Well-to-Well Log Correlation Using Knowledge-Based Systems and Dynamic Depth Warping

Abstract

We present a novel system for well-to-well log correlation using knowledge-based systems and dynamic depth warping techniques. This approach overcomes a major drawback inherent in previous methods, namely the difficulty in correlating missing or discontinuous rock units. The system has three components: (1) A Dynamic Programming algorithm to correlate the logs and to find the minimum-cost or "best" match; (2) A set of "rules" to guide the correlation; (3) A data base that contains the logs and other relevant geologic and seismic information. The Dynamic Programming algorithm calculates the cost of correlating each point in the first well with each of the points in the second well. The resulting matrix of dissimilarity contains cost information about every possible operation which matches the well logs. The cost of matching the two wells is measured by the difference in the log values. The dynamic programming approach allows correlation across geologic structures, thinning beds, and missing or discontinuous units. A path finding algorithm then traces through the matrix to define a function which maps the first well onto the second. The minimum cost path is the optimal correlation between the wells. The system's database contains the well logs themselves and other relevant data including information about the geologic setting, seismic ties, interpreted lithologies, and dipmeter information. Rules operating on the data affect the dynamic programming and path finding algorithms in several ways: (1) Seismic ties or marker beds define a point in the warping path, thereby removing calculations over large portions of the search space; (2) Dipmeter results and knowledge of geologic structure further constrain the path to certain global areas and save calculation time; (3) The system assigns weights to different logs based on log quality and sensitivity; (4) Knowledge of the paleoenvironment allows the program to choose a set of rules (model) which accounts for changes in sediment type or thickness within a field. For example, when the program is operating in a deltaic environment, it will correlate the shales before attempting to correlate the sands. We demonstrate the method with synthetic examples in which the program successfully correlates across geologic structures and pinch-outs. We also applied the program to field examples from two widely separated oil provinces. In both cases, the automated correlation agreed very well with correlations provided by geologic experts.