The authors have declared that no competing interests exist.

Electrical resistivity method is often used to estimate the subsurface structure of the earth. Many inversion algorithms are available to estimate the subsurface features. However, predicting the exact parameter in the non-linear subsurface of the earth is difficult because of its complex composition. Soft computing tools can approximate the subsurface parameters more clearly. Each soft computing tool has certain advantages and disadvantages. A hybrid formation of algorithms will make the decision more appropriate than depending on a single tool. Here in our study the data obtained through Vertical Electrical Sounding has been used to determine the sub surface characteristics of earth viz., true resistivity and thickness. Artificial Neural Networks (ANN) requires certain optimizing procedures. Here in this paper, Genetic Algorithm (GA) is applied to optimize Artificial Neural Networks (ANN). This coupled approach is tested with the field data. Error percentage of algorithm nearly mimics the behavior of earth and is verified. The best performance result shows that this technique can be implemented to estimate the non-linear characteristics of the earth more noticeably.

In recent times, people moved towards advanced techniques to model the complex behavior system associated with nonlinearity, high-order dynamics, time-varying behavior, and imprecise measurements.

One of such non-linear characteristic behavior that we know is the field of Earth Sciences. Because of the complex structural behavior of rocks and soil formations, it is difficult to fix each parameter in a certain range. So the efficient soft computing tool is needed to estimate the subsurface parameters of the earth that are pertinent to the real world.

Geophysical inverse problems atmost are nonlinear in nature and pocessess irregular objective function which is of multimodel type

In recent years, Genetic Algorithm plays a significant role in optimizing various geophysical problems which includes Seismic waveform inversion , Marine Electromagnetic data inversion , Inversion of data obtained through surface waves and that have been proved by many researchers.

Here in our study Artificial Neural Networks (ANN) and Genetic Algorithm (GA) were combined to perform a best unification algorithm for predicting the subsurface features of the earth. Neural networks can approximate the result based on the training. It can exhibit mapping capabilities easily between the input and output patterns. Here synthetic apparent resistivity data were trained with ANN algorithm and tested with the field data. As ANN can learn by examples, the training dataset was developed synthetically and tested. The layer model provides the information about the true resistivity and the thickness of the subsurface layer with error percentage.

Genetic algorithm (GA) is a search algorithm based on the mechanics of natural selection and natural genetics. The central theme of research on genetic algorithms has been robustness, the balance between efficiency and efficacy necessary for survival in many different environments. It is computationally powerful tool that seeks to reproduce mathematically the mechanics of natural selection and natural genetics, according to the biological processes of survival and adaption

Many applications were done on the recent years based on artificial intelligent techniques tocontrol, prediction and inference studies

Genetic Algorithm coupled Neural Networks (GA-ANN)approaches the combined behavior of genetically optimized neural network system. Input data are categorizing to seek the objective function using genetic algorithm which is one of the best global search optimization technique can able to find the global optimum minimum. They can be used to modeling or control linear or non-linear systems using ANN. ANN can generate the synthetic data required for geoelectrical inversion. GA searches the best-fit optimization generated the synthetic data and interpretation was done by the ANN model. The best-fitting individual of the population in GA will predict the output with less error percentage

Our Study area comprises of Rajapalayam-Alangulam Belt of Virudhunagar District , Tamil Nadu State in Indian subcontinent. This area lies in the southern part of the state in the Eastern slopes of Western Ghats. The Study area is endowed with various mineral and water resources. To study the layered structure of the Earth with the delineation of aquifer parameters , the major problem that has to be optimized in this area is the mixed and complex nature of geological formation in that area . Vertical Electrical Sounding method provides a good solution for that. A wide variety of optimization techniques have been used to interpret Vertical Electrical Sounding(VES) data of geophysical prospecting but this coupled approach of GA-ANN shows better results in obtaining information regarding layered structure of the Earth

Geophysical exploration techniques are vibrant and powerful tools that play a vital role in the delineation of aquifer parameters in different geological formations. In particular, The Geophysical method consisting of vertical electrical sounding (VES) survey has been proved to know the variation of resistivity of the aquifer parameters

Artificial neural network can play a major role to dig out the mystery of earth science by means of the previous learned examples. These networks have self-learning capability and are fault-tolerant as well as noise-immune and have applications in various fields

A very important feature of these networks is their adaptive nature, whose ‘learning by example’ replaces “programming” in solving problems. This feature makes such computational models very appealing in application domains where one has little or incomplete understanding of the problem to be solved but where training data is readily available. ANNs are now being increasingly recognized in the area of classification and prediction, where regression model and other related statistical techniques have traditionally been employed. Introducing the layer of neurons involved in the training is shown in (

Here the sigmoid function is used as activation function. It can produce outputs with reasonable discriminating power and its output functions are differentiable, which is essential for the back propagation of errors

f(x)= 1/1-e^{-x}

F

The number of neurons in the hidden layer varies as per the requirement of optimum performance, which could be decided on trial and error basis. Initial weights are assigned at random in the range suitable for the activation function at neurons

In this paper an iterative non-linear feed forward network backpropagation algorithm and the performance was tested.

Various optimization problems that can be solved by the method of Genetic Algorithm (GA), which is based on natural selection particularly of biological evolution. GA is generally mimicking the process of natural evolution, which generates useful solutions for optimization problems using techniques inspired by natural evolution, such as inheritance, mutation, selection and crossover

To ensure their survival, the respective individual Organism must be over reproductive.

