A Novel Flight Dynamics Modeling Using Robust Support Vector Regression against Adversarial Attacks

A linear parameter varying (LPV) flight dynamics model (FDM) is proposed to cater for atmospheric disturbance analysis in special flight conditions. A novel FDM which is capable of addressing the influence of turbulent wind is derived under the wind frame. An affine parameter dependent LPV model with wind effects is built based on function substitution method. The optimal solution for the decomposing function of the LPV FDM is obtained by genetic algorithm (GA). The analysis of dynamic response indicates that the genetic-optimized LPV FDM approximates the nonlinear FDM evidently, since it identifies the instantaneous dynamics and flight states varying in a wide range. The simulations of approach and landing against wind shear show that the genetic-optimized LPV FDM captures the instantaneous dynamic response when flying through turbulent wind, indicating that the LPV model can be further applied to turbulent wind special flight analysis and control law design.

Robust support vector regression in the primal

In this paper, the flight dynamic model of Vertical Take-Off and Landing (VTOL) aircraft which equips a propeller-wing-flap system that produces slipstream is established. The complete aircraft is modeled by combining experimental results, theoretical calculation, and empirical data. Firstly, the propeller deflected slipstream effects and the parameters of ducted fan are obtained by bench test. The DATCOM are used for generating the aerodynamic coefficients. Then, the 6-DoF nonlinear flight dynamic model in both hover and forward flight conditions are deducted by employing the assumption of linear superposition. The gyro effect and gravity are considered. Finally, specific trim conditions in hovering, transitional and level flight states are determined, and dynamic characteristics at trim are studied by both full model simulation and linear system analysis obtained through small-perturbation theory. It is found that the aircraft is critically stable in hovering condition. The damping ratio of short-period mode is relatively small, and the phugoid mode divergence in the transition state. The aircraft is less well damped in the level flight condition. The built model accurately reflects the dynamic characteristics of the VTOL aircraft and provides help for the design of the flight control system afterward.KeywordsSlipstreamVertical Take-off and Landing (VTOL) aircraftBench testDynamic modeling

Processing experimental data with uncertainty is a huge challenge for robust regression modeling and high-accuracy data forecasting. The objective of the present study is the development of a robust support vector regression model for processing observed data of independent variables containing polyhedral perturbation and achievement of high forecasting precision. Firstly, the conception of the data collection of polyhedral perturbation is given and analyzed. Secondly, a novel robust support vector regression model is constructed by replacing the original independent variables in a linear support vector regression model with the independent variables containing polyhedral perturbation. Thirdly, the novel robust linear support vector regression method is also expanded to the non-linear regression model. Both the two models are validated by the linear and nonlinear numerical regression experiments. Comparison of the experimental results show that the proposed method provide more accurate predictions than the traditional regression method for processing data with polyhedral perturbation.

Accurately modeling the DNA sequence preferences of transcription factors (TFs), and using these models to predict in vivo genomic binding sites for TFs, are key pieces in deciphering the regulatory code. These efforts have been frustrated by the limited availability and accuracy of TF binding site motifs, usually represented as position-specific scoring matrices (PSSMs), which may match large numbers of sites and produce an unreliable list of target genes. Recently, protein binding microarray (PBM) experiments have emerged as a new source of high resolution data on in vitro TF binding specificities. PBM data has been analyzed either by estimating PSSMs or via rank statistics on probe intensities, so that individual sequence patterns are assigned enrichment scores (E-scores). This representation is informative but unwieldy because every TF is assigned a list of thousands of scored sequence patterns. Meanwhile, high-resolution in vivo TF occupancy data from ChIP-seq experiments is also increasingly available. We have developed a flexible discriminative framework for learning TF binding preferences from high resolution in vitro and in vivo data. We first trained support vector regression (SVR) models on PBM data to learn the mapping from probe sequences to binding intensities. We used a novel -mer based string kernel called the di-mismatch kernel to represent probe sequence similarities. The SVR models are more compact than E-scores, more expressive than PSSMs, and can be readily used to scan genomics regions to predict in vivo occupancy. Using a large data set of yeast and mouse TFs, we found that our SVR models can better predict probe intensity than the E-score method or PBM-derived PSSMs. Moreover, by using SVRs to score yeast, mouse, and human genomic regions, we were better able to predict genomic occupancy as measured by ChIP-chip and ChIP-seq experiments. Finally, we found that by training kernel-based models directly on ChIP-seq data, we greatly improved in vivo occupancy prediction, and by comparing a TF's in vitro and in vivo models, we could identify cofactors and disambiguate direct and indirect binding.

