Load forecasting is an important basic work to ensure the safety, stability and economic operation of energy system. Its essence is to predict the future load by summarizing the change law between load and time, parameterizing and modeling the law, and learning algorithm based on historical sample data. A short-term electricity, heat and gas load joint prediction method based on deep framework multi task parallel training and learning is proposed. Firstly, the model structure of deep confidence network and multi task learning is introduced. Deep confidence network is an unsupervised learning method for extracting Abstract high-level features, and multi task learning outputs the prediction results as a supervised learning method; Secondly, a multivariate load forecasting system with off-line training and on-line forecasting is established to analyze the input attributes of weather data, historical data, economic data and calendar data, and an index to verify the prediction accuracy of the model is proposed; Finally, the effectiveness of the algorithm is verified by using the actual data of an integrated energy system. The results show that the algorithm has a good application effect in energy demand forecasting.

Integrated energy systems,Deep learning,load forecasting