Fig . 3 The absolute error distribution for the first 228 days of 2022 . The error is calculated as the absolute value of the difference between the total PV yield predicted and measured for a single day . determined and subsequently applied to cases with real PV and load data . This implies that the storage charge / discharge profile remains unchanged , while the real PV and load profiles are incorporated into the system . The energy drawn from the grid and fed back into , it is extracted based on the balance between generation and consumption . This process helps to emphasize the accuracy of prediction methods and identify any substantial difference in predicted peaks that could significantly impact the optimization algorithm . The monthly results ( shown in Fig . 5 ) revealed that , interestingly , the difference between the case with actual PV and load profiles and the case with predicted PV and load is relatively small .
Using only the predicted load data as input for the optimization results in a difference of less than 2 % for all methods , while the best-performing forecasting method in this case ( XGBoost - gbtree ) demonstrates a difference of merely 0.4 % compared to actual values . This emphasizes that employing specific load prediction methods can help reduce consumers ’ electricity bills . As depicted in Fig . 5 , incorporating PV predicted values into the optimization model could lead to an error of approximately 9 %. When both PV and load predictions are added to the optimization , the error ranges from 10-14 %, which is still considered acceptable . This underscores the significant impact of PV yield predictions on consumers ’ electricity savings and highlights the importance of selecting the prediction method carefully . The optimization procedure was carried out for an entire month day by day , and a box plot illustrating the error in the saved amount of electricity bills for the examined forecasting methods combined with optimization method is presented in Fig . 6 .
Fig . 4 Predicting the electric load of the first household in the Open Fluvius dataset for Tuesday , November 8 with the top 5 performing methods . The red line represents the ground-truth and blue the prediction .
Without a PV system and integrated battery , the consumer ’ s monthly electricity bill is € 17 , based on the dynamic tariff . However , by utilizing both the PV system and battery in conjunction with maximizing self-consumption strategies , the consumer can receive a € 47 cashback from the power distributor ( saving of € 64 ). When the electricity rate is flat , optimization does not lower the cost of electricity when compared to the maximum self-consumption approach , and it is not profitable to use this optimization method ( it can only save € 5 / month ). However , when the hourly electricity rate is considered , optimization lowers the monthly electricity cost and saves € 22 more than maximizing self-consumption ( 43 % extra saving ), proving to be advantageous in this case .
Next , the load and PV yield predicted by artificial intelligence is provided as input to the optimization problem to determine the optimal charging / discharging of the battery . For each scenario , the optimal battery charge / discharge profile is
Fig . 5 Comparison of the forecasting methods ’ accuracy in terms of electricity saving .
Fig . 6 Comparison of the forecasting methods ’ accuracy in terms of electricity saving .