XGBoost soft measurement modeling method based on parallel LSTM auto-encoder dynamic feature extraction

Abstract

The invention discloses an XGBoost soft measurement modeling method based on parallel LSTM auto-encoder dynamic feature extraction, and belongs to the field of industrial process prediction and control. The LSTM auto-encoder extracts an encoding vector as a dynamic feature in a manner of restoring an input sequence; the idea of data parallelism and model parallelism is utilized to carry out distributed training on the network; the modeling efficiency is improved, the extracted dynamic features and the original features are combined to train an XGBoost model, and finally the steps of predictingthe test sample, repeatedly extracting and splicing the features, inputting the features into the XGBoost model and the like are carried out until prediction of all the samples is completed. The XGBoost soft measurement modeling method based on parallel LSTM auto-encoder dynamic feature extraction is helpful for solving dynamic problems in process soft measurement based on other models, and meanwhile, two modes of data parallelism and model parallelism are adopted in the network training process, so that the network training speed is increased; and it can be ensured that the precision of theXGBoost model is unchanged or stably improved after the dynamic characteristics are introduced, and the robustness of the model is improved.

Description

technical field [0001] The invention belongs to the field of industrial process prediction and control, and relates to an XGBoost soft sensor modeling method based on parallel LSTM self-encoder dynamic feature extraction. Background technique [0002] The soft measurement of key variables in industrial processes refers to a method of predicting the measured variables by establishing a mathematical model of the measured variables and other variables. It is mainly used to estimate variables that are difficult or costly to measure in actual industrial processes. At present, soft sensor models are mainly divided into mechanism models and data-driven models. Because the derivation process of mechanism models is often very complicated, data-driven soft sensor models are widely used. [0003] At present, the common data-driven soft sensor models mainly include: partial least squares, support vector machine, neural network, decision tree model, etc. These basic models can only meet...

AI Technical Summary

Problems solved by technology

[0003] At present, the common data-driven soft sensor models mainly include: partial least squares, support vector machine, neural network, decision tree model, etc. These basic models can only meet the fitting and prediction of simple data, but in the actual industry In the process, there will be problems such as process nonlinearity, high-dimensional data, data noise, data non-Gaussian distribution, lack of quality variables, state drift and time-varying, process dynamics, and quality variable sampling delays. Improvements or combinations to accommodate more complex modeling problems

[0004] In the above practical problems, the process dynamics are often difficult to solve. The main reason for the process dynamics is that the samples at the current moment may have a relatively implicit relationship with the samples at several previous moments, and in terms of specific sample characteristics It is still difficult to reflect, and the more common solution at present is to use a time series model with a relatively simple structure such as ARIMA, and add the sample information of the previous moment to the model in a linear or non-linear form to improve the prediction accuracy, but this method is often There are two problems. One is that the dynamic characteristics of samples in different working conditions are not the same. For example, in some cases, the samples show linear fluctuation information, while in some cases it is non-linear. Therefore, it is difficult to use a more general time The sequence model is used to synthesize the dynamic characteristics between samples; the second is that with the gradual increase of industrial process data and the formation of industrial big data problems, the relatively simple time series model is difficult to effectively extract sufficient and accurate information from massive data, and the calculation It will also cost a lot of time

Images