Aiming at the prediction model accuracy limitation caused by the difference of peak-valley coupling strength between multiple loads in different periods

paper proposes a deep deterministic policy gradient algorithm based on an integrated energy system multi-load short-term forecasting method with single-task learning (STL)+multi-task learning (MTL)+DDPG adaptive dynamic combination. First

the peak and valley characteristics of multiple electrical

cold and thermal loads were clarified

and the coupling relationship between meteorological data and electrical

cold and thermal loads was extracted from time to time based on the Pearson correlation coefficient to provide effective data supports for the construction of subsequent prediction models. Secondly

in order to solve the problem of peak-valley difference of coupling features in MTL model

a relatively independent STL model was constructed to extract load data from time series features

and at the same time to solve the sensitivity interference of environmental factors on multivariate load changes. Then

the MTL and STL prediction data were reconstructed

and DDPG was used to build a dynamic weight adaptive allocation model

which fitted the prediction results of the sub-model

sensed the external environment changes

and realized the optimal combination of prediction models from moment to moment. Finally

the IES of Arizona State University in Tempe

USA

is taken as an example

the results show that the average absolute percentage error of the proposed method in the forecast of electrical

cold and thermal loads reaches 0.63%

0.98% and 1.08%

respectively

and the prediction accuracy is greatly improved compared with other prediction models.