Prem Kumar Kalra

Load frequency control: a generalised neural network approach

Abstract Variation in load frequency is an index for normal operation of power systems. When load perturbation takes place anywhere in any area of the system, it will affect the frequency at other areas also. To control load frequency of power systems various controllers are used in different areas, but due to non-linearities in the system components and alternators, these controllers cannot control the frequency quickly and efficiently. Simple neural networks which are in common use at present have various drawbacks like large training time, requirement of large number of neurons, etc. The present work deals with the development of a non-linear neural network controller using a generalised neural network. The drawbacks of existing neural networks have been overcome in the generalised neuron structure which has been developed to control the deviations in load frequency of a power system.

Time series prediction with single multiplicative neuron model

Single neuron models are typical functional replica of the biological neuron that are derived using their individual and group responses in networks. In recent past, a lot of work in this area has produced advanced neuron models for both analog and binary data patterns. Popular among these are the higher-order neurons, fuzzy neurons and other polynomial neurons. In this paper, we propose a new neuron model based on a polynomial architecture. Instead of considering all the higher-order terms, a simple aggregation function is used. The aggregation function is considered as a product of linear functions in different dimensions of the space. The functional mapping capability of the proposed neuron model is demonstrated through some well known time series prediction problems and is compared with the standard multilayer neural network.

Improved generalized neuron model for short-term load forecasting

The conventional neural networks consisting of simple neuron models have various drawbacks like large training time for complex problems, huge data requirement to train a non linear complex problems, unknown ANN structure, the relatively larger number of hidden nodes required, problem of local minima etc. To make the Artificial Neural Network more efficient and to overcome the above-mentioned problems the new improved generalized neuron model is proposed in this work. The proposed neuron models have both summation (Σ) and product (π) as aggregation function. The generalized neuron models have flexibility at both the aggregation and activation function level to cope with the non-linearity involved in the type of applications dealt with. The training and testing performance of these models have been compared for Short Term Load Forecasting Problem.

Perception and remembrance of the environment during real-time navigation of a mobile robot

Abstract This paper deals with the advantages of incorporating cognition and remembrance capabilities in a sensor-based real-time navigation algorithm. The specific features of the algorithm apart from real-time collision avoidance include spatial comprehension of the local scenario of the robot, remembrance and recollection of such comprehended scenarios and temporal correlation of similar scenarios witnessed during different instants of navigation. These features enhance the robot’s performance by providing for a memory-based reasoning whereby the robot’s forthcoming decisions are also affected by its previous experiences during the navigation apart from the current range inputs. The environment of the robot is modeled by classifying temporal sequences of spatial sensory patterns. A fuzzy classification scheme coupled to Kohonen’s self-organizing map and fuzzy ART network determines this classification. A detailed comparison of the present method with other recent approaches in the specific case of local minimum detection and avoidance is also presented. As for escaping the local minimum barrier is concerned this paper divulges a new system of rules that lead to shorter paths than the other methods. The method has been tested in concave, maze-like, unstructured and altered environments and its efficacy established.

Image registration using robust M-estimators

In this paper, a method for robust image registration based on M-estimator Correlation Coefficient (MCC) is presented. A real valued correlation mask function is computed using Huber and Tukeys robust statistics and is used as a similarity measure for registering image windows. The mask function suppresses the influence of outlier points and makes the registration algorithm robust to noisy pixels, brightness fluctuations and presence of occluding objects. The superiority of the proposed algorithm, in terms of registration performance and computation time is demonstrated through experimental studies on different types of real world images.

