Load Forecasting of Purchasing Agent Customers Based on Dynamic Adjustment Mechanism of TOU Price

In order to accurately predict the load of power purchasing agents, the influence of multiple factors such as electricity price, weather condition, week, month and holiday/working day on the load was analyzed. Principal component analysis (PCA) was used to reduce the dimension of these factors and eliminate the redundancy of the original sequence. The influence of holiday/working day attributes and weather conditions on load characteristics was analyzed, and the construction strategy of load prediction model candidate dataset (CD) was proposed. On this basis, the dynamic time modeling of multi-variable energy load prediction was carried out by using short term memory (LSTM) network model, and the proposed prediction model was verified by the measured data. The analysis results show that: electricity price, weather condition, week, month, holiday/working day, the cumulative contribution rate of 5 principal components is 85.54%, and they are the main influencing factors of load. After using the candidate data set construction strategy and PCA processing, the pertinence of the training set data is improved, and the accuracy of the prediction results is also improved, with the average absolute percentage error less than 13.03% and the root-mean-square error less than 3.86. The proposed prediction model has good applicability in the load prediction of different typical scenarios.