While quadrotors are becoming more popular, their controllers should be improved. In this study, neuralnetworkcontrol of quadrotors is aimed to obtain an artificial intelligence based controller. Firstly, the quadrotor is modeled according to quadrotor dynamics. Then, PD controllers for x, y, yaw and z control of quadrotor are implemented as classical controllers. The results for these controllers are recorded as training data of NN controllers. As the proposed controllers, NN controllers are trained according to these data and performance of these results are examined. The results verify that NN controllers achieve good trajectory tracking results.
data over the other procedures using traditional neural net- work. Furthermore, enhancements to the SpikeProp learn- ing algorithm are presented. These enhancements provide additional learning rules for the synaptic delays, time con- stants and for the neurons thresholds. Simulated experi- ments have been conducted and the achieved results show a remarkable improvement in the overall performance. This work can also be extended to investigate online learning and address the effect of costs on the decisions in terms of computational time and complexity.
That is, first a control time series u is chosen. Using a model we look at what would happen if this series of control signals is used: the plant output y ˆ can be predicted, then next a cost function (1) is used which shows how good or bad the outcome is. Then let the optimization algorithm choose a new control time series u that is better by means of the cost function (1). Then number of iterations of these steps is done in order to find a series of control signals that is optimal with respect to the cost function (1).
Recently, neuralnetwork technology attracts many attentions in the design of robot controllers. It has been pointed out that multi-layered neuralnetwork can be used for the approximation of any nonlinear function. Other advantages of the neural networks often cited are parallel distributed structure, and learning ability. They make such the artificial intelligent technology attractive not only in the application areas such as pattern recognition, information and graphics processing, but also in intelligent control of nonlinear and complicated systems such as robot manipulators (Sanger, 1994), (Kim and Lewis, 1999), (Kwan and Lewis, 2000), (Jung and Yim, 2001) (Yu and Wang, 2001). A new field in robot control using neuralnetwork technology is beginning to emerge to deal with the issues related to the dynamics in the robot control design. A neuralnetwork based dynamics compensation method has been proposed for trajectory control of a robot system (Jung and Hsia, 1996). A combined approach of neuralnetwork and sliding mode technology for both feedback linearization and control error compensation has been presented (Barambones and Etxebarria, 2002). Sensitivity of a neuralnetwork performance to learning rate in robot control has been investigated (Clark and Mills, 2000).
neuralnetwork (NNs) has the compatibility to improve control of power electronic systems. NNs have self-adapting and super-fast computing features that make them well suited to handling nonlinearities, uncertainties and parameter variations that can occur in a controlled plant . It is used for increasing the processing speed, response, convergence, robustness, accuracy, precision, tracking ability, adaptive ability, steady-state and transient stabilities, etc. –. Neuralnetwork (NN) based algorithms are used to extract required information after processing of signals by learning or training and activation function –. Neuralnetwork based control have created much attention in electrical engineering including power quality problems such as load balancing, reactive power compensation, harmonics elimination and neutral current compensation in a four wire distribution system. In this paper we use three phase four wire distributed system which is mainly concerned about the neuralnetwork controller implemented in a shunt connected compensating device known as DSTATCOM for the extraction of active power and reactive power components of three-phase distorted load currents. Proposed control algorithm is used for PFC and ZVR modes of operation to maintain a balanced and sinusoidal supply current with a self-supporting dc bus of VSC of DSTATCOM, for this purpose Kohonen learning method has been used. Kohonen learning is used to extract the fundamental components of load current in terms of conductance and susceptance.
For all feed forward neural networks, the training sets and the test sets are prepared. The training set consists of the inputs of the neuralnetwork, for example source node, destination node, link costs and queue lengths, and the desired outputs, such as best neighbor node, best path from source to destination. In training stage, random values between – 0.5 and 0.5 are used as initial weights of connections of the neural networks. The neural networks are trained on 30 training sets with the back propagation algorithm for each computer networks (CN1, CN2). For any set, the training is continued until the mean squared error (MSE) becomes acceptable. The neural networks are trained on all possible routing decisions. In this stage, values of learning rate (η) and momentum rate (α) are selected by trial and error, which are given in Table (4). The obtained results are shown in Figures (6)_ (11). These Figures view mean squared error versus number of epochs, where this error is decreased when number of epochs is increased.
