CN-122021404-A - Training method of fresh air inlet concentration prediction model of clean room

CN122021404ACN 122021404 ACN122021404 ACN 122021404ACN-122021404-A

Abstract

The invention relates to a training method of a fresh air port concentration prediction model of a clean room, belongs to the technical field of clean environment monitoring, and solves the problem of poor accuracy of the existing fresh air port concentration prediction. The training method comprises the steps of obtaining outdoor environment parameters and various pollutant concentrations of a fresh air port from historical measurement data to obtain actual measurement data, constructing a fresh air port computational fluid dynamics CFD model, verifying the fresh air port CFD model by using the actual measurement data to obtain a verified fresh air port CFD model, simulating the verified fresh air port CFD model based on the outdoor simulation environment parameters to obtain various pollutant concentrations of the fresh air port simulation, performing data cleaning on the actual measurement data by using simulation data based on a preset data cleaning method to obtain cleaned actual measurement data, and training a concentration prediction model by using the cleaned actual measurement data and simulation data to obtain a trained concentration prediction model. And the concentration prediction precision of the fresh air port is improved.

Inventors

XU XIAOLI
WEI LAN
WANG YAN

Assignees

中国电子工程设计院股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251230

Claims (10)

1. The training method of the fresh air inlet concentration prediction model of the clean room is characterized by comprising the following steps of: The method comprises the steps of obtaining outdoor environment parameters and various pollutant concentrations of a fresh air port from historical measurement data to obtain actual measurement data, constructing a fresh air port computational fluid dynamics CFD model, and verifying the fresh air port CFD model by using the actual measurement data to obtain a verified fresh air port CFD model; Simulating the verified fresh air port CFD model based on the outdoor simulation environment parameters to obtain various pollutant concentrations simulated by the fresh air port, and combining the outdoor simulation environment parameters and the various pollutant concentrations simulated by the fresh air port to obtain simulation data; based on a preset data cleaning method, performing data cleaning on the actual measurement data by using simulation data to obtain cleaned actual measurement data; And training the concentration prediction model by using the cleaned measured data and simulation data to obtain a concentration prediction model after training.
2. The training method of claim 1, wherein the outdoor environmental parameters include one or more of: A temperature; humidity; Wind direction; wind speed; Outdoor potential pollution sources and pollutant emission intensity.
3. The training method of claim 2, wherein the constructing a fresh air port computational fluid dynamics CFD model comprises: Determining a geometric model of a CFD model of the fresh air port based on indoor and outdoor real environments, wherein the indoor and outdoor real environments comprise a clean room and other buildings or building groups in a computing area; selecting different physical models and model parameters of the CFD model of the new air port; Boundary conditions and calculation conditions of the CFD model of the fresh air port are set.
4. The training method according to claim 2, wherein the simulating the CFD model of the verified fresh air port based on the outdoor simulation environment parameters to obtain the simulated fresh air port with various pollutant concentrations comprises: Determining outdoor simulation environment parameters aiming at different temperature, humidity, wind direction and wind speed combinations; inputting each outdoor simulation environment parameter into the verified fresh air port CFD model, and simulating to obtain the corresponding simulated fresh air port with various pollutant concentrations; the outdoor simulation environment parameters include various extreme outdoor environment parameters.
5. The training method of claim 4, wherein the pre-set data cleansing method employs one or more of: A physical rule verification method; an error threshold screening method; a distribution consistency checking method; Correlation analysis method; a data fusion correction method.
6. The training method of claim 1, wherein training the concentration prediction model using the cleaned measured data and the simulation data comprises: carrying out standardized processing on the cleaned measured data and simulation data, and dividing the standardized data into a training set and a testing set according to a preset proportion; training the concentration prediction model by using a preset training algorithm based on the samples of the training set until a preset training cut-off condition is met; and if the test is qualified, the concentration prediction model with the cut-off training is used as a concentration prediction model with the finished training.
7. The training method of claim 6, wherein the predetermined ratio of the training set to the test set is 7:3.
8. The training method of claim 7, wherein the concentration prediction model employs a long-term memory network LSTM.
9. The training method according to claim 8, wherein the preset training algorithm is any one or more of the following: A random gradient descent algorithm; adaptive learning rate algorithm Particle swarm optimization algorithm; And (5) a sine and cosine algorithm.
10. The training method of claim 9, wherein the loss function of the concentration prediction model employs a combination of a mean square error loss function and a relative entropy loss function.

