Automatic Construction And Optimization Of Prediction Model In Large Data Environment

Prediction is a technology which is used in many industries.With the accumulation of data,massive data prediction becomes a problem which has to be resolved for these industries.It is particularly important to have a comprehensive and in-depth understanding of the data processing and its analytical methods to solve this problem.Some unique prediction models are necessary to deal with different forms of data in different industries and specific fields.So data fusion analysis and application of cross disciplines and domains will also become a research trend in the future development of large data analysis.However,there are still no much landmark achievements to the automated construction of predictive model to different fields.It is urgent to find the automatic construction method of forecasting model in order to adapt to the new trend of the development of large data.In recent years,some domestic and foreign experts have done comprehensive work to this problem,including two kinds of relatively stable prediction model which are Gauss process regression model and artificial neural network model.Gauss process is a stochastic process,which is very suitable for dealing with complex problems which has small nonlinear and high dimension samples.The artificial neural network model(ANN)has the ability to search for optimal solutions,do parallel computation and associative storage.ANN also has good self-learning ability.However,both model have its advantages and disadvantages,and there will always be error in the processing of data or the prediction results.Therefore,a single Gauss process regression model and artificial neural network can not obtain better prediction results.By analyzing the advantages and disadvantages of various models,this paper presents a prediction model: PWGB model which can achieves automatic prediction.The PWGB prediction model firstly does data processing with PCA algorithm,and the result will be passed into the improved Gauss regression model as input data.And the improved Gaussregression model will do fitting to these input data to get the basic prediction data and the residual sequence.After that the BP neural network will modify the residual sequence and combine it with the basic prediction data,to produce the final prediction result of PWGB model.Compared with the single model,the PWGB model is more suitable for the combination forecasting model.and the results of comparison experiments further proved that the model has better applicability and can improve the accuracy of the prediction results.