A composite control scheme that deals with longitudinal dynamics problems of the missiles suffering from matched and unmatched disturbances is proposed.First,a Super-Twisting Disturbance Observer (STDO) is used to estimate the unknown disturbance,and an Integral Sliding Mode Controller (ISMC) is designed to compensate for the input disturbance.Subsequently,the differential game theory and Adaptive Dynamic Programming (ADP) algorithm are combined to approximate the optimal control by constructing a single critic Neural Network (NN) to counteract the unmatched disturbance online.The stability of closed-loop system and the convergence of the weights of critic network are proved by Lyapunov stability theory.Finally, the effectiveness of the proposed scheme is proved by designing a longitudinal dynamics model of a missile.

The sensitivity model of multivariate detection is studied,and the model is derived and established by using the detection signal-to-noise ratio,detection probability and false alarm rate.The accuracy of the model is proved by experiments.On this basis,the influence of APD unit number,pulse accumulation times,fill factor and dead time on detection sensitivity under different detection conditions is analyzed,which provides theoretical foundation for the selection of multivariate APD.The results show that:1) When the pulse accumulation number is 6,the larger the fill factor,the smaller the dead time and the better the ranging performance;2) Under high background noise,increasing the number of APD units can significantly improve the detection sensitivity;and 3) Under low background noise,4-unit APD can achieve better ranging effect.

To the problem of many-to-one Pursuit-Evasion (PE) game strategy,an optimal PE control strategy is proposed in the framework of explicit cooperative guidance.Initially,based on graph theory,the many-to-one PE game problem is transformed into a multi-agent consensus control problem.Then,a critic neural network is utilized to solve the control strategies online by Adaptive Dynamic Programming (ADP) technology. The stability of the system is proved by Lyapunov direct method.Considering that the PE strategies always appear in pairs and it is difficult for a single evader to choose escape strategy,an overall escape strategy calculation method using dynamic weighting is proposed.Finally,a two-dimensional multi-missile attack and defense game model is established to ensure the effectiveness of the game strategy.

Regarding the high-conflict problem that can not be solved by conventional DS evidence theory,a new improved DS evidence theory method is proposed.Firstly,Harley-like entropy and Shannon entropy are used to measure the comprehensive uncertainty of evidence,and to describe the non-concreteness and randomness of evidence.Secondly,Pearson correlation coefficient is used to measure the similarity between the evidence.Then,based on the game theory,the combination weights are given to ensure the accuracy and rationality of the corrected weights.Finally,the modified Basic Probability Assignment (BPA) is calculated using the DS combination rule to obtain the fusion result.Through the analysis of the cases in comparison with other methods,it is proved that the proposed algorithm has higher fusion accuracy,faster recognition speed for common sense proposition,and more scientific and effective consideration of problems.

With the increasing proportion of feedback input in battlefield environment,AGNES hierarchical clustering algorithm in unsupervised learning is proposed to improve the traditional theoretical framework of multi-agent alliance.Considering the complexity and fuzziness factors in the environment and the efficiency of different individual sensors,the multi-agent model of the alliance is described and the specific prompting steps of the dynamic alliance detection system under the multi-agent cross prompt are designed.Furthermore,the cognition consistency function combined the unsupervised learning AGNES clustering algorithm with target detection information is improved,so that the system is able to directly establish a dynamic model from the changing environment,thus to achieve the long-term benefit accumulation of the alliance optimization task and make a more favorable decision for the current short-term benefit.Simulation shows that compared with the traditional overall mobilization strategy such as swarm intelligence algorithm,the improved algorithm is more in line with practical needs.

In order to consider the effect of historical information on future navigation results,and use deeper integrated navigation information for fusion,an INS/GNSS deep integrated navigation method based on graph optimization is proposed.By taking measurement information and state propagation as constraint information,the optimal cost function is constructed in time domain,and the optimal state estimation is obtained by using the Levenberg-Marquardt method.The method is evaluated and analyzed by INS/GPS deep integrated navigation system simulation experiment.The simulation results show that compared with the conventional Kalman filtering method,the proposed method reduces the mean position errors of the three axes by 38.5%,21.0%,and 30.9% respectively; and the mean velocity errors are reduced by 31.4%,52.8% and 57.3%.The proposed algorithm can effectively improve the positioning accuracy.

Regarding the problem that the Auxiliary Particle Filter Track-Before-Detect (APF-TBD) can not achieve good detection and tracking performance due to the lack of particles,an auxiliary particle filter track-before-detect based on optimized genetic re-sampling (OGRAPF-TBD) method is proposed.The optimized genetic re-sampling algorithm is applied in re-sampling of APF-TBD.In the re-sampling,high quality particles are chosen according to the weight,and new particles are obtained by crossover of sorting groups and mutation.The method not only maintains the advantage of APF-TBD in improving the accuracy of sampling particles by optimizing the importance distribution function,but also introduces the optimized genetic re-sampling idea into re-sampling,which can effectively solve the problem of particle shortage and increase the number of effective particles.Simulation results show that: compared with APF-TBD and particle filter track-before-detect,OGRAPF-TBD has higher target detection probability and tracking accuracy,and stronger applicability.

