Research On Short-Term Load Forecasting Based On Machine Learning

With the development of society,electricity has become an important part of people ’s daily life.However,with the increase of clean energy such as wind power and solar power,the power grid system is becoming more and more complex.The demand for electricity is also increasing,and the problem of unbalanced supply of power resources is gradually leaking.Power load forecasting is an indispensable part of the power system.Accurate prediction of power load is of great significance for load scheduling,cost saving and power grid improvement.This paper uses the power load data set of the Electrical Cup Mathematical Modeling Contest to complete the task of building a machine learning model for short-term prediction of power load.The main research contents are as follows :(1)To address the problem of low model prediction accuracy,the optimization of the hyperparameters of the A-Bi GRU model using the Mixed Startegy Improved Whale Optimization Algorithm(MSWOA)is proposed.In this thesis,we combine the attention mechanism to assign weights to Bi GRU output states,extract important features from Bi GRU,and improve Bi GRU network performance.MSWOA is used for the hyperparameter selection of the model to improve the prediction accuracy of the A-Bi GRU network.The samples with different features added are trained and validated separately,and comparison experiments are conducted with Bi GRU,A-Bi GRU and WOA-A-Bi GRU.The prediction results show that MSWOA can effectively find the suitable parameters to improve the prediction accuracy and stability of A-Bi GRU network;comparing different samples,the prediction performance of the sample based on the selected features is the best,and the MAPE,MAE and RMSE of MSWOA-A-Bi GRU are 1.935%,145.399 MW and 255.215 MW in this sample,respectively.further enhancing the prediction effectiveness of the model.(2)To solve the problem of unstable model prediction performance,we propose to optimize the network parameters of Extreme Learning Machine(ELM)using Improved Artificial Hummingbird Algorithm(IAHA).The performance of the standard artificial hummingbird algorithm is improved by modifying the initialization method and the update rules of the access table of the artificial hummingbird algorithm.the weights and biases of the hidden layer in the ELM model are chosen randomly,and the prediction performance of the model is unstable,so this thesis optimizes the network parameters of the ELM by IAHA to further improve the prediction performance of the network model.The samples with different features added are trained and validated separately,and the experiments are compared with ELM and AHA-ELM.The prediction results show that IAHA-ELM has better prediction performance and more stable prediction effect compared with other models;comparing different samples,the prediction performance of the sample based on full features is the best,and the MAPE,MAE and RMSE of IAHA-ELM in this sample are 2.008%,156.723 MW and 235.849 MW respectively,which effectively improves the prediction accuracy of ELM.