Research Of 10?320cm Ground Temperature Deduction Model Based On BP Neural Network

According to the missing data problem in the ground temperature observation,as well as National Ordinary Meteorological Observing Station has no deep ground temperature observation service,real-time ground temperature deduction model and deep ground temperature deduction model by the Back Propagation(BP)neural network was proposed in this paper.The primary research contents are as follows:1.One of the design requirements of the ground temperature deduction model is that the input vector contains air temperature and ground temperature observation date.To under this requirement,the elements of the input vectors and output vectors of the real-time ground temperature deduction model and deep ground temperature deduction model were determined respectively,by analyzing the variation characteristics of ground temperature in each layer and analyzing the time span of input vector with Nyquist-Shannon Sampling Theorem.The training function of neural network and the number of hidden layer nodes were determined by single factor method.2.The training set of network was determined by analysis and test.Real-time ground temperature deduction models contain 24 models,which corresponding to the 24 time points per day.As a result,the network could give consideration to both output accuracy and generalization.Debug the neural network parameters,train and test repeatedly,so as to filter out ground temperature models with the best error performance.And then the models were tested with the date of national ordinary meteorological observing station.3.The deep ground temperature deduction models,which had 6 composition schemes of different input and output vectors,were trained and tested respectively to calculate the error of optimal models for each scenario.The two best models with different time spans were selected to compare the network output and observing dates.4.The correction methods of 160cm and 320cm ground temperature calculation data based on moving average filter were studied.The experimental results show:real-time ground temperature deduction models can be used to fill in the blanks of ground temperature observation data,because the output error of their network is small;deep ground temperature deduction models can estimate the deep ground temperature data at a nearby station which has no deep ground temperature observation.