Research On Short-Term Bus Load Forecasting Based On Artificial Intelligence Method

Bus load forecasting is not only the basis of realizing fine management of power grid dispatching,but also an important basis for clearing the power market and forming real-time electricity price.Under the new power system driven by renewable energy,formulating the planned operation mode of power grid not only needs to forecast the total system load,but also needs to analyze the bottom distribution of power load and distributed energy output.Bus load forecasting can provide sophisticated analysis data and results,improving the accuracy of bus load forecasting can effectively promote the consumption of clean energy and reduce the amount of abandoned wind,light and water resources.Therefore,it is an essential part in the demand analysis of power grid regulation and operation.However,the volume of bus load is small,bus load curve trend is weaker than that of system load,and it has the characteristics of nonlinearity and strong fluctuation randomness,so the bus load is difficult to forecast.Therefore,this paper focused on the problem of short-term bus load forecasting,and carried out the research on the method of short-term bus load forecasting based on artificial intelligence.The main work of this paper is as follows:(1)Aiming at the problem of abnormal data in the process of bus load collection and transmission and strong randomness of bus load,the bus load data is preprocessed in two stages.In the first stage,the Chebyshev-Semi-supervised learning-Generative Adversarial Network(GAN)method is used to identify and repair the abnormal bus load data in the process of bus load collection and transmission.In the second stage,the trend characteristics of bus load series are extracted,and the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)in the modal decomposition method and Discrete Wavelet Transform(DWT)and Empirical Wavelet Transform(EWT)in the wavelet transformation method are used to study the bus load decomposition,the input bus load data component is provided for the subsequent forecasting model,and the decomposition results are evaluated by the sample entropy and the variance.(2)In the big data scenario with sufficient bus load data samples,Long Short-Term Memory(LSTM)neural network has the advantage of dealing with long time series forecasting problems.However,when the forecasting object is bus load,there are still problems of insufficient extraction of bus load trend features and difficult determination of hyper-parameters setting.To solve this problem,Bi-directional LSTM(Bi-LSTM)neural network is used to extract the characteristics of the bus load,the mean square error is set as the optimization objective function,and the sparrow search algorithm(SSA)is used to optimize the hyper-parameters of Bi-LSTM network.The components obtained by different data decomposition techniques are forecasted and synthesized respectively,and by comparing to the different benchmark forecasting models to verify the effectiveness of the proposed method.(3)Aiming at the problem of insufficient accuracy and stability of bus load forecasting in small sample scenario,a bus load forecasting method based on DWTMOSMA-SVM is proposed,which takes into account objective functions of forecasting accuracy and stability,and used multiple objective slime mold algorithm(MOSMA)to optimize the penalty factor and kernel function parameters of support vector machine(SVM),a SVM bus load forecasting model suitable for small samples is constructed based on the optimized parameters.DWT,which performed best on the bus data set of this paper,is selected to process the bus load.The bus load data of small samples are input for training and forecasting.The experimental results are compared with different multiobjective optimization algorithms and basic models to verify the effectiveness of the proposed method.(4)Aiming at the problem that bus load forecasting is affected by complex factors,combined with Se2 seq model which is good at dealing with multiple input and output and feature engineering,a multi-feature bus load forecasting method based on Seq2 seq model is proposed.Firstly,the encoder and decoder of seq2 seq model are selected through repeated experiments,and the wind speed,temperature and humidity are selected as the input characteristics of seq2 seq model.Considering the rough granularity of meteorological data,the time continuity periodic coding characteristics are designed,and different feature combinations are set for ablation experiments,The experimental results show that the multi-feature bus load forecasting method based on seq2 seq model proposed in this chapter can take into account the influence of multiple factors on the forecasting,so as to improve the forecasting effect.