An illustration of an FDI attack on a UAV is shown in Fig.1. This figure shows the gates that attacker can use for penetration in the system; like inertial measurement units (IMU), GPS satellites, and communication protocols (CP). In this paper we focused on the IMU sensors. IMU is an integrated circuit contains accelerometers and gyroscope which gives the information related to linear and angular motion of the aircraft. This information is used as a feedback for the control system to modify the aircraft attitude based on the desired manoeuvre.
Bledsoe and Browning in their pioneering work  (around 1959) made the first attempt to base their design of neuralnetwork on mathematical logic concept. More sophisticated networks has naturally been developed subsequently. These include the implementation of Enhanced probabilistic Convergent Networks (EPCN). The EPCN is an enhanced form of PCN . The specific enhancements are as detailed in . EPCN is a feed forward neural networks incorporating supervised learning with the addition that the mathematical logic is minimised even further when EPCN is implemented in a hardware. A harware implementation of ANN offers significant advantages to a purely software implementation due to increased speed. For a weightless NN, the mathematical logic is of a reduced complexity than is the case with alternative NN when implemented in a digital intergrated circuit (IC) - this allows an increase in speed. These advantages, amongst others, motivate the work of this paper.
With increased number of wind generators getting connected to power systems, it is likely that its impact on operation of the systems will be significant. This will be particularly so when large amount of power is transmitted over weaker lines (Koch et al., 2003). Compared to other variable speed systems, the advantages of the DFIG is its economic viability in terms of converter power handling, reduced losses, real and reactive power control capabilities, etc.(Lin et al.,2011a; Mohseni and Islam, 2012). Additionally, they can contribute to improve the damping of the electromechanical oscillations emanating from the synchronous machines in the system (Mendonca et al., 2007). Since the DFIG stator windings connect to the grid directly it makes the machine vulnerable to grid disturbances and low voltage conditions on the grid (Odeku et al., 2012). Controlled power electronic converters in the rotor circuit makes the DFIG behave like an inverter, inertia of which has virtually no coupling with the rest of the generator system. Hence with increased wind penetration the effective system inertia is decreased and it can affect reliability of the system significantly when subjected to severe disturbances (Gautam et al., 2009; Campos-Ganoa et al., 2013).
The RSOM-WSN architecture proposed in this thesis is based on the recurrent self organizing map, which is an uncommon neuralnetwork designed to process and recognize multi-dimensional temporal sequences. Mathematical analysis of the RSOM neural algorithm was necessary to establish a solid framework for comparison to state-of- the-art algorithms as well as to understand the applications and environments in which the proposed architecture would succeed. Due to the massively parallel nature of wireless sensor networks, simulations were also necessary to investigate emergent behaviors that were too complex to address analytically. To conduct the simulations in question, this author developed a purpose-built, object-oriented simulator that was highly configurable and designed to investigate several types of network behavior. Simulations investigated network configuration, node mobility, neural activations and resilience against signal corruption caused by noise.
Program modules in contemporary environments must be so set to automatically report ”non-attractive sets” (as we have noted - it is not difficult their description) and to provide the user reliable solution to the problem. Goulianis et al.  showed that the proposed adaptiveneuralnetwork algorithm is characterized by fast convergence for high–dimensional systems.
In recent years, artificial intelligence theory is being applied to SMC. Neural networks (NNs), Fuzzy logic and Neuro-fuzzy are combined with SMC and used in non-linear, time variant and uncertain plant. Some researchers applied fuzzy logic methodology in SMC to reduce the chattering (Barrero et al., 2002) and other researchers have applied sliding mode methodology in fuzzy logic controller (FLC) to improve the stability of system (Hang et al., 2003; Aloui et al., 2011). In Sun et al. (2011), the authors have addressed the robust trajectory tracking problem for a robot manipulator in the presence of uncertainties and disturbances. A neuralnetwork based sliding mode adaptive control (NNSMAC) which is a combination of sliding mode technique, neuralnetwork approximation and adaptive technique is designed for trajectory tracking of the robot manipulator. Authors, in Lin and Lenk (2008), Moradi and Malekizade (2013) and Rossomando et al. (2013) developed a design method of recurrent fuzzy neural networks (RFNN) control system for MIMO non-linear dynamic system. In (Guoling et al., 2014), the authors have addressed hybrid terminal sliding mode surfaces and a new fast decoupled terminal sliding mode control (FDTSMC) scheme.
