Predicting energy consumption of smart military facilities and locations using Deep Learning hyper parameter optimization based on artificial intelligence models

Energy limitation as one of the crises that societies are facing has attracted a lot of attention. Since buildings have a significant contribution to energy consumption, the problem of predicting energy consumption of buildings has become a necessity in the field of energy efficiency. In order to achieve an effective solution to solve this problem, a system based on stacking was proposed in the form of the first proposed model, which uses XGBoost and MLP methods in the first level so that it can enjoy the advantages of both methods. The predictions made by each of these methods, in the second level, are used as input for another XGBoost algorithm, which is considered as a meta-learner. The meta-parameters of the baseline techniques were optimized using successive halving search. Further, the research results showed that to solve this problem, deep learning methods have a significant performance compared to other methods. In this way, in the form of the second proposed model, a vote-based solution was presented in which three CNN models with different structures and a DNN method were used. According to the good results recorded by each of the deep learning methods, the second proposed model can be proposed as the final model and provide a robust system to solve this problem. The proposed models were applied to the WiDS Datathon dataset and provided good performance. The first proposed model achieved a value of 0.34 in the R2 criterion. The value of R2 in the best learning method of the second proposed model is equal to 0.34, but the second proposed model, which will be a robust model against new data, has achieved a value of 0.33.

Predicting energy consumption of smart military facilities and locations using Deep Learning hyper parameter optimization based on artificial intelligence models Keywords:

Artificial Intelligence , Deep Learning , CNN , DNN , Energy Consumption , Stacking , Regression