Research On Short-term Load Forecasting In Open Market Environment Of Power Retail Side

Short-term power load forecasting has a significant influence in the dispatching,production and planning of power system.Accurate load forecasting is helpful for decision maker to make correct decision plan and it is suitable for the stable operation of power system.In the open market environment of the electricity retail side,the market competition of multiple electricity retail subjects brings the fluctuation of electricity price,which makes the electricity consumption behavior of users guided by electricity price change accordingly,making the load curve more random and complex and bringing more uncertainty to short-term load forecasting.Existing deterministic point prediction methods cannot effectively simulate and capture the change characteristics of load,and cannot accurately provide comprehensive predictive value information.Therefore,this paper proposes to study the short-term load forecasting under the open market environment of the electricity retail side,and proposes a short-term load forecasting method of multiquantile outlier robust extreme learning machine(MQR-ORELM)based on electricity price-oriented demand response.Firstly,considering the influence infactors of load forecasting.In addition to the traditional load forecasting factors such as meteorological factors and daily type factors,the electricity price-oriented demand response influencing factors are also considered,and an electricity price-oriented demand response model based on the real-time electricity price is built to evaluate the influence.In this way,the real-time electricity price-oriented demand response factor is obtained,preparing for the next step of model prediction.Then,quantile regression and outlier robust extreme learning machine(ORELM)are combained together to realize the prediction of the multi-segmentation scene.The model predefine different quantile values to realize the transformation of prediction scenarios,so as to capture more uncertain factors into the prediction process and obtain the prediction results under multi-quantile.At the same time,the model preserves the robustness of ORELM and better deals with the random interference factors in the electricity retail side.What's more,considering the interference components in the load curve,In order to extract more accurate load modes,a time-varying filtering mode decomposition(TVFEMD)method carried out on the historical load curve decomposition.Then each submodel is abtained by the proposed method,and the final prediction results is obtained by superimposing individual predicted values.In order to provide more comprehensive and intuitive predictive value information,this paper adopts the prediction results of multiple quantile models with double kernel probability density estimation to form the probability density picture,which intuitively displays the predicted results of all possibilities and the probability of occurrence.Eventually,the data published by AEMO of Australia is used to verify the validity of the proposed model.Results show that: 1)Proposed electricty price-oriented demand response model can effectively realize the estimation of real-time electricity price-oriented response demand degree.2)Compared with the benchmark model,the prediction accuracy of the MQR-ORELM proposed in this paper is higher,and it can achieve high-precision prediction in multi-quantile scenarios.3)The proposed probability density estimation can intuitively and effectively show the probability distribution of all predicted values,and further verify the validity of the predicted accuracy of the MQR-ORELM model.4)The simulation analysis of holiday and seasonal load verifies the general applicability of the model to special load curve.