Increase in the power electronic devices had leads to the harmonic contamination. Harmonic analysis is done to know about the origin and cause of the harmonics in the power system. The low order harmonics is monitored because these are very dangerous and cause serious power quality issues. Wavelet networks (WNs) is an effective version of nonlinear signal processing techniques in recent years an adaptive wavelet neuralnetwork (AWNN) is the most appropriate for prevailing low-order harmonics estimation. Odd-harmonic components of the voltage/current signal are decomposed into the frequency bands by using the above technique. Instead of one complete cycle data for estimating the harmonics the proposed scheme only requires an only half-cycle data point. The back propagation is used for training of the network parameters which is a easy, fast converging and reliable learning algorithm. The experimental signal which is obtained is examined with the projected method. The output result conforms that AWNN technique is effective in estimating the lower order harmonics, inter-harmonics if they are deviated from the fundamental frequency.
Abstract. The Generation and load balance is required in economic scheduling of the generating units and in electricity market trades. Energy forecasting became very important to mitigate some of the challenges that arise from the uncertainty in the resource. The paper presents a short term forecasting of hourly electric power load. Historical data are sourced from Global Energy Forecasting Competition 2017 (GEFCom2017). An adaptive learning algorithm is derived from analysis of system stability to ensure convergence of training process. The simulations with different initial state of network structure demonstrate that training error steadily decrease with an adaptive learning factor starting at different initial values whereas errors behave volatile with constant learning factors.
prediction model depends on a clear understanding of the problem and what input and output parameters are to be used. At this point in the initial selection step, the concern is on the raw data from which a variety of parameters will be collected and developed, because these parameters will form the inputs of the neuron network model for training. In short, input data should cover all key factors that are related with IMD process in the case study. Moreover, the past IMD reliability failure lesson-learnt cases’ root-cause parameters should also be considered as key input factors to improve forecasting performance. For example, the Vramp test early fail (EF) results should be selected as key factors of input and output for neuron network model training. Figure 3(A) shows typical Weibull plot and Figure 3(B) shows typical EF IV curves of IMD Vramp test. The products with EF indicate that reliability risk is high. In our case, total 124 data sets are collected and 11 of them have EF. This includes about 2 years of reliability WLR monitoring history data.
determined with adaptive mutation PSO algorithm, and it can solve the shortcomings that learning process is easy to fall into local optimal value and learning rate is slow when the initial weights is random, also avoid the disadvantages for the traditional PSO search precision is low and late iterative efficiency is not high. Then analyzing the selection of the PSO inertia weight, and choosing the weight adjustment strategy which suitable for synchronous control system. Finally, in the process of PIDNN training, network input layer and output layer weights are set for fixed values according to the actual system. It greatly reduces the computational complexity and improves the practicability of synchronous controller. Optimization of adaptive mutation PSO - PIDNN controller has good adaptability and quickness, no overshoot by simulation analysis. In addition, it solves the disadvantages of the traditional PID control parameters selection difficultly, and can effectively realize the motor synchronization control performance, can meet the engineering requirements about synchronized control system.
