Research On Domain Adaptation Algorithm Based On Extreme Learning Machine And Its Soft Sensor Application

The process industry is a pillar industry that occupies an important strategic position in the national economy.It has significant characteristics such as frequent changes in production raw materials,violent fluctuations in working conditions,and complex internal operating mechanisms.Due to the key process parameters affecting product quality,such as equipment operation status,process parameters and raw material composition of production process,real-time and comprehensive detection cannot be realized by hardware sensors,which makes it difficult to establish accurate prediction models for process control in industrial process.The process industry is moving toward high efficiency,green and intelligentTaking the basic equipment wet ball mill,which is common in the process industry,as an example.It is difficult to measure the mill load according to the material balance of the grinding process and the metal balance by establishing mechanism model,due to the comprehensive complex dynamic characteristics of the grinding process itself and the dynamic changes of the external disturbance factors.Limited by the closedness and rotation of the running process,the working characteristics of continuous operation make the critical load parameters required for the optimal control of the mill difficult to monitor,has low precision and unstable performance.Soft sensor is an effective means to solve this problem in industrial practice.However,the ball mill also has the characteristics that the material changes in the process industry cause frequent fluctuations in the working conditions,which leads to the misalignment of the traditional machine learning-based soft measurement model.How to obtain effective information for modeling unknown conditions from the known source modeling conditions and achieve the purpose of synergy has become an urgent problem to be solved.Based on the above analysis,this paper takes the domain adaptive learning strategy as the starting point,and studies the soft sensor method of the critical load parameters of the wet ball mill under the unknown mode in the multi-case task.The main research work is as follows::(1)When multiple signals such as vibration and vibration are obtained before and after the change of working conditions,multi-view learning is introduced into the domain adaptation strategy modeling process.From the perspective of the model,the two-view domain adaptation model is integrated to solve the model misalignment problem when the working conditions changed.Constructing multi-view based domain adaptive Extreme Learning Machine to predict mill load parameters and compare with traditional soft measurement modeling method.(2)For the common multi-case task that only obtains a single signal,from the characteristic point of view,the deep Extreme Learning Machine self-encoder is used to extract the common features between the fields in an unsupervised manner,and the weakly paired maximum association in the deep feature space.The variance analysis extracts the relevant information of different working conditions,so that the source domain data closely matches the target domain data,and the accurate prediction of the critical load parameters under the unknown mode is realized(3)There are distribution differences for different fields of data,and it is difficult to obtain enough sample tags to establish a soft measurement model problem in the target domain.The model parameters migration and feature mapping are used to build the model,and the subspace alignment is used to narrow the data distribution difference while using a small number of target domains.The tag data adjusts the source domain model to adapt it to the target domain task,and establishes the soft load model of the critical load parameters of the wet ball mill under the unknown mode.