The process of laser angle deception jamming is analyzed from the energy level.Firstly,the power density transmission formulas of targets with different shapes are derived,their energy transmission performance is obtained via simulation analysis.The energy transmission in the process of laser angle deception jamming is analyzed and the expression of protected airspace is deduced based on energy suppression coefficient.Finally,the influencing factors of protected airspace are simulated.The results show that the protection angle is affected by the incident distance of the indicating laser,interfering laser as well as the curvature characteristics of the real target.

In order to improve the tracking accuracy of fighter in the scene with illumination variation, scale variation,occlusion,and deformation,a kernel correlation filtering algorithm based on multi-feature fusion is proposed.The color feature,texture feature,and convolution feature are fused,and the maximum response value outputted after fusion is the detected fighter position.In addition,the scale filter is adopted to estimate the fighter size.The concept of sidelobe ratio is introduced to judge the fighter occlusion situation while the template is updated.Experimental results show that the accuracy and success rate of the algorithm reach 77.6% and 73.3% respectively.In the cases of fast motion,background clutter,deformation,target out-of-view,illumination variation,rotation,and low resolution,etc.the tracking accuracy and tracking success rate are ranked first compared with the KCF,DSST and MOSSE algorithms.

In order to improve the rapidity and accuracy of the response of the servo system of a certain obstacle breaking weapon,the neural network sliding mode control is studied.Combined with the model of servo system,the Ridgelet Recurrent Neural Network(RRNN) is introduced to dynamically and adaptively approximate the model,which can effectively improve the speed of response and robustness.Through the Ridgelet Recurrent Neural Network Sliding Mode Controller (RRNN-SMC),the influence of uncertain factors such as load disturbance and parameter change is effectively overcome.Finally,the Particle Swarm Optimization (PSO) algorithm is applied to optimize the ridgelet parameters and link weights,which can effectively reduce the influence of sliding mode chattering.The simulation results show that the method can ensure the stability of the servo system,accelerate dynamic real-time response and improve the precision of servo control.

For the control problem of a class of high-order unknown nonlinear systems with unmatched interference,a virtual recursive control method based on Auto-Coupling Proportional-Integral(ACPI) control theory is proposed.By introducing an unknown state and the concept of total disturbance,the unmatched nonlinear system is mapped into an equivalent linear unknown system,and a controlled error system excited by total disturbance in reverse phase is constructed.The ACPI virtual instructions with speed factor as the core linkage factor are constructed in combination with ACPI control theory.In order to effectively avoid the problem of “differential explosion” in back-stepping control,the differential of each virtual instruction is defined as unknown bounded disturbance.Simulation results show that ACPI control system not only has good dynamic quality and steady-state performance,but also has good robustness against disturbance,so it has a wide application prospect in the control field of high-order nonlinear unknown systems.

In the wavelet threshold function,the image cannot be restored optimally due to the discontinuity between signals and the error between the estimated wavelet coefficients and the wavelet coefficients of the original signal.To solve the problem,an improved Collaborative Quantum Particle Swarm Optimization (CQPSO) method is proposed to optimize the wavelet function.The method introduces adaptive shrinkage expansion factor on the basis of the CQPSO,optimizes the adjustment factors and thresholds in wavelet threshold function by the improved CQPSO.The simulation images and data show that the algorithm successfully reduces the distortion of useful signals,and improves the effect by 5%~10% compared with other algorithms under the evaluation standard of signal-to-noise ratio.Under the evaluation standard of Root Mean Square Error(RMSE),the effect of the algorithm is improved by 16% to 20% compared with that of other algorithms, which has better practical value.

The siamese network tracking algorithm can turn tracking problem into similarity matching problem,which has attracted widespread attention.However,most algorithms based on siamese network cannot be applied to devices such as mobile terminals or embedded devices with insufficient computing power.Thus,a lightweight,high-speed tracking algorithm based on siamese network is proposed.The algorithm deploys MobileNetV2 as the backbone network,which has good feature extraction capabilities and less parameters,and the network parameters are further reduced by group convolution,Crop and other operations to improve the network operation speed.SE module is added to the inverted residual structure to dynamically adjust the weight of the model,highlight important information.Through the fusion of information of different layers,the expression of the target multi-scale semantic information is improved. The results of the target tracking benchmark library OTB100 and VOT2018 show that,compared with the existing general tracking algorithms,the proposed algorithm not only maintains high accuracy,but also runs at a speed of 170 frames per second,which has a good engineering application prospect.