The dynamics of robot manipulators are highly nonlinear with strong couplings existing between joints and are frequently subjected to structured and unstructured uncertainties. Fuzzy Logic Controller can very well describe the desired system behavior with simple “if-then” relations owing the designer to derive “if-then” rules manually by trial and error. On the other hand, Neural Networks perform function approximation of a system but cannot interpret the solution obtained neither check if its solution is plausible. The two approaches are complementary. Combining them, Neural Networks will allow learning capability while Fuzzy-Logic will bring knowledge representation (Neuro-Fuzzy). This paper presents the control of six degrees of freedom robot arm (PUMA Robot) using Adaptive Neuro Fuzzy Inference System (ANFIS) based PD plus I controller. Numerical simulation using the dynamic model of six DOF robot arm shows the effectiveness of the approach in trajectory tracking problems. Comparative evaluation with respect to PID, Fuzzy PD+I controls are presented to validate the controller design. The results presented emphasize that a satisfactory tracking precision could be achieved using ANFIS controller than PID and Fuzzy PD+I controllers
81 wavelets, fractals and predictive coding described in this paper. One of the advantages of doing so is that implementation of various techniques can be standardized on dedicated hardware and architectures. Extensive evaluation and assessment for a wide range of different techniques and algorithms can be conveniently carried out on generalized neuralnetwork structures. With dedicated hardware implementation, the massive parallel computing nature of neural networks is quite obvious due to the parallel structure and arrangement of the neurons within each layer. In addition, neural networks can also be implemented on general purpose parallel processing architectures or arrays with programmable capability to change their structures and hence their functionality . At present, research in image compression neural networks is limited to the mode pioneered by conventional technology, namely, information compacting (transforms) + quantization + entropy coding. Neural networks are only developed to target individual problems inside this mode [9, 1, and 23]. Typical examples are the narrow channel type for information compacting and LVQ for quantization, etc. Although significant work has been done towards neuralnetwork development for image compression, and strong competition can be forced on conventional techniques, it is premature to say that neuralnetwork technology can provide better solutions for practical image coding problems in comparison with the traditional techniques. Further research can also be targeted to design neural networks capable of both information compacting and quantizing. Hence the advantages of both techniques can be fully exploited. Therefore, future research work in image compression neural networks can be considered by designing more hidden layers to allow the neural networks go through more interactive training and sophisticated learning procedures. Accordingly, high performance compression algorithms may be developed and implemented in those neural networks. Dynamic connections of various neurons and non-linear transfer functions can also be considered and explored to improve their learning performances for those image patterns with drastically changed statistics.
Two learning methods are generally used in the adaptive TS model to specify the relationship between input and output and to determine the optimized distribution of MF. The TS model utilizes a combination of the least-square method and the back-propagation gradient descent method for training the FIS membership function parameters to identify patterns hidden in a given training dataset [19–22]. Several methods can be utilized for setting up the MF of the ANFIS (e.g., grid partition and subtractive clustering). Regarding the combination of grid partition and ANFIS, grid partition divides the input vector into a number of fuzzy regions using paralleled axis. However, in this method, fuzzy rules increase exponentially when the amount of input variables increases. Therefore, the application of grid partition in ANFIS is not recommended for large input variable problems .
Abstract: Since the electro-hydraulic servo shaking table exists many nonlinear elements, such as, dead zone, friction and blacklash, its acceleration response has higher harmonics which result in acceleration harmonic distortion, when the electro-hydraulic system is excited by sinusoidal signal. For suppressing the harmonic distortion and precisely identify harmonics, a combination of the adaptive linear neuralnetwork and least mean M-estimate (ADALINE-LMM), is proposed to identify the amplitude and phase of each harmonic component. Namely, the Hampel’s three-part M-estimator is applied to provide thresholds for detecting and suppressing the error signal. Harmonic generators are used by this harmonic identification scheme to create input vectors and the value of the identified acceleration signal is subtracted from the true value of the system acceleration response to construct the criterion function. The weight vector of the ADALINE is updated iteratively by the LMM algorithm, and the amplitude and phase of each harmonic, even the results of harmonic components, can be computed directly online. The simulation and experiment are performed to validate the performance of the proposed algorithm. According to the experiment result, the above method of harmonic identification possesses great real-time performance and it has not only good convergence performance but also high identification precision.
WOS filters are special subset of stack filters. Each stack filter is based on a positive Boolean function and needs much computation time to achieve its Boolean computing. This makes the stack filter uneasy to use on application. Until now, the computation time has been only marginally im- proved by using the conventional design approach of stack filter or neuralnetwork. Although the adaptiveneural fil- ter can e ﬀ ectively remove noise of various kinds, including Gaussian noise and impulsive noise, its learning process in- volves a great deal of computational time. This work has proposed a new designing technique to approximate opti- mal WOS filters. The proposed technique, based on thresh- old composition, uses a dichotomous approach to reduce the Boolean computing from 255 levels to two levels. Then the technique of SVMs is used to get an optimal hyperplane to separate those two levels. The advantage of SVMs is that the risk of misclassifying is minimized not only with the exam- ples in the training set, but also with the unseen examples of the test set. Our experimental results have showed that im- ages were processed more eﬃciently than with an adaptiveneural filter.
