How to use Artificial Intelligence to Predict and Optimize Thermal Energy Storage Systems ?

Keywords: Thermal Energy Storage (TES); Artificial Intelligence; Energy efficiency; Optimization; Model Predictive Control (MPC); Artificial Neural Networks (ANNs)

Introduction to Thermal Energy Storage Systems (TESS)

Thermal Energy Storage Systems (TESS) are advanced technologies that store thermal energy for later use in heating, cooling, or power generation. They help manage energy by storing excess energy when it's available (like from renewable sources) and releasing it when needed, improving energy efficiency and reducing reliance on traditional power sources. TESS comes in different types, storing energy through temperature changes, phase transitions, or chemical reactions. These systems are crucial in various sectors and play a key role in optimizing energy usage and promoting sustainability.

Optimizing Thermal Energy Storage Systems (TESS) for energy efficiency holds significant importance for several reasons:

1. Enhanced Renewable Integration: TESS optimization enables better integration of renewable energy sources like Concentrated Solar Power (CSP) by storing surplus energy and releasing it when demand peaks. This maximizes the utilization of renewable energy, reducing dependency on non-renewable sources.
2. Energy Cost Reduction: Efficient TESS management leads to reduced energy consumption during peak hours, minimizing reliance on expensive energy sources. This, in turn, lowers energy costs for consumers and businesses.
3. Grid Stability and Reliability: TESS optimization contributes to grid stability by balancing energy supply and demand. It helps mitigate fluctuations, ensuring a more reliable and stable power supply, especially during high-demand periods.
4. Environmental Impact: By optimizing TESS, overall energy consumption decreases, resulting in reduced greenhouse gas emissions and environmental impact. It aligns with sustainability goals and promotes a greener energy infrastructure.
5. Improved Operational Flexibility: Optimized TESS systems offer increased flexibility in managing energy needs. They can respond quickly to varying energy demands, providing heating, cooling, or power as required, enhancing overall operational efficiency.
6. Long-term Sustainability: As the world aims for sustainable energy solutions, optimized TESS contributes to a more sustainable energy landscape. It facilitates efficient utilization of resources and reduces waste in the energy generation and consumption process.

Revolutionizing Thermal Energy Storage: AI's Role in Prediction and Optimization

The integration of Artificial Intelligence (AI) into Thermal Energy Storage Systems (TESS) revolutionizes energy management by leveraging predictive analytics, machine learning, and adaptive control. AI algorithms, ranging from neural networks to genetic algorithms and particle swarm optimization, enable TESS to forecast energy demand, optimize charging and discharging rates, and identify efficient system configurations. This integration ensures precise energy utilization, reduces operational costs, and enhances grid stability by enabling adaptive responses to changing environmental conditions and energy needs. Ultimately, AI implementation in TESS represents a pivotal leap toward sustainable energy practices, promising greater efficiency and resilience in energy storage and distribution systems.

In the figure below, we will cover the background of different AI methods commonly used in TESS.

Optimization Techniques

Particle Swarm Optimization (PSO):

• Concept: Inspired by the social behavior of bird flocking or fish schooling, PSO is a metaheuristic optimization algorithm.
• Application: PSO is used to find the optimal solution in a search space by iteratively adjusting a population of candidate solutions based on their fitness.
• In TESS: PSO can be applied to optimize parameters within thermal energy storage systems, such as determining the optimal charging and discharging rates for energy storage mediums, maximizing energy efficiency, and minimizing losses.

Genetic Algorithm (GA):

• Concept: Based on the principles of natural selection and genetics, GA mimics the process of evolution to find optimal solutions to problems.
• Application: GA operates by evolving a population of potential solutions through selection, crossover, and mutation to reach an optimal or near-optimal solution.
• In TESS: GA can be employed to optimize TESS configurations, such as determining the best material for energy storage, finding optimal system configurations, or optimizing control strategies for energy release.

Predictions Techniques

Artificial Neural Networks (ANN):

• Concept: ANNs consist of interconnected nodes (neurons) organized in layers (input, hidden, output) that process information. They learn by adjusting connection strengths between neurons.
• Application: ANNs are used for pattern recognition, classification, regression, and complex data processing tasks.
• In TESS: ANNs can be applied to model complex relationships within thermal energy storage systems. They learn from historical data, weather patterns, and system behavior to predict future energy storage performance, optimize charging/discharging strategies, or identify optimal system configurations for improved efficiency.

