Research And Application Of Thermal Process Modeling Method Based On Artificial Neural Network

With the improvement of information technology,it is more and more convenient to obtain and store the operation data in the thermal process.As a non-linear neural network modeling method,the Auto Association neural network has become a research hotspot.In this paper,the modeling,fault diagnosis and migration methods of auto associative neural network are studied.The main contents are as follows:Aiming at the problem that the thermal process modeling is greatly affected by the sample distribution,a high-quality sample extraction method is proposed.Based on the principal component,the system level steady-state factor is calculated,which is used as the steady-state weight for sample reduction.Furthermore,a small number of samples are oversampled according to the reduced sample weight,and finally a balanced high-quality sample is obtained.A gas turbine system is taken as an example to verify the effectiveness of the proposed method in modeling and diagnosis.The results show that the training speed is faster when using highquality samples for modeling,and it can improve the model accuracy degradation and uneven problems caused by sample imbalance,and it can provide good modeling samples for real-time diagnosis.In order to suppress the residual pollution in the process of fault diagnosis,a fault diagnosis method based on compensation auto associative neural network(CAANN)is proposed.By adding a compensation layer to the network model,the model is compensated first and then reconstructed to achieve accurate fault location and se FARation.In addition,to solve the problem of poor real-time performance of CAANN,a rapid branch and bound compensation auto associative neural network is proposed(RB&B-CAANN).The results of simulation examples and gas turbine examples show that RB&B-CAANN can greatly reduce the rate of fault misdiagnosis and speed up the diagnosis on the basis of ensuring the rate of fault diagnosis,which can meet the requirements of online fault diagnosis.This paper proposes a spatial alignment migration learning algorithm based on residuals Alignment(RSA)to realize the knowledge transfer and information fusion between the simulation data and the actual operation data of thermal process.The system model established by this method not only has a wide range of operating conditions,but also can ensure the consistency between the model and the actual operating conditions,so as to greatly improve the practical application effect of the model.The results show that the migration data has the generalization characteristics of simulation data and thermal characteristics of operation data.Based on SIS of power plant,the fault early warning platform is developed.The platform uses the RB&B-CAANN method proposed in this paper for real-time fault early warning,and gives the software design idea,system architecture and operation process.