Description

Training method of fresh air inlet concentration prediction model of clean room Technical Field The invention relates to the technical field of clean environment monitoring, in particular to a training method of a fresh air inlet concentration prediction model of a clean room. Background With the development of high-end industries such as electronic information, aerospace and the like, the influence of chemical pollutants in a clean room environment is increasingly prominent. The fresh air port is used as a key node for indoor and outdoor air exchange, and the pollutant concentration dynamic change is driven by multiple factors such as outdoor wind speed, wind direction and the like, so that the following bottlenecks are faced: Data scarcity, high actual measurement cost, limited sample size and difficult coverage of space-time dynamic change of fixed point position monitoring. For example, outdoor meteorological conditions (such as sudden change of wind speed) cause concentration fluctuation of a fresh air port, but the traditional method relies on a small amount of historical data, so that the prediction error is larger; The model has insufficient robustness, the existing machine learning model has poor generalization capability under variable working conditions or extreme meteorological conditions, is easy to be interfered by noise, and has low matching degree between a prediction result and a physical rule; the physical constraint is missing, the conventional data driving method ignores the fluid mechanics principle, so that abnormal data is not cleaned, and the reliability of the model is affected. Certain electronic factories cannot early warn in time due to pollution of fresh air ports, huge losses are caused, and limitations of the prior art are highlighted. Therefore, a new technical solution for predicting the concentration of the fresh air inlet and the fresh air outlet indoors is needed. Disclosure of Invention In view of the above analysis, the embodiment of the invention aims to provide a training method of a fresh air port concentration prediction model of a clean room, which is used for solving the problem of poor accuracy of the existing fresh air port concentration prediction. The embodiment of the invention provides a training method of a fresh air inlet concentration prediction model of a clean room, which comprises the following steps: The method comprises the steps of obtaining outdoor environment parameters and various pollutant concentrations of a fresh air port from historical measurement data to obtain actual measurement data, constructing a fresh air port computational fluid dynamics CFD model, and verifying the fresh air port CFD model by using the actual measurement data to obtain a verified fresh air port CFD model; Simulating the verified fresh air port CFD model based on the outdoor simulation environment parameters to obtain various pollutant concentrations simulated by the fresh air port, and combining the outdoor simulation environment parameters and the various pollutant concentrations simulated by the fresh air port to obtain simulation data; based on a preset data cleaning method, performing data cleaning on the actual measurement data by using simulation data to obtain cleaned actual measurement data; And training the concentration prediction model by using the cleaned measured data and simulation data to obtain a concentration prediction model after training. Based on a further improvement of the training method, the outdoor environment parameters include one or more of the following: A temperature; humidity; Wind direction; wind speed; Outdoor potential pollution sources and pollutant emission intensity. Based on the further improvement of the training method, the construction of the fresh air port computational fluid dynamics CFD model comprises the following steps: Determining a geometric model of a CFD model of the fresh air port based on indoor and outdoor real environments, wherein the indoor and outdoor real environments comprise a clean room and other buildings or building groups in a computing area; selecting different physical models and model parameters of the CFD model of the new air port; Boundary conditions and calculation conditions of the CFD model of the fresh air port are set. Based on the further improvement of the training method, the method simulates the verified fresh air port CFD model based on the outdoor simulation environment parameters to obtain the simulated fresh air port concentration of various pollutants, and comprises the following steps: Determining outdoor simulation environment parameters aiming at different temperature, humidity, wind direction and wind speed combinations; inputting each outdoor simulation environment parameter into the verified fresh air port CFD model, and simulating to obtain the corresponding simulated fresh air port with various pollutant concentrations; the outdoor simulation environment parameters include va