Analysis of the Effect of Rate of Extraction on Strain Development in Interaction Subsidence Data Using a Neural Network

In 1993 an opportunity presented to monitor the effects of undermining an abandoned roadway by a longwall panel in the Sydney. N.S. coal field. Interaction subsidence data was collected by survey leveling to benchmarks in the undermined gateroad as well as by a standpipe system equipped with 15 vibrating wire piezometers measuring changes in elevation in the gateroad relative to an outby header tank. The piezometers were monitored at 30 minute intervals using a datalogger. while the leveling surveys were conducted weekly over the life of the project. The monitoring program resulted in the collection of a very large both of data. relating vertical subsidence to the relative position of the underlying face. The most basic analyses derived a travelling subsidence curve from the leveling data only. which hinted at a relationship between rate of extraction in the panel below and the amount of strain developed in the undermined strata. The complexity of the relationships between the amount of subsidence obtained, the location of' the individual monitoring stations, the rate of face advance below and the potential for "periodic" subsidence due to geological conditions has made analysis difficult. and the amount of data has compounded this problem. Recently. an analytical approach using a commercial neural network analysis package has been applied. which appears to have resulted in the ability to draw sonic meaningful conclusions from the data set. In this paper a short history regarding the collection of the data set is followed by a brief' description of neural networks. This is followed by a description of the several analyses performed. some of the results obtained, and conclusions drawn.