Transfer prediction of underground pressure for fully mechanized mining face based on MRDA-FLPEM integrated algorithm

Based on the working resistance monitoring data of hydraulic support, the perceptual prediction of strata behavior in the fully mechanized mining face is an effective means to realize the early warning and early response of periodic pressure on the roof of working face, and plays an important role in improving the adaptability of hydraulic support and optimizing the quality of surrounding rock control dynamically.However, underground environment is special, different positions in the working face lead to an inconsistent data distribution of work resistance of hydraulic support due to variable working conditions.Simultaneously, the production process has the characteristics of nonlinearity and uncertainty, which seriously affect the accuracy of the data-driven mining pressure prediction model.Therefore, for the problems that prediction accuracy is affected by inconsistent working resistance data distribution due to the frequent changes of working conditions, a novel robust regression model of ground pressure for fully mechanized mining face considering the domain adaptation of data distribution based on manifold regular domain adaptation integrated function link prediction error method(MRDA-FLPEM) algorithm is proposed.Firstly, the manifold regular domain adaptation algorithm is introduced to obtain the feature map matrix, which can map the data distribution structure of source and target domains to a public space uniformly so that the consistency of the data distribution under multiple working conditions can be maintained.Then the function link prediction error method is used to establish a mining pressure prediction model based on the migrated data of source domain, and improve modeling based on the target domain data, which can reduce the influence of the time series data distribution difference of multiple working conditions on the accuracy of the prediction model.The effectiveness of the proposed integrated algorithm is validated from a fully mechanized mining face, and the results demonstrate that the proposed algorithm can improve the influence of data distribution differences in multi-conditions on the prediction model, improve the robustness and generalization ability of the model, which can establish a foundation for the subsequent analysis of strata behavior law in the fully mechanized mining face, adapt to the changes of environment in advance, and guide the normal mining operation.