Detection, Tracking and Avoidance of Multiple Dynamic Objects

Real-time motion planning in an unknown environment involves collision avoidance of static as well as moving agents. Strategies suitable for navigation in a stationary environment cannot be translated as strategies per se for dynamic environments. In a purely stationary environment all that the sensor can detect can only be a static object is assumed implicitly. In a mixed environment such an assumption is no longer valid. For efficient collision avoidance identification of the attribute of the detected object as static or dynamic is probably inevitable. Presented here are two novel schemes for perceiving the presence of dynamic objects in the robots neighborhood. One of them, called the Model-Based Approach (MBA) detects motion by observing changes in the features of the environment represented on a map. The other CBA (cluster-based approach) partitions the contents of the environment into clusters representative of the objects. Inspecting the characteristics of the partitioned clusters reveals the presence of dynamic agents. The extracted dynamic objects are tracked in consequent samples of the environment through a straightforward nearest neighbor rule based on the Euclidean metric. A distributed fuzzy controller avoids the tracked dynamic objects through direction and velocity control of the mobile robot. The collision avoidance scheme is extended to overcome multiple dynamic objects through a priority based averaging technique (PBA). Indicating the need for additional rules apart from the PBA to overcome conflicting decisions while tackling multiple dynamic objects can be considered as another contribution of this effort. The method has been tested through simulations by navigating a sensor-based mobile robot amidst multiple dynamic objects and its efficacy established.

On Efficient Learning Machine With Root-Power Mean Neuron in Complex Domain

This paper describes an artificial neuron structure and an efficient learning procedure in the complex domain. This artificial neuron aims at incorporating an improved aggregation operation on the complex-valued signals. The aggregation operation is based on the idea underlying the weighted root power mean of input signals. This aggregation operation allows modeling the degree of compensation in a natural manner and includes various aggregation operations as its special cases. The complex resilient propagation algorithm (C-RPROP) with error-dependent weight backtracking step accelerates the training speed significantly and provides better approximation accuracy. Finally, performance evaluation of the proposed complex root power mean neuron with the C-RPROP learning algorithm on various typical examples is given to understand the motivation.

Experimental studies with a generalized neuron-based power system stabilizer

Artificial neural networks (ANNs) can be used as intelligent controllers to control nonlinear, dynamic systems through learning, which can easily accommodate the nonlinearities and time dependencies. However, they require large training time and large number of neurons to deal with complex problems. To overcome these drawbacks, a generalized neuron (GN) has been developed that requires much smaller training data and shorter training time. Taking benefit of these characteristics of the GN, a new power system stabilizer (PSS) is proposed. Results show that the proposed GN-based PSS can provide a consistently good dynamic performance of the system over a wide range of operating conditions.

Curve and surface reconstruction from points: an approach based on self-organizing maps

Abstract Modeling of shapes for free form objects from point cloud is an emerging trend. Recognition of shape from the measured point data is a key step in the process of converting discrete data set into a piecewise smooth, continuous model. Shape recognition is to find the topological relation among the points, and in case of thick unorganized point cloud, the step requires both thinning and ordering. The present paper outlines a new approach based on growing self-organizing maps (GSOM) for piecewise linear reconstruction of curves and surfaces from unorganized thick point data. Inferences on selection of self-organizing map (SOM) algorithm parameters for this problem domain have been derived after extensive experimentation. A better quality measure to evaluate and compare various runs of SOM for the domain of curve and surface reconstruction has also been presented.

Some new neural network architectures with improved learning schemes

Abstract Here, we present two new neuron model architectures and one modified form of existing standard feedforward architecture (MSTD). Both the new models use self-scaling scaled conjugate gradient algorithm (SSCGA) and lambda–gamma (L–G) algorithm and entail the properties of basic as well as higher order neurons (i.e., multiplication and the aggregation functions). Of these two, compensatory neural network architecture (CNNA) requires relatively smaller number of inter-neuronal connections, cuts down on the computational budget by almost 50% and speeds up convergence, besides, gives better training and prediction accuracy. The second model sigma–pi–sigma (SPS) ensures faster convergence, better training and prediction accuracy. The third model (MSTD) performs much better than the standard feedforward architecture (STD). The effect of normalizing the outputs for training also studied here shows virtually no improvement, at low iteration level, say ∼500, with increasing range of scaling. Increasing the number of neurons beyond a point also shows to have little effect in the case of higher order neuron.The numerous simulation runs for the problem of satellite orbit determination and the complex XOR problems establishes the robustness of the proposed neuron models architectures.