A Predictive Analysis Of Air Temperature Based On Improved EMD-PSO-ANN Model

Being able to predict the temperature more accurately is of great significance for human beings: from the macro level,it can guide human production more effectively;from the micro level,it can affect people’s daily life.The methods of temperature prediction mainly include numerical prediction method and data mining method.Numerical prediction method has extremely high requirements on hardware and software equipment.In addition,complex multi-order partial differential equations need to be established and numerically solved.Compared with this method,data mining method does not require harsh hardware support.Furthermore,through the full use of historical data,it can reveal the long-term intrinsic rules and laws of temperature changes.Using data mining methods,an EMD-PSO-ANN prediction model is established for the historical data of the daily maximum temperature of Xi’an City from 2011 to2019.The model first performs empirical mode decomposition(EMD)on historical temperature,and then establishes a three-layer artificial neural network(ANN)prediction model for each order of the intrinsic mode functions(IMF)obtained by EMD.The activation function of the hidden layer nodes is in the form of radial basis function(RBF).Then the model parameters are optimized by particle swarm optimization(PSO)algorithm.When giving the final error evaluation of different models,the method of nested cross-validation(Nested C-V)is adopted.On improving the standard PSO algorithm,the relative measurement index is put forward creatively to measure the degree of aggregation of particles.By specifying the threshold value of the particle aggregation,the timing of restarting the search procedure and randomly redistributing particles is controlled.After simulation research,this improvement is proved to be able to mitigate the impact of the particle swarm trapped in the local optimal solution in the search process.Through the comparison of simple ANN model,PSO-ANN model and EMD-PSO-ANN model,the EMD-PSO-ANN model proposed has the best prediction result,the MAPE error of which is 6% lower than that of ANN model and 4% lower than that of PSO-ANN model.