Research On Short-term Load Forecasting Based On Variational Mode Decomposition And Improved LSTM

Accurate short-term power load forecasting is important for maintaining safe and stable operation of the power system and keeping a dynamic balance between the supply and demand side.Since the load values are easily influenced by a variety of external factors,the load series exhibit a certain degree of nonlinearity and nonstationarity.To improve the accuracy of short-term power load forecasting,This paper combines signal decomposition,intelligent algorithms and neural network technology for short-term power load forecasting.The main work is as follows:(1)Introducing the basic concepts related to load forecasting.The intrinsic characteristics of the electric load and the influence of external factors on the load value are analyzed.Aiming at the complex fluctuation of electric load sequence,variational mode decomposition(VMD)is introduced to decompose it into several relatively stable subsequences.In contrast to empirical mode decomposition(EMD),The results show that VMD has better filtering and anti-modal overlap performance.Aiming at the fact that the VMD decomposition effect is easily affected by the number of modes k and the penalty factor α,the adaptive adjusted inertial weight particle swarm optimization(APSO)is used to optimize the parameters.The sub-modal envelope entropy is used as the fitness function,and the minimum value of the sum of all modal envelope entropy is the optimization target,and then the optimal solution of k and α values is obtained.Finally,the parameter-optimized VMD is used to process the real load sequence of a certain area,and the results show that the decomposition effect is better.(2)In view of the diversity of load influencing factors,the correlation between load value and its influencing factors is analyzed,and the main influencing factors are selected according to the absolute values of Pearson coefficient,Spearman coefficient and Kendall coefficient.Aiming at the randomness of the parameter selection of the long short-term memory neural network model,the sparrow search algorithm(SSA)is used to optimize the number of neurons in the hidden layer of LSTM,the number of iterations and the learning rate.The superiority of SSA algorithm in optimization speed and accuracy is verified by two test functions.(3)Combined with the actual data,the simulation experiment is carried out in PyCharm.Comparing the SSA-LSTM short-term load forecasting model proposed in this paper with the improved LSTM model under other optimization algorithms for simulation experiments,the RMSE,MAE,and MAPE values of the model prediction results in this paper are the smallest among all models,which verifies that the SSA algorithm optimizes LSTM parameters more superior.For the parameter-optimized VMD algorithm proposed in this paper,the simulation experiments are compared with the traditional EMD and VMD algorithms under the same conditions,The results show that the decomposition performance of the algorithm in this paper is better,and the prediction performance of the model is further improved.Finally,the APSO-VMDSSA-LSTM short-term power load forecasting model is established by combining the improved methods in the third and fourth chapters,The comparison of multiple models shows that the model has better fitting effect and higher prediction accuracy.In order to improve the practicability of the model,C # language is used to build a short-term power load forecasting platform to realize data management,model training and result display.