Wind Power Prediction Based On Multivariate Feature And Economic Scheduling With Uncertainty Of Source And Load

The rapid penetration of renewable energy which represented by wind power in the power grid is the inevitable requirement of socio-economic development.Compared with traditional thermal power generation,the output power of wind turbine has strong fluctuation.Therefore,the accurate prediction of wind power is an important means to improve the permeability of wind power and ensure the safe operation of power grid.In this paper,short-term load prediction and wind power prediction are carried out for large-scale connection of wind farms with power grid from the perspective of supply and demand respectively.On this basis,the influence with dual uncertainty of source and load on day-ahead scheduling of power system is discussed.The main work and innovation achievements are as follows:Based on the analysis of the forecasting requirements and characteristics of power load time series,an improved model of Elman neural network is proposed,which not only simplify the network structure and training process,but also considering the influence factors of time,weather and society on load forecasting.In view of the difficulty of parameter optimization in neural networks,an improved grey wolf optimization algorithm(CIGWO)based on random cosine function and chaotic mapping is proposed.The combined model(CIGWO-Elman)is applied to the short-term prediction of actual load demand in a certain area.The numerical results show that the load prediction model proposed in this paper has better prediction accuracy than the basic Elman network and the neural network improved by the original grey wolf optimization(GWO),particle swarm optimization(PSO)and differential evolution(DE)algorithms.In view of the high fluctuation of wind power output,the characteristic attributes that affect wind power output are analyzed and studied based on multivariate characteristics,so as to better learn the potential rule of wind power output under strong randomness.Firstly,a hierarchical wind pattern classification model was established through data discretization and model classification,and the accuracy of the prediction model was improved by reducing the fluctuation range of wind power.Secondly,an improved random forest method based on Poisson distribution is proposed to realize the parallel sampling process,and the model is applied to the selection of multivariate characteristics that affect the output of wind turbines.Taking the SCADA system that actually put into operation in a wind farm as the research object,two new characteristics of equivalent wind speed variation rate and yaw coefficient were proposed through data analysis,further reflecting the influence of wind speed and direction variation trend on wind power.Finally,the characteristics of wind power dataset were extracted from the time-frequency domain by the complete EEMD with adaptive noise(CEEMDAN)method and Pearson correlation verification.The result of the example shows that the submodels based on wind regime classification have better prediction performance combined with feature selection and feature construction.On the basis of the above analysis of multi-angle and multi-dimensional characteristics,the idea of multi-horizon wind power prediction modeling is proposed.That is,historical wind power time series and characteristic variables are all taken as the influence characteristics of wind power output in the target period,so that the prediction model can learn trend information from input characteristics of different combinations and forms.Based on Transformer network,propose an improved gated unit(T-GRUs)which can handle non-uniformly spaced time series and a multivariable attention mechanism based on time convolutional network(TCN).Taking them as the input layer of encoder and decoder,it is helpful for the prediction model to learn the important information inside the feature set,so as to obtain more accurate prediction results.The numerical results show that compared with the features contained in the simple historical time series,the wind power prediction algorithm considering multi-horizon has better performance,especially when the wind power changes at adjacent time points,the proposed algorithm has higher prediction accuracy.Aiming at the influence of uncertainty with load demand and wind power output on power system scheduling,an economic scheduling model based on Wasserstein’s metric and distributionally robust optimization algorithm was proposed.The confidence interval of load demand and the fuzzy set of wind power prediction error are constructed respectively to achieve the formal description with daul uncertainty of source and load.On this basis,a two-stage economic dispatching model of power system with uncertainty and chance constraint is established.By means of Wasserstein’s metric features,piecewise affine function mapping,and the reconstruction and approximation of conditional value-at-risk(CVaR)driven by wind power error data,the difficult-to-solve infinite dimensional optimization problem is transformed into a linear programming problem.Then,a hybrid algorithm based on simplex method and improved differential evolution algorithm is proposed to solve the simplified economic dispatching model of power system.Combined with the simplex method’s direct and simple features,and the differential evolution algorithm’s outstanding memory function,the hybrid algorithm has better optimization ability.This algorithm is applied to solve the classical power system scheduling optimization problem.The results show that under the premise of uncertainty between load demand and wind power output,Wasserstein’s measurement radius,sample size of random variables and the performance of optimization algorithm all have a certain influence on the optimization of target cost.The short-term load prediction algorithm,wind power day-ahead prediction model and the economic dispatching model of power system with uncertainty of source and load proposed in this paper.It can promote the consumption of wind power,and play a positive role in promoting the construction of wind power,photovoltaic power generation and other new energy sources.