Multivariate drilling time sequence prediction method based on mixed leakage integral CRJ network

Abstract

The invention discloses a multivariate drilling time sequence prediction method based on a mixed leakage integral CRJ network. The method comprises the following steps of firstly, obtaining a time sequence data sample and performing data preprocessing on the data sample; secondly, selecting a high-correlation variable of a prediction variable as a network input by using a grey correlation algorithm; then, optimizing a traditional CRJ network model; combining a hybrid leakage integral neuron with stronger memorability mutually with a CRJ network to improve the dynamic characteristic and the prediction performance of the network; obtaining an optimal combination mode and an optimal leakage rate through a comparison experiment; constructing the mixed leakage integral CRJ network model based on an experiment result; and finally, using the trained mixed leakage integral CRJ network for carrying out time sequence prediction on the key variables of the drilling process. through the predictionresult, the change condition of the parameters are known in advance, and a corresponding adjustment strategy can be adopted in advance, and therefore it is ensured that drilling engineering is carried out safely and efficiently.

Description

technical field [0001] The invention relates to the technical field of drilling time series prediction, in particular to a multivariate drilling time series prediction method based on mixed leakage integral CRJ network. Background technique [0002] Complex systems widely exist in many fields such as meteorology, hydrology, industry, and information science. They have multi-variable dynamic evolution behaviors and multi-level structures, and most of them present complex characteristics. It is usually difficult to obtain accurate analytical models. Therefore, it is of great practical significance to use data-driven technology to predict the multivariate complex time series observed in the system and analyze the evolution mechanism of the system. [0003] In the establishment of time series forecasting models, neural network models can approximate nonlinear functions with arbitrary precision, and only need less statistical knowledge to obtain ideal forecasting results, so they...

AI Technical Summary

Problems solved by technology

The traditional CRJ network uses a single hyperbolic tangent neuron, the model memory ability is weak, and it is difficult to meet the requirements of complex multivariate time series prediction tasks

Images