Research On Short-term Load Forecasting Based On MDS And PSO-GRU Neural Network

Short-term power load forecasting is the basis for ensuring the stable operation of the power system.Aiming at the changing laws and characteristics of modern power load,this paper aims to improve the accuracy of short-term power load forecasting,and establishes a gate control based on multidimensional scaling(MDS)and particle swarm optimization(PSO)optimization.The short-term load forecasting model of Gated Recurrent Unit Neural Network(GRU)neural network.First,in order to effectively improve the accuracy of data preprocessing,this paper analyzes the characteristics of short-term power load forecasting and the influencing factors of power load through historical data before data preprocessing.The following conclusions can be drawn: short-term power load forecasting is inaccurate,The characteristics of limitation,timeliness and multiple schemes;historical load,date type,weather and temperature are the main factors that affect the accuracy of short-term power load forecasting.It has laid a certain theoretical foundation for the following data preprocessing and data dimensionality reduction of power load influencing factors.Secondly,in view of the huge data of short-term power load forecasting factors that lead to too long model training time and low fitting accuracy,this paper studies the basic principles and characteristics of four dimensionality reduction methods.Select the dimensionality reduction rate,dimensionality reduction loss and dimensionality reduction quality as the evaluation indicators of the dimensionality reduction methods,and make a preliminary comparison of the dimensionality reduction methods;based on the selected data set in this article,the four dimensionality reduction methods are established to establish a load forecasting simulation model,and through examples Simulation analysis proves that the MDS algorithm has good dimensionality reduction ability when processing the data in this article;the basic principles of Long Short-Term Memory(LSTM)and GRU neural network models are studied,and the MDS algorithm will be processed The data is brought into the LSTM model and the GRU model,and the load forecasting capabilities of the two are compared through example simulation,which proves the superiority of the GRU model in short-term load forecasting.In addition,in order to balance the learning ability of the network and the complexity of training,the PSO algorithm is used to optimize the hyperparameters of the neural network,and a shortterm power load forecasting model based on the PSO algorithm to optimize the GRU neural network is established.Finally,in order to solve the above problems,this paper establishes a short-term power load forecasting model based on multi-dimensional scaling analysis and particle swarm optimization to optimize the gated recurrent unit neural network.According to the ten-month actual historical load data and its influencing factor data in a certain area of Hubei Province,a case simulation was carried out in MTALAB.At the same time,it was compared with the GRU model,LSTM model and PSO-GRU model.The results prove that the forecasting scheme proposed in this paper has Higher prediction accuracy.