Thermal energy storage using phase change materials (PCM) offers significant potential for load shifting and grid support, but optimizing these systems remains challenging due to complex multi-objective trade-offs and forecast uncertainty. Researchers have developed an advanced control framework that addresses these limitations through machine learning and evolutionary algorithms.
The load prediction component uses a temporal convolutional network combined with bidirectional long short-term memory architecture enhanced by attention mechanisms. This hybrid neural network model achieves exceptional accuracy with a coefficient of determination of 0.992, reducing prediction errors to 233 watts RMS and 183 watts mean absolute error—substantially outperforming conventional forecasting methods.
For scheduling optimization, the team implemented a multi-strategy particle swarm optimization–differential evolution hybrid algorithm. This approach incorporates nonlinear adaptive parameter tuning, hybrid mutation strategies, and external archive mechanisms to escape local optima and explore the solution space more effectively. Testing demonstrates superior Pareto front performance in both convergence speed and solution distribution compared to traditional optimization algorithms.
The integrated system delivers substantial operational benefits. Optimal scheduling strategies derived from this framework reduce energy consumption by up to 27.6% and operating costs by 28.4% compared to conventional scheduling approaches. These improvements stem from better exploitation of favorable thermal storage windows and more precise load matching.
The research validates the framework across realistic operational scenarios and provides actionable insights for system designers. Potential applications include district heating networks, industrial process heat management, and building energy systems where thermal storage can defer peak loads and improve overall efficiency. As renewable energy penetration increases, such optimization tools become essential for grid operators seeking to maximize the value of distributed thermal storage assets while maintaining system reliability and economic performance.