The quantity of available input data to anticipate the desirable output was evaluated using the M-test. The M-test results help to determine whether there were sufficient data to provide an asymptotic Gamma estimate and subsequently a reliable model. The M-test analysis results are presented in Figure 5. As it can be witnessed from the figure, the data standard error for 1360 numbers is the lowest. This number has been selected as the training network data number. The remaining data were used to test the model and estimate the results.
In case of flood/flooding in urban areas, the operation strategy for sewerage systems in Taipei City is to set up pumping stations which are the major hydraulic facilities for inner rainwater discharges. Undoubtedly, pumping stations play a key role in flood reduction in metropolitan areas. Nev- ertheless, fast rising peak flows resulting from urbanization and climate change are highly challenging to existing sew- erage systems. In fact, the current pumping operation pro- cedure depends more highly on the experiences of local op- erators than on the pumping operation standards. In other words, there are no explicit guidelines for pumping opera- tions. Operators have to stand by prior to the coming of extreme rainfall events and keep monitoring and operating until storms’ departure. It is time- and human resources- consuming with no guarantee of safe pumping operations because only the information of current water level measure- ments is available for operators. Therefore, it is necessary to construct an efficient and accurate pumping operation model to simulate the drainage mechanism for discharging rainwa- ter in advance. Furthermore, the advantages of building a suitable and successful pumping operation prediction model for a sewerage system are to increase its storage capacity prior to peak flows by reducing water levels in advance and to decrease flood/flooding probability by speeding up discharge rates during storm periods. To achieve this goal, two rule- based fuzzy neural networks are introduced in this study by taking the predictive water levels into account to effectively
Several strategies of the neuraladaptive control exist which we quote: direct adaptiveneural control, indirect adaptive neuronal control, adaptiveneural internal model control, adaptive depth control based on feedforward ne- ural networks, robust adaptiveneural control, Feedback- Linearization based neuraladaptive control, adaptive ne- ural network model based nonlinear predictive control [5-13]. Each strategy has neuraladaptive control archi- tecture, the algorithms used during the calculation of the parameters and stability conditions. It has three types of neuraladaptive control architectures. The first type of architecture consists of a neural controller and a system to be controlled. The second type of neural architecture includes a controller, a system to be controlled and his neural model. The third type of architecture is composed of a neuronal controller, one or more robustness filter, a system to be controlled and his neural model.
The comparison of the results revealed that the suggested model could increase the forecast accuracy and perform better than MLP, MLR and AR models (Kisi 2008). Rainfall - runoff modeling using a combination of wavelet - neuralnetwork for the Ligvan Chai (Tabriz, Iran) catchment has been studied. The results showed that the proposed model can be used to predict long-term and short-term precipitation(Nourani et al. 2009). Approach improvement based in the precipitation-runoff modeling using a combination of artificial neuralnetwork-Wavelet is performed, which shows that the model which precipitation and discharge data, as an input entered, outperformed than the model which just precipitation was entered as an input (Chua and Wong 2010). A method based on trans-form discrete wavelet and artificial neural networks to predict applied flow in seasonal river in semi-arid watershed in Cyprus were presented. Wavelet coefficients as an input Levenberg Marquardt (LM) artificial neuralnetwork models was used to predict the flow. The relative performance of the wavelet-neuralnetwork (WANN) and artificial neuralnetwork (ANN) models was compared to lead times of 1 and 3 days flow forecasting for two different rivers. In both cases, neuralnetwork-Wavelet model for flow predictions, was more accurate than artificial neuralnetwork. The results indicate that wavelet-neuralnetwork models are a promising new method of short-term flow forecasting in non-perennial rivers in semi-arid watersheds such as those found in Cyprus (Adamowski and Sun 2010). Two hybrid methods of artificial intelligence for modeling rainfall - runoff for two watersheds are presented in Azerbaijan, Iran. The first model was SARIMAX-ANN (artificial neuralnetwork- Seasonal Auto Regressive Integrated Moving Average with exogenous), and the second model was wavelet - neuralnetwork system - Adaptive Fuzzy (ANFIS). The results showed that although the proposed model can predict both short-term and long-term runoff according to seasonal effects, but the second model is relatively better. Because in this proposed model due to use of multiple scales of time-series, rainfall - runoff data has been applied as an input layer of adaptiveneuralnetwork - fuzzy system(Nourani et al. 2011).