Full-space State Prediction Of Digital Twin Under Limited Perception

In the special era when the new type of coronavirus pneumonia is raging around the world,various regions continue to improve home isolation and traffic control and epidemic prevention measures,resulting in the problem that the indoor environment is aggravating and harming human health.Indoor environmental monitoring systems are often installed in buildings to improve energy efficiency,ensure good human health and safety,and facilitate the development of digital twin buildings.Only a small number of sensors can be installed in the building,so the indoor environmental monitoring system can only sense the specific information of the installation location of the sensor.How to quickly and accurately predict the environmental factors(such as humidity)of the whole indoor space is an urgent problem to be solved.The fast prediction model based on computational fluid dynamics needs to maintain a balance between real-time performance and accuracy;the method based on spatial interpolation has a low accuracy rate of interpolation results due to limited expert experience;the method based on machine learning has prior analysis of indoor environment cases.Problems such as training,these trained models can only be used in the specified application environment,resulting in the prediction performance of the model will be reduced with the change of indoor environmental conditions.Aiming at the problem that it is difficult to predict the entire space environment information in real time and accurately in buildings,this paper proposes a real-time prediction scheme of indoor state based on machine learning,called ML-PIS.The scheme first builds a spatial graph model to represent the correlation between any two spatial positions,then designs a cross-sample training algorithm to learn the coefficients of the monitoring positions in the spatial graph model,and then predicts the state value of any indoor position.The underlying sensor group is arranged in the scene,and the performance of the proposed scheme is evaluated in many aspects with real and reliable data.The main research contents and achievements include:(1)Design an indoor state perception system based on STM32.Aiming at the problem that real-time acquisition of indoor multi-area state data is difficult,an indoor state perception system is developed in this paper.The system realizes the real-time collection of the environmental data information of the indoor state,which is convenient,reliable and stable,and lays a foundation for the verification of the indoor real state data for the proposed scheme.(2)A spatial graph model for indoor state prediction is proposed.For most indoor environment monitoring locations and unmonitored locations,the weights are expressed in a single way.In this paper,it is proposed that the monitored location has different degrees of influence factors on the unmonitored location,and a multi-dimensional indoor spatial graph model is constructed,which can effectively characterize the correlation between multiple regions.(3)An indoor state prediction scheme based on spatial graph model and machine learning is proposed.It is difficult for the power function p of Inverse Distance Weighted(IDW)to accurately deal with the complex and changeable indoor state environment problems.In this paper,the fixed p of the expert experience is changed into a variable factor that can be learned and optimized,and the cross-sample learning algorithm is used to calculate the real-time optimal variable factor value to improve its adaptability in the indoor state environment.Based on the indoor state perception system,this paper sets up a variety of evaluation indicators to conduct full-space experimental verification of the indoor state environment.The indoor environment monitoring system can continuously collect environmental data as the training samples of ML-PIS,which makes the method in this paper effectively overcome the limitation of the model to the specified application environment,and also provides more accurate indoor environment information for digital twin buildings.