transmission records, and saves key operation logs on the intranet or blockchain for subsequent audits and responsibility tracing. Specific embodiment three:

[0193] Based on the technical solutions of Specific Implementation 1 and Specific Implementation 2, further application cases are given:

[0194] 1. An important celebration event was held in a central square of a city, and it was necessary to effectively prevent the potential risk of "illegal flying" of drones. The security department deployed multi-base station cognitive radars and optoelectronic cameras around the square, and used electromagnetic environment scanning devices to understand the noise floor and available frequency bands at the event site. Before the event started, an unknown small drone entered the air above the square. The system identified the target drone's reliance on Beidou plus inertial navigation through the track characteristics of the target drone. The digital twin model loaded the flight control logic of this type of drone and simulated it. The adversarial AI generated interference strategies for its key visual inertial fusion module. The cross-domain multi-source deception module randomly synthesized false Beidou navigation signals and injected micro-power, combined with geomagnetic distortion and communication link data tampering, to guide the target drone to gradually deviate from the predetermined trajectory. The reconfigurable antenna array continuously adjusted the beam pointing to cover the area where the target drone was located through the attitude information transmitted back in real time by the cognitive radar. When the target has an abnormal attitude but has not yet triggered the return mechanism, the bio-inspired algorithm module quickly calculates a more accurate power and phase update plan to continuously maintain the interference effect, and finally lands the drone in a designated safe area. Afterwards, the system enters standby mode and encrypts and stores the entire set of interference operation records on the intranet for auditing and security review.

[0195] 2. A protective belt is set up along the border, and comprehensive defense is required for small and medium-sized drones that may cross the border. The defender deploys cognitive radars and magnetoacoustic arrays in a multi-base station manner at the border to identify low-altitude and low-speed aircraft. One day, a drone was found trying to bypass the conventional cordon. The system detected that it relied on inertial navigation and GPS fusion and had a certain anti-interference ability at the transmitting end. The digital twin model loads the common flight control structure and anti-interference algorithm of the drone. After testing multiple methods in a simulation environment, the adversarial AI determines to prioritize interference with its inertial navigation correction process. The cross-domain multi-source deception module uses software radio equipment to synthesize fake GPS signals and applies interference in a narrow beam through a reconfigurable antenna array. The visual deception module projects interference patterns on the adjacent ground to disrupt its visual perception. If the drone detects interference and tries to accelerate to escape, the system immediately switches to an emergency interference plan, increases the transmission power and expands the interference bandwidth, and the geomagnetic coil array also enhances the magnetic field distortion intensity. In the end, the drone lost its stable posture and fell into the safe area of ​​the border line. The entire intervention process only outputs high power in a very short time, effectively reducing the impact on normal communications in the surrounding area.

[0196] 3. A defense circle is set up outside the nuclear power plant, and real-time prevention and control is required for drone attacks that high-value targets may face. The perception module includes multi-base station radars, high-precision infrared observers and distributed magnetometer devices distributed around the power plant, which are used to detect airspace movements around the clock. Once a suspicious drone is found approaching the power plant, the digital twin module immediately loads the common navigation fusion algorithm of the drone, and the environmental model takes into account the reflection of buildings around the power plant and the complex wind field factors. In the simulation, the adversarial AI identified that the weakness of the drone lies in the fusion inertial correction step between the barometer and GPS, so the deception module injects false pseudocode into the satellite navigation frequency band, and the geomagnetic coil array creates magnetic field distortion and inserts false calibration instructions into the barometer communication data at critical moments. The reconfigurable antenna array outputs interference signals in the main beam direction according to the positioning parameters continuously transmitted back by the cognitive radar, while maintaining a very low emission intensity in directions unrelated to the safe area of ​​the power plant to avoid interference with normal equipment. If the drone is forced to enter a hovering or hovering state, the bio-inspired algorithm combines the adversarial AI output to further optimize the parameters, completely guiding it to a nearby preset landing strip or triggering its self-destruction program. After the interference is completed, the system will encrypt the entire operation log and write it into a secure database for the anti-terrorism department to conduct post-analysis and summary.

