Energy Cost Minimization in Data Centers Using Hybrid CNN–LSTM-Based Workload and Price Prediction under Deregulated Power Markets

Abstract

The growth of cloud computing infrastructures has resulted in significant energy consumption and cost of operation for data centers, especially in deregulated electricity markets with high price volatility and uncertainty. We proposed a hybrid deep learning system combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for predicting the workload and electricity price jointly, which allows the cost-aware and adaptive optimization of electricity. The model utilizes very large and high-resolution time-series data that is directly extracted from realistic workload patterns and deregulated price signals in the market, capturing both temporal dependencies and stochastic price fluctuations. CNN captures short-term workload bursts and local variations, while LSTM layers capture the long-term temporal correlations, which enhance the predictive performance. A dynamic price-aware scheduling mechanism is developed that combines the predicted workload and price signals for making optimal energy consumption decisions in a discrete-time simulation environment. The proposed framework is experimentally validated using extensive data sets and compared to the baseline models, such as LSTM, GRU, and ARIMA, which show that the proposed method has significantly reduced the forecasting error and operational cost. In particular, the model shows significant gains in energy cost savings, peak load reduction, energy efficiency, and high service reliability. Results from statistical validation are significant (p<0.01), and robustness analysis with various stress conditions, such as workload fluctuations and price fluctuations, proves the stability. In addition, multi-objective evaluation by using Pareto analysis demonstrates that the proposed method has a better balance between accuracy, cost efficiency, and stability. The results demonstrate that the proposed CNN–LSTM framework is a scalable, reliable and cost-effective solution for intelligent energy management in modern data centers under dynamic market conditions.