Improving Clustering-Based Forecasting of Aggregated Distribution Transformer Loadings With Gradient Boosting and Feature Selection

Abstract

Load forecasting is more important than ever to enable new monitor and control functionalities of distribution networks aiming to mitigate the impact of the energy transition. Load forecasting at medium voltage (MV) level is becoming more challenging, because these load profiles become more stochastic due to the increasing penetration of photovoltaic (PV) generation in distribution networks. This work combines medium to low voltage (MV/LV) transformer loadings measured with advanced metering infrastructure (AMI) and machine learning (ML) algorithms to propose a new clustering based day-ahead aggregated load forecasting approach. This four-step approach improves the day-ahead load forecast of a city. First, MV/LV transformer loadings are clustered based on the shape of their load pattern. Second, a gradient boosting algorithm is used to forecast the load of each cluster and calculate the related feature importance. Third, feature selection is applied to improve the forecast accuracy of each cluster. Finally, the day-ahead load forecast of all clusters are aggregated. The case study presented uses 519 measured MV/LV transformer loadings in a city to perform 30 day-ahead load forecasts. Compared against the day-ahead aggregated load forecast without clustering, the average normalized root mean squared error (NRMSE) reduced 12.7 %, the average mean absolute percentage error (MAPE) reduced 18.2 % and the average Pearson Correlation Coefficient (PCC) increased 0.37 %. The 95 % confidence interval of the difference between the average NRMSE, MAPE and PCC without clustering and with the proposed method indicates a statistically significant improvement in accuracy.

Medium‐term forecasting of daily aggregated peak loads from heat pumps using clustering‐based load duration curves to calculate the annual impact on medium to low voltage transformers

“…Figure 8 shows the logarithmic function used as representation function f 2 ( D α ) that describes the correlation most accurately: $$${f}_{2}\left({D}_{\alpha }\right)=0.15\cdot \log \left({D}_{\alpha }\right)+0.26={\overline{H}}_{agg\mathit{. }peak}$$$ A logarithmic relationship was also found for the calculation of the coincidence factor from charging of electric vehicles supporting the general relationship between the slope of an NLDC and the coincidence factor even if the shape of the load profiles are significantly different [45]. This supports the explanation in Section 3 that the proposed calculation of the coincidence factor based on the shape of an NLDC is generalisable to all HPs without the need for all detailed technical properties or load profiles.…”

Load Forecasting of Distributed System Host Based on Linear Time Series Model

“…Moreover, due to energy consumption, output fluctuation and other factors, the node voltage and current are unstable or even serious problems, resulting in an increase in the probability of system collapse or abnormal operation of the host shutdown, thus affecting the stability and reliability of the entire network. Since the time series model method is a random simulation based on probability and statistical characteristics, it does not take into account the impact of time factors on the actual output power change, so this method is subjective to a certain extent [13][14]. But doing so will also cause errors, because the algorithm requires a large amount of data in the calculation process.…”

Control Method of Complex Environmental Engineering Machinery Cooling System Integrating Deep Learning and Feature Clustering

“…Better clustering results than any single cluster clustering result. In order to achieve this goal, it is an important guarantee that cluster fusion has a good and important guarantee to generate cluster members with differences and design a consensus matrix that conforms to data characteristics [9][10].…”