[0197] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a reference structure" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0198] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-system navigation deception system for unmanned aerial vehicles based on intelligent algorithm optimization, characterized by: The system comprises: The perception module is used to detect and identify the UAVs in the target airspace through multi-source information. The perception module includes: A cognitive radar is used to obtain the distance, speed and direction information of the target UAV; the cognitive radar is distributed in multiple base stations; Photoelectric / infrared sensors are used to capture the shape, posture and visual features of the drone; Magnetic and acoustic sensors to detect magnetic field fluctuations or noise generated by the operation of drone motors; Electromagnetic environment scanning device, used to collect the distribution of local satellite navigation frequency bands, ground cellular networks and other radio frequency signals in real time; Digital twin and adversarial AI module, used for model simulation based on UAV flight control and navigation fusion logic, adversarial simulation and optimal deception strategy generation. Digital twin and adversarial AI module includes: Digital twin model module, used to build a simulation environment for the dynamics and flight control algorithms of the target UAV; Adversarial AI engine, used to search and output multi-dimensional deception parameters in digital twin environments through adversarial attacks, reinforcement learning, or generative adversarial networks; The cross-domain multi-source decoy module is used to generate and transmit composite decoy signals targeting drones. The cross-domain multi-source decoy module includes: Multi-system navigation signal synthesis module, used to synthesize and synchronize the standard false navigation signals of GPS, GLONASS, Beidou, and Galileo at the baseband level, and combine the ground cellular network and satellite-based broadband communication signals; The sensor-level deception module is used to simulate or inject the visual, geomagnetic, and air pressure sensor data of the drone to build multi-level hybrid deception; The communication link jammer module is used to tamper with, delay, forge and interfere with the communication data between the drone and its ground station or satellite link; Reconfigurable smart antenna array for flexible control of decoy signal pattern, polarization and phase, including: Metamaterial units for secondary reflection, beamforming or phase modulation of electromagnetic waves; A real-time beam controller that adjusts the operating mode of the antenna array or RIS panel based on the tracking data from the perception module and the output of the digital twin and adversarial AI modules; A biologically inspired intelligent algorithm module, which is used to perform parallel search and optimization in high-dimensional, multi-objective deception parameter space, and assist in completing real-time adaptive deception strategies; The system security concealment module is used to manage the command issuance, data exchange and concealed transmission of the entire system, including: Encrypted access control unit, used to authenticate and encrypt the interactions between each submodule of the system to prevent unauthorized access; Micro-power spread frequency hopping unit to reduce the risk of decoy signals being detected or located; Redundant backup and disaster recovery units are used to maintain the core functions of the system when key nodes fail or are attacked.

2. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to claim 1 is characterized by: The digital twin model module in the digital twin and adversarial AI module further includes: Environmental modeling unit, used to simulate the terrain, buildings, electromagnetic reflection and meteorological data of the actual deployment area; The flight control logic and sensor fusion modeling unit is used to simulate the calculation process of UAV inertial navigation, satellite navigation, vision, air pressure and geomagnetic fusion.

3. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to claim 1 is characterized by: The adversarial AI engine uses a generative adversarial network and an adversarial attack algorithm to generate adversarial samples for the deep learning and data fusion models of the target drone, which are used to introduce confusion interference into the visual and inertial sensor channels.

4. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to claim 1 is characterized by: The cross-domain multi-source deception module includes a geomagnetic field anomaly simulation module, which is composed of a number of controllable electromagnetic coil arrays and is used to apply local magnetic field distortion within the target airspace, causing the geomagnetic navigation of the drone to drift significantly.

5. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to claim 1 is characterized by: The reconfigurable smart antenna array adopts a metamaterial unit and has an electrically tunable phase-shifting array, and achieves phase synchronization under the guidance of cognitive radar return data.

6. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to claim 1 is characterized by: The bio-inspired intelligent algorithm module adopts a neuromorphic computing approach and has global search and fast convergence effects when processing navigation signal power allocation, phase control, and communication interference strategies.

7. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to claim 1 is characterized by: The system also includes a hardware acceleration and disaster recovery platform, which includes GPU, FPGA, and ASIC nodes. The hardware acceleration and disaster recovery platform accelerates large-scale real-time simulation and AI training, and deploys hot backup servers in the cloud to achieve system recovery when a system failure occurs.

8. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to any one of claims 1 to 7, characterized in that: The system comprises the following steps: Sp1, detection and identification step: monitor the target airspace through the perception module to obtain the type, navigation mode, posture and position of the target UAV; Sp2, digital twin modeling and adversarial deduction steps: Based on the target drone type and environmental information, the digital twin model is loaded, and the adversarial AI engine deduces the optimal hybrid deception strategy in a simulated environment; Sp3, decoy signal generation step: the cross-domain multi-source decoy module synthesizes and transmits multi-standard false navigation signals according to the deception parameters output by the adversarial AI, and simulates or forges sensor channels such as vision, geomagnetism or air pressure; Sp4, beam steering and power control step: Through the reconfigurable smart antenna array or RIS, the decoy signal is injected into the receiving end of the target UAV with a specific pattern and power, and the beam shape and phase are dynamically adjusted according to the real-time tracking information; Sp5, closed-loop monitoring and strategy iteration step: After the target drone begins to deviate or fly abnormally, continue to monitor its trajectory and attitude changes, compare the system's expected results, and use quantum or biological heuristic algorithms to update deception parameters within milliseconds or seconds; Sp6. Complete the intervention steps: When the target drone is induced to a safe area, forced to land or disabled, the system enters the standby or withdrawal mode, and stores and analyzes the process data for subsequent strategy optimization.

9. The UAV multi-system navigation deception system based on intelligent algorithm optimization according to claim 8 is characterized by: The adversarial AI engine first identifies the importance of the inertial navigation characteristics and visual modules of the target UAV, and preferentially generates adversarial attack methods for the key perception channel. The reconfigurable intelligent antenna array uses the airspace situation information transmitted in real time by the cognitive radar during beamforming to dynamically avoid interference with non-target areas or legitimate aircraft. During the strategy iteration process, if the target UAV is detected to trigger an autonomous fail-safe mode or escape maneuver, it immediately switches to a preset emergency interference plan to continue to maintain the controlled or harmless state of the target UAV. After the Sp6 completes the intervention, the system automatically calls the security and concealment module to securely erase the deceptive signal transmission records, and saves key operation logs on the intranet or blockchain for subsequent auditing and responsibility tracing.

Images