Various sensors in flight control system play a vital role in stability and control of the aircraft,and are critical for the safety of aircraft.The conventional redundancy method has the characteristics of “high safety and low cost-efficiency”.Improving the safety of the system through redundancy design has brought some problems on the aircraft weight and structural design,system integration,maintenance and inspection costs.Aiming at various sensor failures in civil aircraft flight control systems,the typical failure mode is introduced,as well as the design considerations on failure tolerance.The research and application of model based methods and information based methods in related fields are elaborated.Finally,the present challenges and future developments of sensor fault diagnosis are analyzed,and the great potential of advanced fault diagnosis methods in aircraft flight control systems is clarified.

In many-to-many radar countermeasure signal-level simulation system,the close coupling between the radars and the jammers often results in poor extensibility.A signal descriptor method is proposed to solve the problem.Firstly,the theory and method of digital simulation of coherent pulse train target echo are analyzed.On this basis,the simulation process of digital signal is decomposed,the components of signal descriptors are extracted,and then five kinds of signal descriptors are obtained,and how the signal descriptors adapt to the diversity of signal styles is discussed.Finally,the effectiveness of the method and its adaptability to many-to-many radar countermeasure signal-level simulation are verified via the simulation examples with two radars and two jammers.

There is no shape,texture or color information in the observation image of the dim target under the starry background,and it has high similarity with stars and noise.Background suppression and noise elimination is the basis of retaining the target information to the maximum extent.Based on the analysis of the histogram characteristics of the observation image of small and dim target under the starry background,a starry background parameters estimation method is proposed,which continuously discards the high-gray areas and reduces the sample variance,and the background parameters are used to suppress noise.Experiments show that the method can filter out most of the noise,retain and enhance the target to the maximum extent under the influence of complex background noise and stars.

In binocular camera calibration,BP neural network is affected by initial weights and thresholds.A method based on BP neural network optimized by the improved genetic algorithm is proposed to solve the problem.The crossover and mutation probabilities of the genetic algorithm are improved.The direct contrast of world coordinates is adopted to make the data more intuitive.The experiment shows that higher accuracy can be achieved.

Regarding the large amount of samples,random truncation and multiple data sources of the outfield use & support data in airborne electronic countermeasure system,a method for continuous evaluation of electronic countermeasure system reliability based on outfield use and support data is proposed.Firstly,the steps of continuous evaluation of the reliabilty of the airborne electronic countermeasure system is analyzed.Secondly,the collection,analysis and verification methods of outfield use and support are given.Then,the equipment-level and system-level reliability evaluation models based on use and support data are constructed.Finally,the correctness of the proposed method is verified via an case analysis of a certain type of electronic countermeasure system.

In view of the significant progress of deep learning in the field of image recognition,in the background of unmanned helicopter autonomous landing,SSD network is adopted to identify the landing mark for the complex landing environment and landing mark design.Aiming at the disadvantage of low recognition rate of small targets in SSD network,SSD network is improved based on deep residual network and feature pyramid network structure,and the ResNet101 is used instead of VGG-16 network,the traditional up-sampling structure is improved by using feature pyramid network structure,which integrates the high-level semantic information of the detection network into the low-level feature information.Finally,the recognition effect of the improved network is verified via experiments.

In order to solve the problem of model uncertainty and external wind disturbance in the process of quadrotor attitude control,an inner and outer-loop control method is proposed.The active disturbance rejection controller is designed in the inner loop,in which the extended state observer and nonlinear state error feedback controller are adopted to estimate and compensate for the total disturbance of the system in real time.The non-singular terminal sliding mode controller is designed in the outer loop to improve the response speed of the system,and the stability of the control algorithm for the inner and outer loops is verified.The simulation results of attitude angle tracking show that the designed controller has high tracking accuracy and good anti-interference ability,and can effectively realize the attitude control of quadrotor.

In order to improve the stability precision of the electro-optical stabilized platform,a nonlinear integral sliding mode control method is proposed,which can solve the problems of the steady-state error of the conventional sliding mode control,as well as large overshoot,long adjustment time and poor transient performance of the conventional integral sliding mode control.Adopting new reaching law can improve the response speed and reduce the chattering.Moreover,the nonlinear integral sliding mode surface is added to sliding mode observer to improve the anti-disturbance ability of the system.Simulation results show that the proposed method can track the input instructions more rapidly and accurately without overshoot in comparison with conventional sliding mode and integral sliding mode.By using this method,the stabilization accuracy of the line of sight is improved by 28.43% compared with that of the conventional sliding mode.Therefore,the control algorithm designed not only improves the dynamic performance of the system,but also improves the stability accuracy of the platform.

The Enhanced Flight Vision System(EFVS)is a complex airborne electronic system,for which the conventional fault analysis is complex and error prone.On this basis,the concept of formal modeling is introduced and a formal modeling method of a EFVS based on fault coupling model is proposed.The fault propagation mode of the system is determined via functional interaction model of the EFVS.The system architecture and data flow are abstracted,and the fault coupling formal model is established hierarchically.The model checking tool is introduced to verify the system model,and automatic operation is carried out in combination with typical system fault states,so as to obtain the minimal cut set of the fault tree in fault state.The results show that the method is highly automated,which is helpful to improve the efficiency of the fault analysis process of enhanced flight vision system.