From last few decades traditional ON-OFF controllers and conventional Proportional plus Integral plus Derivative controllers (PID) were used for controlling the Air Conditioning system. But these controllers don’t always produce desired and fast response due to dynamic characteristics of the systems. And tuning of conventional PID controllers was difficult. So in recent years intelligent control system are designed for controlling the air conditioning (AC) system. Neuralnetwork controllers are best option for controlling the system. By choosing suitable neural networks, learning method and input- output data the neuralnetwork can be learn the system states, to predict the future behavior of the system. Neuralnetwork controllers give the fast response and these controllers are reliable and robustness. The system which are linear or non-linear can be controlled by neuralnetwork because of neural networks have the ability to approximate arbitrator value through self-learning property. Due to this property of neuralnetwork the designed controllers make up the system uncertainties and system nonlinearities and make system response stable.
the information. However, early TCP implementations had very bad retransmission behavior. When this packet loss occurred, the end points sent extra packets that repeated the information lost, doubling the data rate sent, exactly the opposite of what should be done during congestion. This pushed the entire network into a 'congestion collapse' where most packets were lost and the resultant throughput was negligible. Congestion is directly related to the amount of traffic that is present in the network. More the traffic more will be congestion. Congestion results when the traffic on the network exceeds the channel capacity. A router is able to handle optimum amounts of traffic but if the traffics exceeds the channel capacity, the entire system collapses and all or a majority of packets transmitted will be lost. To avoid congestion, we either increase the resource or decrease the load. Factors resulting in congestion are limited queue size, insufficient memory ,low band width, low CPU processing speed of router etc. II. R OUTING WITH CONGESTION CONTROL
Underwater glider is an important equipment for ocean research, water quality detection and other ocean missions. It needs very high precision requirements to meet underwater glider motion control. When the position of buoyancy system changes, the balance parameters will change significantly. This paper presents a method for calculating the balance parameters of underwater glider based on neuralnetwork. In order to verify the effectiveness of the neuralnetworkcontrol, the South China Sea experiment was carried out. By comparing the analysis results with the actual situation, the experiment shows that the neuralnetwork model is feasible.
The sliding mode control is robust to plant uncertainties and insensitive to external disturbances. It is widely used to obtain good dynamic performance of controlled systems. However, the chattering phenomena due to the finite speed of the switching devices can affect the system behavior significantly. Additionally, the sliding control requires the knowledge of mathematical model of the system with bounded uncertainties. Another method, popular in recent years, is based on [7-10].
It is well known that imperfect robot model will lead to degradation of tracking performance. So it is necessary to approximate the un-modeled dynamics and external disturbance. Next, the neuralnetwork and disturbance observer are employed to approximate the un-modeled dynamic and external disturbance respectively
Because of its capability to approximate arbitrary non- linear mapping, neuralnetwork has been actively used in nonlinear system identification and control [9-12]. The multilayer feed-forward neuralnetwork (MFNN) is one of the most widely used neural networks as a system model in the design of a model-based controller. A dy-
Dynamic Positioning (DP) of ships is a control mode that seeks to maintain a specific position (stationkeeping) or perform low-speed maneuvers. In this paper, a static NeuralNetwork (NN) is proposed for control allocation of an over-actuated ship. The thruster force and commands are measured during a trial run of the simulated vessel to gather data for training of the NN. Then the network is trained and used to transform the virtual force commands from a motion controller into individual thruster commands. A standard Proportional Integral Derivative (PID) controller, using wave-filtered position and heading measurements, is implemented as motion controller for each Degree Of Freedom (DOF) of the ship. For a DP application the controllable DOFs are the translational motion in surge and sway directions, as well as the rotation about its up/down axis. Simulation tests were performed to verify the feasibility of this approach.
