Optimal demand response programs selection using CNN‐LSTM algorithm with big data analysis of load curves

Abstract

One of the problems in implementing DRPs is the lack of sufficient understanding of consumers' behaviour when implementing DRPs. This paper compares consumers' load patterns annually by the improved Weighted Fuzzy Average (WFA) K-means clustering method. According to the collected data, DRPs are discussed annually using a combined Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). To make CNN-LSTM a practical algorithm for executing DRPs, Time Series Prediction (TSP) operations must be performed, and input data must be memorized. Finally, according to the TSP done for the obtained data, the implementation of DRPs in practice is facilitated. Then, the DRPs executed with the Deep Learning (DL) model and the power model will be prioritized by Technique for Order of Preference by Similarity to the Ideal Solution (TOP-SIS) method, and decision indicators will be determined and weighted using the Shannon entropy method. The numerical studies section shows that the CNN-LSTM algorithm is able to simulate DRPs with the Mean Absolute Percentage Error (MAPE) almost below 1% in the residential cluster and the highest MAPE in the commercial cluster of 15%. Nevertheless, this algorithm could easily give the same answer as the DRPs power model in the optimal selection of programs.