A Neuro-Computing Based Model for Anomaly Recognition in Geochemical Exploration

Mansour Ziaii

Faculty of Mining Engineering and Geophysics Shahrood University of Technology

IRAN

Ali A. Pouyan, member IEEE Faculty of ICT and Computer Engineering

Shahrood University of Technology IRAN

Mehdi Ziaii, member IASP

Semnan Science and Technology Park Shahrood, IRAN

Abstract: - In this paper an abstract model for anomaly recognition in geochemical exploration has been developed based on neural networks. Traditional geochemical exploration methods are based on multivariate statistical analysis, which suffer several shortcomings including lack of geo-statistical generalized approach for separating anomalies from background. These shortcomings make the interpretation process time consuming and costly. We have proposed a novel approach for quantitatively recognition between blind anomalies and false anomalies’ patterns using back propagation artificial neural networks with fuzzy C-means cluster analysis. It has been revealed that the resultant output has been significantly improved, comparing traditional methods. The major advantage of the proposed method is that it computationally enables us to distinguish zone of dispersed ore mineralization from blind mineralization without exploration drilling. Key-Words: - Neural networks, anomaly, dispersed mineralization, pattern recognition, exploration, zonality 1 Introduction A major aspect to be considered in the interpretation of secondary halos is the erosion level of the mineral deposit as it affects the size and extent of anomalies in soils. This point has been conceptualized in different paradigms known as model of blind economic mineralization, model of outcropped economic mineralization and zone dispersed mineralization (ZDM). Soil anomalies associated with outcropped economic mineralization would be stronger than the economic mineralization and this may be erroneously assumed to be more promising than others unless the erosion levels are taken into account. Soil anomalies based on ZDM model might well be similar in intensity to that associated with the blind mineralization. And if it is not properly interpreted, it could lead to fruitless exploration. Root zones (ZDM) of some types of ore

deposits typically have a different metal association from the ore zone and leakage (upper) zones [1], [6], [7]. These associations may be helpful in identifying the relationship of a soil anomaly to mineralization [5], [4]. Characterizing horizons of erosional surfaces (HES) of a steeply dipping ore body and its primary halo in enclosing rocks is a problem with no direct and known solution. To date, there are only two generally reliable ways of acquiring knowledge on HES. These are laboratory measurements and vertical zonality coefficient interpretation. Laboratory measurement of the cores obtained from the field or core archives provides precise (assuming adequate equipment) vertical zonality coefficient of values [14]. These are used in geochemical simulation studies as well as any other design and development studies on the field. The other method for HES determination is geochemical

Proceedings of the 10th WSEAS International Conference on SYSTEMS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp98-102)

model of mineralization in bedrock and soils. However, geochemists outside of Russia do not appear to have the expertise necessary to make such interpretation [8]. In this paper, we have introduced a new method for HSE determination. The proposed method, which is based on artificial neural networks is quite inexpensive comparing with the traditional methods. Furthermore, it does not require production interruption and provides vertical zonality coefficient values. These values are comparable to those obtained by laboratory measurements of cores. A feasibility study based on the proposed method for HES estimation, shows effective and useful results. In has been demonstrated that in porphyry copper mineralization, application of a carefully designed and elaborated neuro-fuzzy model can provide acceptable results. 2 Study Area The study area is located in northwest Iran and located 75 km northwest of Ahar (Sungun and Astamal valley). The Sungun Cu porphyry–skarn deposit and Astamal Cu porphyry mineralization are located in the Cu metallogenic zone of Ahar [12], [13]. There are many ore deposits in this region both in Iran [2] and Armenia. Ahar region is part of the Alpine Eurasian metallogenic belt which extends from Greece towards Iran. The porphyry, skarn and vein-type mineralization in this region are attributed to the Pyrenean orogenic phase [3]. At Sungun, as shown in Fig. 1, the Oligo-miocene subvolcanic plutons which form part of the Cenozoic magmatic belt of Iran, have intruded Oligocene acid-intermediate volcanic rocks, Eocene terrigenous sediments and Cretaceous limestone. The plutonic rocks of Sungun (with average composition of monzodiorite) are equivalent to the volcanic rocks. The Sungun deposit shows features of both porphyry and skarn mineralization. The Sungun stock is a complex intrusive body which crops out over an area of about 1.5 by 2.3 km. the stock consists The formation of the Astamal porphyry is due to the intrusion of the Astamal pluton, of Oligocene age into Cretaceous limestone. The average composition of this pluton is granodiorite. It is part of the Shavardagh Batholith which has formed many ore deposits north of Ahar.

