Research On The Cross Prediction Model Of Steel Performance And The Technological Parameters

With the rapid development of automobile, shipbuilding, high-speed rail, aerospace and soon, social demand for high performance steel is increasing. How to reduce the steel productioncost, improve production efficiency and quality assurance has become imperative. Practicesshow that nonlinear mapping relation between steel performance and technological parametersin the process can’t be described by common mathematical model. The BP neural network hasgood performance in dealing with complex nonlinear problem because it needn’t to presetfunction, and has outstanding performance in modeling fitting and forecasting the stronglynonlinear data. But BP neural network has defects in selection initial weights and thresholdvalues and easy to fall into the local optimal solution, now the BP neural network model topredict steel performance is still not perfect.To overcome the inherent defects of BP neural network algorithm, enhance its applicationof quality for steel production practice and accident investigation, an improvement actions toBP neural network algorithm is proposed. Through the permutation and combination, sixcross prediction models, of three sets of data including steel performance, temperature controltechnology and chemical composition addition technology, are built. This method combinedniche technology and Tabu search with Genetic Algorithm (GA), the selection of BP neutralnetwork initial weights and threshold values is optimized and local optimum can be avoided.Cauchy error estimator instead of traditional mean-square deviation estimator is used toimprove error statistical method of BP neutral network, and the influence of abnormal inputelements on results is reduced. Results show that all the model test precision is higher aftertraining, the overall average relative error is small, good prediction ability and practicalfeasibility of the presented method is demonstrated.