Noise prediction of chemical industry park based on multi-station Prophet and multivariate LSTM fitting model

Qingtian Zeng,Chunguo Li,Hua Duan,Geng Chen,Yu Liang

doi:10.1186/s13634-021-00815-6

Qingtian Zeng, Chunguo Li + Show 3 more

Open Access

https://doi.org/10.1186/s13634-021-00815-6

Copy DOI

Abstract

With the gradual transformation of chemical industry park to digital and intelligent, various types of environmental data in the park are extremely rich. It has high application value to provide safe production environment by deeply mining environmental data law and providing data support for industrial safety and workers’ health in the park through prediction means. This paper takes the noise data of the chemical industry park as the main research object, and innovatively applies the 3σ principle to the zero-value processing of the noise data, and builds an LSTM model that integrates multivariate information based on the characteristics of the wind direction classification noise data combined with the wind speed and vehicle flow information. The Prophet model integrating multi-site noise information was adopted, and the Multi-PL model was constructed by fitting the above two models to predict the noise. This paper designs and implements a comparative experiment with Kalman filter, BP neural network, Prophet, LSTM, Prophet + LSTM weighted combination prediction model. R2 was used to evaluate the fitting effect of single model in Multi-PL, RMSE and MAE that were used to evaluate the prediction effect of Multi-PL on noise time series. The experimental results show that the RMSE and MAE of the data processed by the 3σ principle are reduced by 32.2% and 23.3% in the multi-station ordered Prophet method, respectively. Compared with the above comparison models, the Multi-PL model prediction method is more stable and accurate. Therefore, the Multi-PL method proposed in this paper can provide a new idea for noise prediction in digital chemical parks.

Highlights

With the spread of 5G high-speed transmission technology, chemical industrial complexes are entering the Era of Internet of Things (IoT) through sensors [1]
4.3 Multi‐PL model based on Prophet and long shortterm memory (LSTM) combination Based on the characteristics of the Prophet and LSTM models, we propose the Multi-PL model to make up for the limitations of a single model, and can effectively use the park information and the advantages of the two models to achieve higher-precision noise prediction
In the multi-site ordered Prophet method, the root mean square error (RMSE) and mean absolute error (MAE) of the data processed by the 3σ principle are reduced by 32.2% and 23.3%; compared with single-site data, the RMSE and MAE predicted by using the multivariate data set in the LSTM model are reduced by 9.3% and 15.9% dB, respectively

Summary

Introduction

With the spread of 5G high-speed transmission technology, chemical industrial complexes are entering the Era of Internet of Things (IoT) through sensors [1]. More and more researchers are capturing complex time series distribution patterns based on hybrid forecasting models in order to obtain better forecast accuracy and performance [22]. The prediction accuracy obtained by applying the mixed model in the above literature is better than that of the single model, so the mixed model will be the key method to solve the problem of time series prediction of park noise. Based on the existing sensor distribution and traffic data in the chemical park, this paper builds a scene model suitable for the distribution characteristics of the park, constructs a noise multivariate data set and a multi-station data set according to the scene, and introduces the 3σ criterion to deal with the zero value of noise in order to improve the prediction accuracy.

System model and data set

Periodicity of data

Multi‐PL model based on Prophet and LSTM

Findings

Conclusions