As more and more data are available, training a machine learning model can be extremely intractable, especially for complex models like Support Vector Regression (SVR) training of which requires solving a large quadratic programming optimization problem. Selecting a small data subset that can effectively represent the characteristic features of training data and preserve their distribution is an efficient way to solve this problem. This paper proposes a systematic approach to select the best representative data for SVR training. The distributions of both predictor and response variables are preserved in the selected subset via a 2-layer data clustering strategy. A 2-layer step-wise greedy algorithm is introduced to select best data points for constructing a reduced training set. The proposed method has been applied for predicting deck’s win rates in the Clash Royale Challenge, in which 10 subsets containing hundreds of data examples were selected from 100k for training 10 SVR models to maximize their prediction performance evaluated using R-squared metric. Our final submission having a R2 score of 0.225682 won the 3rd place among over 1200 solutions submitted by 115 teams.

Unmanned Aerial System (UAS)-based phenotyping of soybean using multi-sensor data fusion and extreme learning machine

It is key index of air-jet texturing yarn quality such as air-jet texturing yarn strength and so on. It can well control air-jet texturing yarn quality by predicting yarn strength and so on. Generally, it is normal used to predict yarn strength such as Multiple Linear Regression (MLR), Support Vector Regression (SVR) and shallow Artificial Neural Network (ANN). Because the processing of air-jet texturing yarn production has time sequence, the paper proposes a new deep neural network, it is Long Short-Term Memory neural network (LSTM). It used 1800 sets of data to train LSTM, SVR and ANN. It tested LSTM, MLR, SVR and ANN with 200 sets of data. Experimental results show that LSTM neural network is the best accuracy among these four algorithms.

A dynamic particle filter-support vector regression method for reliability prediction

This paper presents a real-time closed-loop rotorcraft simulation framework using HeliUM-A, a high-fidelity flight dynamics analysis, and a Simulink®-based flight control system model. Serial optimization and parallel computing techniques are introduced in HeliUM-A to achieve real-time speeds. A customized ordinary differential equation solver with parallel load balancing enables accelerated time marching simulations. Software interfaces are introduced to encapsulate HeliUM-A into a Level-2 S-function Simulink® block. Using standardized Simulink® ports, control inputs, rotor/body states and their time derivatives as well as relevant output quantities are communicated in-memory between Simulink® and HeliUM-A for closed-loop execution. This encapsulation retains the parallel computing improvements in HeliUM-A when executed through MATLAB, Simulink® or through the compiled executable automatically generated by the Simulink Coder. The framework is demonstrated on a coaxial compound scout helicopter with a pusher-propeller. Closed-loop vehicle and rotor responses are compared between two flight dynamics models: a stitched simulation model based on linearized models extracted from HeliUM-A, and the original HeliUM-A model. Time domain and frequency sweep comparisons showed excellent agreement between the stitched and nonlinear HeliUM-A models demonstrating that the new HeliUM-A based closed-loop framework shows expected behaviour. The HeliUM-A bare airframe model can be used to simulate various rotorcraft configurations in edge-of-the-envelope flight conditions, whereas the stitched models are accurate for center-of-the-envelope conditions.

