Data identification method and system for high-noise industrial process

Abstract

The invention discloses a data identification method and system for a high-noise industrial process. The method comprises the following steps: acquiring historical time series data of an industrial process; preprocessing the historical time series data; based on a noise period in the historical time series data, determining a Butterworth filter; filtering the preprocessed historical time series data through the Butterworth filter; constructing a frequency domain transfer function model; training the frequency domain transfer function model through the filtered historical time sequence; and converting the trained frequency domain transfer function model into a time domain to obtain a time domain transfer function model as a system identification model, and identifying the data of the high-noise industrial process by using the system identification model. According to the method, online prediction can be carried out in an industrial control environment, and a prediction result is obtained in real time so that the requirements of an actual industrial process are met, the frequency characteristics of the industrial process are effectively analyzed, and high-frequency fluctuation noiseis identified and filtered explicitly.

Description

technical field [0001] The invention relates to the field of industrial control, in particular to a data identification method and system for high-noise industrial processes. Background technique [0002] In recent years, with the vigorous development of industrial intelligence and big data technology, more and more intelligent detection and control facilities such as DCS and PLC have been integrated into the industrial production process, which has improved the automation level of the industrial field. This also brings opportunities for grasping the operation rules of complex equipment, predicting it based on past data and existing status, and then estimating the output. However, due to the complexity of the application environment, the sensor data in the actual environment will inevitably be affected by high-frequency noise such as electromagnetic interference and power grid fluctuations. As a result, the volatility of the recorded data in the system is much higher than th...

AI Technical Summary

Problems solved by technology

The linear regression method has low complexity and is not suitable for complex industrial processes; the sliding autoregressive method and the state space model are developed from linear regression, assuming that the same variable x has a linear relationship between each moment, and using the previous state at each moment to predict the current state, This method can eliminate random fluctuations in forecasting, but it is often difficult to represent non-zero mean and periodic noise

Neural networks such as BPNN are difficult to capture the long-term dependence of the system, and the interpretability and robustness are not strong.

Methods such as support vector machines can reduce the dimensionality of high-dimensional features, but the necessary kernel function or distance function cannot be automatically selected.

The above-mentioned industrial process modeling algorithms still have two common problems. One is that they cannot explicitly identify periodic noise. When the amplitude of the noise component is large, the system cannot correctly identify the target model; Only discrete-time models can be built for discrete data, and these models often need to be retrained and predicted when the system control or sampling frequency changes

Images