all the other conditions, the neuralnetwork controller shows a fast response speed with low current and voltage oscillations. The standard vector controller shows similar performance for power transmission control between the two VSC stations (Fig. 10). Compared to the neuralnetwork vector controller, the standard vector controller shows higher oscillations (Figs. 9c and 10c, Figs. 9d and 10d). This is due to the fact that the control action generated by the standard PI controller is determined by the error between the control parameter and the corresponding reference value. Hence, there must be overshoot and settling time issues associated with a PI-based controller. However, the neuralnetwork controller is designed and trained based on the DP-based optimal control principle. For an ideal optimal controller, a reference command can be reached immediately without any delay and overshoot. But, this cannot be achieved practically because of physical system constraints. The neuralnetwork controller tries to approximate an ideal optimal controller within the physical system constraints. Therefore, the neuralnetwork controller has the advantages for fast power transmission control between VSC stations with small oscillations as shown by Figs. 9 and 10. B. Power Transmission and PCC Voltage Control
This paper brings out an innovative idea called as mouse for handless human (MHH) to use the camera as an alternative to the mouse. The mouse operations are controlled by the hand gesture captured by the camera Gesture Recognition Technology for Games Poised for Breakthrough on the website of Venture Bet ( http://venturebeat.com) .In the near future, for example, users will likely be able to control objects on the screen with empty hands, as shown in Figure 1.
The main problem in controlling DC motor as a position control drive is to achieve the desired position, reduce the steady-state errors and oscillation problem. Most of position controllers present several problems because of the nonlinearities system. Moreover, these nonlinearities are often unknown. These nonlinearities system affect the performance of position control of the DC motor system. The commonly used Proportional-Integral-Derivative (PID) controller is simple and easy to practice, but this controller suffers from poor performance if it has nonlinearities and uncertainties issues. The PID controller is not able to work well for non-linear system, and particularly complex and vague system that has no precise mathematical models. To overcome these difficulties, the neuralnetwork controller is developed in this project to ensure the DC motor is operable in any circumstance of situations.
300 rpm at 8.4E-3 s It’s observed that the proposed two methods of neural controller succeeded to enhance the ripples. This reduction was around 1.26% with the value of ripples equal 0.0196 N.M for the neuralnetwork, 0.46% with the value of ripples equal 0.0072 N.M for neuralnetwork based estimator for torque constant and stator winding resistance. Moreover, this result is compared to other publication for more validation. V. Petrovic et al , Torque ripple percentage was ≈ 4% which is still higher than a proposed two methods of neuralnetwork. While W. Qian and Panda , Torque ripple percentage was 3.9% which is still higher than a proposed two methods of neuralnetwork. P. Mattavelli et al. , Torque ripple percentage was ≈ 3.8% which is still higher than the proposed two methods of neuralnetwork. H.M. Hasanien et al. . Their result was 12%. Torque ripples in compression to our two methods of neuralnetwork are actually higher. M.Tarnik et al. , Torque ripple percentage was ≈ 4% which is also higher than the proposed two methods of neuralnetwork. Table 4, presents the torque ripples than percentage, which have been calculated from different works. The table shows that the proposed control reduced the percentage torque ripples to a good value rather the previous work -. We can further reach to optimize the output for torque and reduce the amount of percentage error by combining two neural networks first method, and second method.
Motivated by the above discussion, this paper puts forward an adaptive neural- sliding model compensation control scheme. A RBF neuralnetwork is used to approximate the unknown nonlinear dynamics of the robot manipulators. But the local generalization of the RBF network is considered by the paper, because accuracy of control system is effected by approach errors of neuralnetwork, Sliding model controller is designed to eliminate approach errors to improve control accuracy and dynamic features. The adaptive laws of network weights are designed to ensure adjustment online-time, offline learning phase is not need; Globally asymptotically stable(GAS) of the closed-loop system is proved based on the Lyapunov theory. The simulations show this controller can speed up the convergence velocity of tracking error, and has good robustness.
After the neuralnetwork architecture is modelled, the next stage defines the learning model to update network parameters. By this learning capability, it makes the ANN suitable to be implemented for the system with motor parameters which are difficult to define and vary against with environment. The training process minimizes the error output of the network through an optimization method. Generally, in learning mode of the neuralnetwork controller a sufficient training data input-output mapping data of a plant is required. Since the motor parameters of the induction motor drive vary with temperature and magnetic saturation, the online learning Back propagation algorithm is developed. Based on the first order optimization scheme, updating of the network parameters is covered. The performance index sum of square error is given by