The rod ejection accident (REA) is the design-basis reactivity initiated event and an important aspect for a pressurized water reactor (PWR). The consequence of REA is that it introduces a large positive reactivity insertion in a core, which leads to a fast large power excursion and other parameters changing. Thus, it is important to understand the uncertainty in the parameters of reactor core when REA happens. This paper applies support vector regression (SVR) to analyze accident scenarios with control rod ejection. SVR is an approach based on machine learning and soft computing. SVR, by definition, is an application of support vector machine (SVM) to nonlinear regression problem. Furthermore, the objective of this paper is to train SVR model to identify both safe and potentially unsafe power plant conditions based on real time plant data. The data is obtained from computer generated accident scenarios and is divided into two datasets, training datasets and test datasets. The training dataset are used to train the SVR model and the test dataset are used to test the validation of this model. And then the results obtained by SVR model are compared with that of artificial neural network (ANN) model. The comparison results show that SVR model has superior performance over ANN model and agree well with the general understanding. Because the proposed methodology achieve accurate results, it is likely to be suitable for other data processing of nuclear engineering.

Aircraft flight dynamics trim is the basis for the large civil aircraft flight dynamics characteristics analysis. In this paper, the full nonlinear flight dynamics model of civil aircraft is systematically deduced, an iterative trim method for the six-degree-of-freedom flight dynamics model based on the Downhill Simplex optimization algorithm is designed, and the trim accuracy and computational efficiency of the Downhill Simplex optimization algorithm are analyzed and the variation of trim variables under different initial weights is studied. Results show that the Downhill Simplex optimization algorithm can effectively solve the six-degree-of-freedom full nonlinear aerodynamics trim problem of large civil aircraft under the condition of straight forward and level flight, and the trim deviation can be controlled within the range of ±0.05N/Nm within 500 steps. The flight dynamics iterative trim method based on the Downhill Simplex optimization algorithm designed in this paper can provide reference and technical support for civil aircraft flight simulation, flight dynamics characteristics and flight quality research.

Collaborative Filtering (CF) is a successful technology that has been implemented in E-commerce recommender systems. However, the risks of shilling attacks have already aroused increasing concerns of the society. Current solutions mainly focus on attack detection methods and robust CF algorithms that have flaws of unassured prediction accuracy. Furthermore, attack detection methods require a threshold to distinguish normal users from fake users and suffer from the problems of false positive if the threshold is too high and false negative if too low. This paper proposes a soft-decision method, Neighbor Selection with Variable-Length Partitions (VLPNS), to reduce false positive rate through marking suspicious fakers instead of deleting them directly such that misclassified normal users can still contribute to the similarity calculation. The method works as follows: First, it gets user's suspicion probability by applying SVM. It then generates partitions of variable sizes from which different numbers of neighbors can be selected by using the bisecting c-means clustering algorithm. Finally, it chooses neighbors considering the user's suspicion degree and similarity with target user at the same time. Theoretical and experimental analysis show that our approach ensures an excellent prediction accuracy against shilling attacks.

In this article, we implement three types of data-driven algorithms for the inverse design of the polarization rotator (PR) unit cell. Toward this end, a novel configuration pattern of the PR unit cell is proposed, which is reshapable to different geometries with a large number of design variables and recasts as a regression problem. The state-of-the-art algorithms including the neural network (NN), the deep neural networks (DNNs) with multiple hidden layers, and the support vector regression (SVR) are experimented with cross validations for ensuring the prediction generalizability. Averaged over all the experiments with competitive performances, the highest prediction accuracy about 95.23% was achieved for the SVR algorithm. This demonstrates the enormous capability of the data-driven algorithms in the geometrical dimension prediction of the unit cells for any given frequency band designated in the radar range (X, Ku, K, and Ka). The proposed inverse design procedure can expedite, facilitate, and to some extent replace the conventional and time-consuming manual design approaches with electromagnetic (EM) simulation software. Although these models can be very efficient in practice, they might be vulnerable against adversarial attacks that craft fake inputs to purposely fool the victim regressors toward adversary's wishes. This poses security concerns for the learning-based algorithms and might negatively affect their prediction reliabilities in runtime. In this article, we only characterize the existence of the adversarial attacks for the regression models using the fast gradient sign method. Our conducted experiments uncover that the fooling rate of all the aforementioned cutting-edge NN and DNN-based regressors is above 98% and this rate for the SVR model is about 11% better than other models.

Real time estimation of impaired aircraft flight envelope using feedforward neural networks