1. In the next generation, some minor stochastic changes are seen.

2. The more successful will be better adaptive individuals which will be able to pass their genes to the successive generations.

To optimize a particular problem GA has five major steps, that are as follows;

A random initial population of the problem that has to be optimized is generated by the means of random generator. Initial population is represented by the set of parameter values which describes the problem. These set of parameters are then transformed into a string of binary bits (0’s and 1’s) of specific length. At each and every generations, the chromosomes are tested for fitness using fitness function which is achieved by decoding binary strings into set of parameter values

By the means of various ranking schemes which results that good chromosomes have higher chances to be in the next generation. Rank selection schemes performs the operation over the population on the basis of fitness values. In particular over various ranking schemes tournament selection is preferable

The process of cross over and mutation is performed over existing generation to create new generation which is based on crossover and mutation probability.

The new generation is formed by replacing the currently existing generation. The children of the successive generations are decided based upon the fitness value through fitness function provided.

The process of GA stops when at least one among the various stopping criteria such as Maximum number of generations, Maximum time limit, Maximum fitness limit etc., met.

Conventional search methods are not robust. The robustness behavior of GA has been used successfully in many applications. Far more desirable performance would be obtained in this Robust Scheme. Simple genetic algorithm that yields good results in many practical problems is composed of three operators.

1. Reproduction

2. Crossover

3. Mutation

Reproduction is a process in which individual strings are copied according to their objective function values. The operator, of course, is an artificial version of natural selection, a Darwinian survival of the fittest among string creatures. In natural populations fitness is determined by a creature’s ability to survive predators, pestilence, and the other obstacles to adulthood and subsequent reproduction. In our case the apparent resistivity data has been chosen. The best fit optimized true resistivity data was the output that better survives in the selection operator. Reproduction process with individual strings is on the basis of apparent resistivity data. The dataset provides the necessary population generation needed for the natural selection of the algorithm.

Genetic Algorithm (GA) description:

step1: Generation = 0;

step2: Initialize M (Generation);

Evaluate M (Generation);

step3: While (GA has not converged or terminated)

Generation = Generation + 1;

Select M (Generation) from M (Generation - 1);

Crossover M (Generation);

Mutate M (Generation);

Evaluate M (Generation);

End (while)

step4: Terminate the GA.

Many optimization algorithms can be used to interpret the Vertical Electrical Sounding data. But most of the optimization algorithm doesn't behave in the way to represent the exact lithological features. But when the optimization algorithm concepts are integrated, the performance of the architecture has been improved. One of such method is integrating the GA and ANN. Many works based on Geoelectrical resistivity data has been going on to converge the interpreted results to a unique solution. Earlier in 1930’s many researchers investigated the inverse problem

Since the last decades the optimization techniques, especially global optimization techniques, are widely used for the inversion of resistivity data and other geophysical related problems.

Sen et al. in 1993

Stations | Location | Conventional Method | Genetically optimized ANN | ||

True Resistivity | Depth | True Resistivity | Depth | ||

Station 1 | Longitude 9.39494 | 68 | 1.5 | 27.9 | 2 |

Latitude 77.611 | 53.8 | 22.4 | 44.46 | 10 | |

936 | 48.35 | 25 | |||

71.59 | 30 | ||||

82.42 | 50 | ||||

405.5 | 60 | ||||

120.8 | 80 | ||||

148.36 | |||||

Station 2 | Latitude 77.6463 | 20.9 | 1.24 | 4.7 | 2 |

Longitude 9.37541 | 40.3 | 9.29 | 30.33 | 10.5 | |

218 | 37.09 | 30 | |||

105.83 | |||||

Station 3 | Longitude 9.36231 | 15 | 4.53 | 132.47 | 2 |

Latitude 77.6926 | 49.4 | 14.4 | 86.21 | 15 | |

253 | 85.71 | 30 | |||

96.16 | 45 | ||||

106.76 | 50 | ||||

128.1 | 80 | ||||

168.5 | |||||

Station 4 | Longitude 9.47087 | 32.5 | 1.61 | 32.8 | 2 |

Latitude 77.6269 | 57.9 | 21.8 | 65 | 25 | |

426 | 366 |

For the field data validation, VES data collected from Rajapalayam-Alangulam belt has been chosen for evaluating the performance of GA coupled ANN algorithm.

Thus, on summarizing hybrid GA-ANN model always surpasses other algorithms, especially the converging technique. The predicted accuracy of GA-ANN expresses best performance result of more than 90%. Finally, the output layer parameters were modeled. In this model, it converges more quickly and reduces the complexity involves in the non-linear problem. Specialized behavior of GA-ANN algorithm proved that this can be applied to complex geological data and the results are satisfactory and also it can be employed to all kinds of non-linear geophysical inversion problems.

GA-ANN coupled algorithm addresses the geoelectrical data inversion using hybrid approach. In this study, optimization of neural networks is done by genetic algorithm and the local convergence is made using multi node neural networks. On comparing with the traditional approach, this hybrid algorithm has a suitability to predict the subsurface layer parameters effectively. Errors and noises are easily removed through this hybrid approach because of the optimization of objective function of neural networks is made by genetic algorithm.