Separation And Extraction Of Measured Time Series Data And Advanced Prediction Of Mineral Pressure

In recent years,the rapid development of artificial intelligence technology and its wide application in coal mining have made it popular in mine pressure prediction research,but mine pressure prediction research based on machine learning and deep learning technology is still in its infancy.At present,most of the current mine pressure prediction research work based on artificial intelligence technology uses different algorithms or improved models to achieve high-precision prediction of a single or several data in a short period of time,ignoring the impact of the measured mine pressure data processing on the mine pressure data prediction.And there is no research to show that continuous and long-term mining pressure data can be predicted ahead of time.However,compared with the prediction of a single or several data in a short period,the realization of long-term advance prediction of mine pressure data has more profound practical significance in analyzing the development trend of mine pressure and guiding safe production.For this reason,this paper takes the measured rock pressure data of hydraulic supports as the research object,and adopts the methods of theoretical analysis,numerical analysis and laboratory experiments to propose a separation and extraction processing method of time series data,and carry out long-term advanced prediction of rock pressure data.Research.And obtained the following main research results:(1)A separation and extraction method suitable for time series data processing is proposed.The method has the characteristics that the extracted data is still time series data,the data set after extraction is composed of multiple sets of time series data,and the position number of each data on the time axis before and after extraction can be calculated by the extraction interval number through formulas,etc.The sequential and reverse sequential separation and extraction of sequence data,the sequential sequential extraction is used to extract the training set and test set sample data for model training,and the reverse sequential extraction is used for the extraction of large-duration advance prediction sample data.(2)It is found that the weakening of the overall performance of the model is related to the destruction of hidden information in the measured data.The overall performance of the model is evaluated from the two aspects of model training effect and prediction performance.The results show that the model training effect and prediction performance both decrease with the increase of the time interval of the sample set,and finally fluctuate around a certain level.As the time interval increases,the hidden information in the measured mineral pressure data is also destroyed,which makes the model unable to obtain sufficient regular information from the sample data.(3)The destruction law of the hidden information in the measured mine pressure data is revealed,and a calculation method of the reasonable time span for predicting the mine pressure data with a large duration ahead of time is given.Based on the ratio of the variation of the function value that cannot be explained by the model regression relation in the test sample set to the total variation,the damage of the hidden information in the measured rock pressure data by the separation and extraction method is studied,and it is further confirmed that the selected hydraulic support measured data conforms to the The reasonable time span of the large time-ahead forecast required for the actual forecast is 455 minutes(7 hours and 35minutes).(4)A reverse check calculation model of the minimum data volume of the validation set that satisfies the extraction of the large-length advance prediction sample set is established.In order to realize the long-term advanced prediction of the mine pressure data within the reasonable time span of the determined advanced prediction,the maximum number of intervals,the dimension of the sample eigenvalues and the number of label values in the model training set are extracted according to the separation and extraction method that meets the actual prediction requirements.It is tested whether the data volume of the set can meet the extraction of sample data for long-term advance prediction.(5)It has realized the long-term advance prediction of the working resistance data of the hydraulic support for the next 455 minutes(7 hours and 35 minutes).On the basis of studying the destruction of hidden information in the measured mine pressure data,based on the single-step prediction of the long short-term memory neural network model group,the long-term advance prediction of the hydraulic support working resistance data with a time span of 455 minutes(7 hours and 35 minutes)is carried out.According to the evaluation and analysis of the long-term advanced prediction results and the comparison with related research results,it is shown that the long-term advanced prediction results can be applied to actual prediction,and can provide a reference for the study of the mine pressure manifestation law.Based on the proposed separation and extraction method of time series data processing,this paper presents an implementation scheme of continuous and long-term advance prediction of mineral pressure data,and realizes the long-term advance prediction of mineral pressure data for the first time.The obtained research results further verify the feasibility of realizing continuous and long-term advance prediction of mine pressure data.At the same time,the proposed time series data separation and extraction processing method also provides a new idea for further research on long-term advance prediction of other time series data.