Research And Implementation Of Prediction Algorithms Suitable For Energy Fluctuation

In this thesis,an energy consumption prediction system and two prediction algo-rithms suitable for energy consumption fluctuation were implemented.First of all,a multi-layer perceptron algorithm based on the improved particle swarm optimization was pro-posed.The traditional particle swarm optimization algorithm has the shortcoming of being easily trapped in the local optimal solution.In this thesis,traditional particle swarm opti-mization algorithm was improved.And it can optimize the multi-layer perceptron algo-rithm by using its optimization features.Then,a Lasso-SVR hybrid prediction algorithm based on improved genetic algorithm optimization was proposed.The algorithm mainly used Lasso regression linear approximation ability and SVR nonlinear fitting ability to predict energy consumption data.Finally,an energy consumption prediction system was implemented.The main contents are as follows:(1)The algorithm of multi-layer perceptron was studied.And a particle swarm op-timization algorithm improved by using good-point set theory and Levy flights theory was proposed to optimize the multi-layer perceptron.At the same time,the particles were divided into a plurality of particle groups to ensure the diversity of particles.Compared with the static velocity adjustment of the traditional particle swarm optimization algo-rithm,the improved particle swarm optimization algorithm performed dynamic velocity adjustment according to the fitness of the particle.The experiment showed that the im-proved particle swarm optimization algorithm was more difficult to fall into the local op-timal solution.Therefore,when the algorithm was applied to the parameter adjustment of the multilayer perceptron,it was easier to find the global optimal solution.The multi-layer perceptron algorithm based on improved particle swarm optimization algorithm has a good prediction effect.(2)The traditional Lasso regression and SVR algorithms were studied,in which a hy-brid prediction algorithm combining the two was proposed.Genetic algorithms were also used for parameter adjustment.Traditional genetic algorithm has a problem of slow con-vergence and need to be improved.In this thesis,the selection,exchange and mutation of genetic algorithm were optimized,and the genetic algorithm was optimized by simulated annealing algorithm.The experiment showed that the algorithm accelerated the conver-gence and shortens the running time.Compared with the classical Lasso regression and SVR algorithm,the Lasso-SVR hybrid prediction algorithm based on improved genetic algorithm has better prediction effect.(3)The entire energy consumption prediction system was designed.Starting from the main functions,it introduces the realization of user management function,predictive func-tion and data preprocessing function.Finally,the system is tested.By studying the prediction algorithm for energy consumption fluctuations,it helps en-ergy managers to discover the anomalies of useless energy as well as helps energy suppliers optimize their energy supply strategies.Therefore,the research on energy consumption pre-diction algorithm is of great significance to China's energy conservation and consumption reduction policies.