Fig. 1. Geological map of Sungun copper deposit (NW-Iran)

3 Methodology In this section we first explain the traditional methods used for HES estimation. Then we will discuss the proposed approach for quantitatively recognition between blind anomalies and false anomalies’ patterns using back propagation artificial neural networks with fuzzy C-means cluster analysis. 3.1 Traditional method Calculating vertical zonality coefficient from primary geochemical halos has been practiced since drilling stage of exploration became available. These calculations assume a linear or non-linear model of the relationship between vertical zonality coefficients and depth of mineralization responses. It should be mentioned that we use linear or non-linear model to emphasize that in these calculations it is assumed that a known function (linear or non-linear) is sufficient for modeling the relationship between these geochemical halos of parameters and the aforementioned depth mineralization responses.

The primarily method for HES determination is geochemical model of copper porphyry mineralization in bedrock and soils in the Sungun deposits. As it is shown in figure 3, the graph of multiplicative vertical geochemical elements

Proceedings of the 10th WSEAS International Conference on SYSTEMS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp98-102)

distribution pattern in ores and primary halos (geochemical modeling) from three porphyry – copper deposits is practically a straight line with a mineral scatter of observation points.

Fig. 2. Geochemical model for porphyry Copper deposits

The model shown in Fig. 2 has been constructed

for the Aktogy (Kazakhstan), Asarel (Bulgaria) and Tekhut (Armenia) porphyry-copper deposits. Despite of considerable difference among their geological locations, the points that are strictly placed on the straight line suggest the existence of quantitatively uniform vertical geochemical zonality structure of primary halos at the deposits of this ore formation. It can be deduced from Fig. 2 that such a zonality implies the same levels (upward and downward) of mineral deposits and halos of a given ore formation. Such halos of a given ore formation are characterized by strictly definite vertical geochemical zonality coefficient values.

The practical significance of this quantitatively uniform geochemical zonality will be evident if we bear in mind that it makes it possible to evaluate a level of erosion cut of any geochemical anomaly in a given formational type; like a porphyry-copper one in this particular case. A gradient characterizing of the vertical zonality coefficient allows us to reliably differentiating among various mineralization and its primary halos: supra-ore, upper-ore, min-ore, lower-ore, and under-ore.

The above-described quantitatively uniform geochemical zonality of porphyry-copper mineralization based on the results of the detailed

geochemical modeling of only three deposits was used in interpretation of geochemical sampling results in the Sungun ore district. Using the traditional method resolves the known problems of exploration of blind mineralization and identification of zones of dispersed ore mineralization (ZDM) in the field of geochemical exploration. However, traditional method of multivariate statistical analyses, used to recognize anomalous mineralization in mining geochemistry, have several shortcomings: (1) it is difficult to isolate anomalies where the data are not normally distributed, (2) it is difficult to separate distinct anomaly populations corresponding to distinct formation mechanisms while separating anomalies from background, (3) it can not be used to present illustrations of multivariate anomalies on contour maps, and (4) it is not suitable for preparation of multivariate pattern recognition. 3.2 The Proposed Neuro-Fuzzy Model Neuro-fuzzy techniques can be considered as a hybrid discipline between neural networks and fuzzy logic, not only bring out the best of the both techniques, they also facilitate a comprehensive sensitivity analysis common with neural network without going into elaborate sensitivity analysis associated with fuzzy logic. Neuro-fuzzy modeling is an approach for the fusion of neural networks and fuzzy logic modeling. These two modeling techniques complement each other. The neuro-fuzzy approaches employ heuristic learning strategies derived from the domain of neural network theory to support the development of a fuzzy system. It is possible to completely map neural network knowledge to fuzzy logic [9]. A combination of neural network and fuzzy logic techniques should help overcome the shortcomings of both techniques [11]. Neuro-fuzzy techniques can learn a system s behavior from a sufficiently large data set and automatically generate fuzzy rules and fuzzy sets to a pre-specified accuracy level. Also, they are capable of generalization, thus overcoming the key disadvantages of the fuzzy logic-based approaches, viz., self-learning, inability to meet pre specified accuracy, and lack of generalization capability [10]. The fuzzy c-means (FCM) algorithm partitions a data set into a predefined c number of clusters. It is a data clustering technique wherein each data point belongs to a cluster to some degree that is specified by a membership grade. It provides a method that shows how to group data points that populate some multidimensional space into a specific number of

Proceedings of the 10th WSEAS International Conference on SYSTEMS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp98-102)

different clusters. Clustering is a mathematical tool that attempts to discover structures or certain patterns in a data set, where the objects inside each cluster show a certain degree of similarity. A more detailed discussion of FCM and examples is given [9], [10]. In this study, a four-layer neuro-fuzzy network has been considered, as shown in Fig. 3. The nodes of the first layer represent the inputs. The activation functions of the second layer nodes and the third layer acts as a hide node so that input layer provides the fuzzy rule (FCM) base. The output of this layer determines the activation level at the output memberships. As ordinary neural nets, the neuro-fuzzy one learns from a training data set, Tansig functions and rules, by means of a back-propagation artificial neural network (BP-ANN) algorithm. Tansig is a neural transfer function. Transfer functions calculate a layer's output from its net input.

