Method To Predict Cementing Quality With Method Of Gray Neural Network

There are many factors affecting the cementing quality, relating various aspects of cementing operations, it is not the single factor affecting the cementing quality to decide the cementing quality, but it is the result of the interaction of the various factors. So, factors affecting the cementing quality are analyzed more subtly, comprehensively and systematically in this paper, and the prediction model of cementing quality is established to forecast the cementing construction plan, to provide theoretical basis for the plan, to provide guarantee for the good cementing quality.Firstly, Three main factors affecting the cementing quality——geological conditions, drilling and cementing technology are expounded in this paper. The relationship between factors affecting the cementing quality and cementing quality is built by statistical analysis. Secondly, based on the field data, the main factors affecting the cementing quality are found by, which are used for the input parameters of the prediction model of cementing quality. Then, the GM(0,5) grey prediction model of cementing quality is built, in which cementing quality is the mother sequence, the main factors affecting the cementing quality which get by the method of grey relational analysis are the child sequences; the BP neural network prediction model of cementing quality is built, in which the cementing quality is the output parameter, the main factors affecting the cementing quality are the input parameters; but, because the stimulant result of grey prediction model of cementing quality is not good, and in the BP neural network prediction model of cementing quality, convergence speed is slow, training time is long, and the learning efficiency is low. the complementary and the possibilities of combination between the two models are analyzed in this paper, the cementing quality prediction model combined the theories of grey system and artificial neural network is built, the model adds a gray layer of cementing quality data in the front of the input layer of BP neural network to accumulatively calculate cementing quality data, by which the randomness of the original data is weakened; and the model adds a white layer of output result to the output layer of the BP neural network to get the predicted value, the forecast accuracy of the model is improved, and the learning speed and generalization ability of the cementing quality prediction model are improved.