Research On Short-term Power Load Forecasting Based On Combined Model

Short-term power load forecasting is an important part of the secondary energy allocation process in the power grid,and is equally important for the stable operation of the power system.With the development of open power system and the concept of smart grid being proposed,the traditional forecasting methods can hardly adapt to the increasing requirements for forecasting accuracy under the massive data.In recent years,the application of machine learning in short-term power load forecasting studies has also increased gradually because of its strong self-learning capability.In order to improve the prediction accuracy,the Least Squares Support Vector Machine(LSSVM)is used as the base model,and determine kernel parameters and penalty function by Improved Sparrow Search Algorithm(ISSA)in this paper.Firstly,introduces the development of short-term power load forecasting and classical forecasting models.Then describes the data pre-processing process.The sequence is repaired both vertically and horizontally after identifying the data of anomalies and missing in the sequence.Variational Modal Decomposition(VMD)is used in the process of denoising and smoothing of data sequences.The pre-processing process was validated with actual data.Next,the influencing factors of short-term load are analyzed.similar days are determined by a similarity measure with linear information granulation.The final inputs to the prediction model are determined as six similar day variables and one temperature variable.LSSVM has good performance in terms of noise resistance and operation speed,but the prediction effect is more influenced by the initial parameters.To address this situation,the Sparrow Search Algorithm(SSA)is studied and improved by three aspects in this paper.Obtain the Improved Sparrow Search Algorithm(ISSA).The initial parameters of LSSVM are optimized with ISSA to build the combined prediction model of ISSA-LSSVM.In order to verify the prediction effect of the combined model,representative dates of different seasons and two typical holidays are predicted separately.Then,an error prediction model is built by the validation set data in order to make a first correction to the prediction results.The data of the predicted fluctuation points by trend extrapolation were used to correct the prediction results for a second time.The simulation results show that the prediction error is substantially reduced after two corrections.Short-term power loads are highly volatile.In order to show the fluctuation characteristics of the electricity load comprehensively,this paper also applies the joint prediction model based on the short-term load average and the fluctuation range of the electricity load:(FIG-ISSA-LSSVM)for the prediction study of the fluctuation range of the load.The fuzzy granulation theory is used to granulate the original data sequence into three subseries of maximum,minimum and mean values.Predictions are made for each of these subsequences.Forecasting and analyzing the load of the day from several aspects.