Optimizing Depth Confidence Network Based On Particle Swarm Optimization Short-term Load Forecasting

Power system short-term load forecasting is safeguard an indispensable part in the smooth running of the grid,and the accuracy of load forecasting of power grid operation is of great significance.However,influenced by various factors,load forecasting error cannot be completely eliminated in practice,leading to the difficulty of the accuracy of load forecasting to achieve one hundred percent,therefore,how to improve the accuracy of load forecasting effectively,to the power grid operation standards,has always been the hot spot of the academic research.In many prediction models,neural network has strong learning ability and generalization ability,so it is widely used in the field of load forecasting.Based on the latest research results of neural network,this paper deeply analyzes the characteristics and influencing factors of load forecasting,and makes relevant research with the historical load data of a county in Sichuan province.The main contents are as follows:(1)relevant theories of power system load prediction are studied,and the periodic characteristics and composition structure of power load are analyzed.In order to verify the good performance of the neural network in the field of load forecasting,three common neural network models were used to predict the load value of a county in Sichuan province in the next three days.According to the prediction results,all three models can be applied to load prediction,but there are still large errors in the prediction results.Through analysis and comparison,it is concluded that the BP neural network prediction curve is relatively good and has a large optimization space.(2)in order to improve the accuracy of load prediction,the deep confidence network(DBN)was adopted to improve the efficiency of training by greedy pre-training restricted boltzmann machines layer by layer,aiming at problems such as the single structure of BP neural network,easy to get into local optimization,and the input layer only considers the historical load data.At the same time,in view of the problem that the input layer only considers the historical load data and does not consider other influencing factors,several factors that have great influence on short-term load prediction are listed and added to the input layer.(3)aiming at the initial randomization of weights in the DBN pre-training process,resulting in slow overall convergence speed and other problems.A method of DBN optimization based on improved particle swarm optimization(pso)is proposed.Firstly,the learning factors and inertia weights in the particle swarm optimization(pso)algorithm are dynamically adjusted.Then,the improved pso algorithm is combined with the deep confidence network(DBN),and the optimization function of the pso is used to adjust the initial weight of the deep confidence network(DBN)pre-training.(4)using the historical load data of a county in Sichuan province as training samples,a prediction model based on DBN was established.Through error analysis of the predictionresults and comparison with other models,the pso-dbn model proposed in this paper is verified to be accurate and effective in the field of short-term load prediction.(5)in order to apply this theory to engineering practice,according to the design requirements for a short-term load forecasting software,the software to "Visual Studio 2010" as a development platform,using CSharp language programming,and selects the SQL Server2008 database development,basically achieved data management,data analysis,load analysis of load forecasting function,can achieve a simple and friendly human-computer interaction effect.