Fig. 7. The proposed neuro-fuzzy network architecture

Use of neuro-fuzzy model requires similar steps as neural networks. Development of a neuro-fuzzy model is comprised of three steps: learning, validation, and application. The entire data sets were divided into three groups: learning, validation and application. During the learning step the neuro-fuzzy networks were provided with various combinations of data for pattern recognition purposes and the neuro-fuzzy network modified its internal representation by changing the values of its weights to improve the mapping of inputs to outputs relationships. During the validation step, the network was fed a set of data as new input, and the net mapped the inputs to output relationships based on previously learned patterns without its weights.

Once the learning and validation steps were completed, the application data, which were much larger than the learning and validation data sets, were used to generate geochemical exploration blind mineralization from ZDM. This approach was established on the basis of geochemical characteristics and origin of the various populations in geochemical data, such as background, blind anomalies and ZDM anomalies (false anomalies). The topology of the BP-ANN with FCM was optimized using the outputs of the BP-ANN and the correct rate. As shown in Fig. 3, the output of the BP-ANN is in the form of two distinct types of false anomalies (False) and blind mineralization (True). The application of this method does not need any information related to input data distribution. 4 Conclusion In this study, we have proposed a model for anomaly recognition in geochemical exploration based on fuzzy neural networks. Traditional geochemical exploration methods are based on multivariate statistical analysis, which suffer several shortcomings including the need for information regarding the input data distribution and lack of geo-statistical generalized approach for separating anomalies from background. These shortcomings make the interpretation process time consuming and costly. The proposed approach for quantitatively recognition between blind anomalies and false anomalies’ patterns using back propagation artificial neural networks with fuzzy C-means cluster analysis. It has been shown that the resultant output has been significantly improved, comparing traditional method. The major advantage of the proposed method is that it computationally enables us to make distinction between zone of dispersed ore mineralization and blind mineralization without exploration drilling. References: [1] Solovov, A.P., Geochemical prospecting for

mineral deposits, Mir, Moscow, 1987, 288pp. [2] Karimzadeh S.A., Garnet composition as an

indicator of Cu mineralization: evidence from skarn deposits of NW Iran, Journal of Geochemical Exploration Vol.81, 2004, 47-57.

[3] Bazin, D., Hubner, H., Copper deposits in Iran: Report No.13. Geological Survey of Iran. 1969, 29pp.

Proceedings of the 10th WSEAS International Conference on SYSTEMS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp98-102)

[4] Beus A.A., Grigorian S.V., Geochemical Exploration Methods for mineral Deposits, USA, 1997, p.287, Applied Publishing Ltd.

[5] Grigorian S.V, Mining Geochemistry. Nedra Publishing House, Moscow, pp.294. (in Russian)

[6] Ovchinnikov L.N, Grigorian S.V. Geochemical Prospecting for Ore Deposits. Geology Rev .V.20, N12, 1978.

[7] Grigorian S.V ,Geochemical Prospecting for Hidden Ore Deposits .Moscow , 1992 (in Russian)

[8] Levinson A.A., Introduction to exploration geochemistry. .Second Edition applied publishing Ltd. Wilmette, USA. (The 1980 supplement).

[9] Bezdek, J.C. Pattern Recognition with Fuzzy objective Function Algorithms, Plenum, New York, 1981.

[10] Bezdek,J.C., R.Ehrich, and W.Full (1984), FCM: The fuzzy c-means clustering algorithm, Comput. Geosci. 10,191-203.

[11] Dixon B., Applicability of neuro-fuzzy techniques in predicting ground-water vulnerability: a GIS-based sensitivity analysis, Jurnal of Hydrology 309(2005)17-38.

[12] Ziaii M., Lithogeochemical exploration methods for cupper-porphyry deposit in Sungun, NW Iran, Unpublished M.Sc. thesis. Geochemistry faculty of MSU, Moscow, 1996.

[13] Ziaii M., Application of GIS technology in regional exploration programs, II international conference" GIS in Geology" 2004,

[14] Grigorian S.V, Ziaii M., Computing methods for determination of geochemical haloes background, international symposium "Applied Geochemistry" in CIS, 1997(in Russian).

Proceedings of the 10th WSEAS International Conference on SYSTEMS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp98-102)