Research On Data Prediction Based On Artificial Neural Network

With the improvement of human data acquisition ability,a large number of mea-surement data have been accumulated in all walks of life.How to discover and mine the inherent law of things'development from these massive data and predict the development trend or state in a certain period in the future is the core issue of data science and prediction research.Data forecasting mainly includes deterministic fore-casting and uncertain forecasting.Deterministic prediction is mainly applied to the classification and regression of deterministic data,while uncertainty prediction is ap-plied to the situation where there are uncertain factors in practical problems.Unlike point estimation of deterministic prediction,uncertainty prediction mainly evaluates the accuracy and uncertainty of prediction by prediction interval.Based on the arti-ficial neural network model,this paper studies the data prediction methods,and the main contents are as follows:First,classification prediction based on deep belief network(DBN)and whale op-timization algorithm(WOA).The layer-by-layer training mechanism of deep learning technology is used to learn the deep nonlinear network,so as to realize the charac-teristic of complex function approximation.Based on the deep belief net.work(DBN),a DBN-WOA model for deterministic data prediction is established by improving the whale optimization algorithm(WOA),and the classification problem is surely predicted and evaluated.Firstly,a deep confidence network with one input layer,two hidden layers and one output layer is constructed,and the output vectors of the hidden layer and the output layer are calculated by using Sigmoid activation func-tion.Secondly,taking the accuracy of the current classification prediction as the opti-mization target,WO A is used to optimize the weight matrix and bias vector between adjacent layers of DBN,and obtain the DBN meeting the confidence level.Finally,three groups of data are selected as case studies to compare the classification accu-racy of DBN-WOA model,classical DBN model,BP model and LSTM model.The results show that compared with the other three models,DBN-WOA model has high-er accuracy and stronger generalization ability,and is a reliable data classification prediction model.Second,regression prediction based on DBN and WOA.The Sigmoid activation function of the network output layer in DBN-WOA model is changed into a linear activation function such as pureline function,the regression prediction of determin-istic data is carried out.The take prediction error is taken as optimization target to optimize DBN network through WOA,so as to improve prediction accuracy.Three groups of data are selected to make case study by comparing the regression errors of DBN-WOA model,classical DBN model,BP model and LSTM model for the same data.The results show that the regression accuracy of DBN-WOA model based on swarm intelligence algorithm is lower than that of other models based on gradient descent algorithm.Third,interval prediction based on DFNN,WOA and LUBE.In view of the uncertainty of data prediction,a new interval prediction model WDL is designed by combining WOA,DFNN and LUBE methods.Firstly,a double-output feedfor-ward neural network is constructed,with the larger output as the upper bound of the prediction interval and the smaller output as the lower bound.Secondly,tak-ing the interval coverage width criterion(CWC)as the network optimization objec-tive,aiming at its discontinuous and nondifferentiable characteristics,the parameters of the double-output feedforward neural network are optimized by improving the whale optimization algorithm.Finally,the prediction performance of WDL model is evaluated by 10 general data sets,and are compared with GDL(GA+DFNN+LUBE)and PDL(PSO+DFNN+LUBE)models.Furthermore,the influence of data denoising on the prediction performance of WDL model is explored through data set parti-tioning technology.The results show that,WDL model can obtain higher prediction interval coverage and narrower normalized average width of prediction interval,which significantly reduce the uncertainty of data prediction,and can improve the accuracy of prediction and the performance of uncertain prediction model.