Adaptive Neuro-Fuzzy Logic (ANFIS):

• Concept: Combines neural networks and fuzzy logic to create adaptive systems that can learn and make decisions based on input data.
• Application: ANFIS adapts and learns from input-output data relationships, creating fuzzy inference systems that model complex relationships.
• In TESS: ANFIS can be utilized to model and predict thermal energy storage behavior based on historical data, weather patterns, and system parameters, aiding in forecasting energy storage performance.

Support Vector Machine (SVM):

• Concept: SVM is a supervised learning model used for classification and regression analysis.
• Application: SVM works by finding the optimal hyperplane that best separates data into different classes or predicts continuous values.
• In TESS: SVM can be employed for predictive modeling in TESS, assisting in tasks such as predicting optimal charging/discharging cycles, temperature profiles, or energy storage behavior based on various input parameters.

Case Study

Prediction and optimization of building energy systems integrated with the TES component :

This study aims to propose a Model Predictive Control (MPC) scheme using AI to optimally control building energy systems integrated with the TES component. The feasibility of the proposed AI-based MPC scheme was investigated by the experimental system by comparing the results with a classical Rule-Based Control (RBC) scheme that prioritizes TES charging and discharging operations.

The experimental system was operated by five different operation modes as follows: the stop and standby status, TES charging mode, cooling mode by TES discharging operation, cooling mode by TES discharging and chiller operation in parallel, and cooling mode by chiller operation. When the system was operated by the TES charging mode, the outlet temperature of the chiller for producing chilled water was fixed at 5 ?C. Otherwise, during the cooling mode, the chiller outlet temperature was set at 7 ?C. The above different operating modes were implemented in the experimental system automatically by controlling the on-off status of the system component and the open-close status of 2-way valves (MV1–MV4) to switch the water loop flow in the system based on the control table described in Table 3.

As a baseline for comparing the performance of the developed AI-based Model Predictive Control (MPC) strategy, the experimental system was also operated by the conventional RBC strategy. The operational flowchart from the RBC strategy can be found in Fig. 3. The RBC strategy was defined as a priority-based control to prioritize the charging and discharging operation of the TES.

To construct a reliable and computationally manageable MPC controller, the system’s behaviors under varying operational disturbances were predicted by Artificial Neural Network (ANN) models.

Predictions using Artificial Neural Network (ANN)

The behavior of five different prediction targets was predicted from ANN models during 4 hours. The prediction targets were the water tank temperature during high, medium and low cooling demand, the outdoor temperatures, the chiller power consumption and pumps on the primary side.

Water tank temperature prediction :

The control results of the water tank temperature on the secondary side in a high-level cooling demand schedule during the day are shown in Fig. 4. During the cooling mode, the AI-based MPC strategy controlled the water tank temperature near the cooling set temperature of 26 ?C, whereas, in the RBC strategy, the water tank temperature greatly deviated from the setpoint temperature because a high thermal load level continuously occurred but the TES tank cooling capacity was limited, and the discharging operation was conducted simultaneously with the cooling operation due to the predefined operation rules.

As shown in Fig. 5, under the medium-level cooling load condition, the water tank temperature on the secondary side was stably controlled near 26±0.5 ?C in both the RBC and AI-based MPC strategies. The secondary water loop was controlled by a local PID controller, and a slight delay and the oscillation of the temperature near the setpoint occurred in both cases.

When the low-level cooling load profile was tested, both the RBC and AI-based MPC strategies stably regulated the water tank temperatures near the cooling setpoint temperature of 26 ?C. Because the assumed cooling demand profiles during a day were on average slightly small considering the nominal capacity of the chiller and TES tank, the water tank temperature tended to be controlled in a region slightly lower than the cooling setpoint temperature of 26 ?C, as shown in Fig. 6.

Outdoor, tank temperatures and chiller power predictions :

As a result, there was an error identified between the measured outdoor temperatures and forecasted outdoor temperature from the National Weather Service as shown in fig.7. The Mean Square Error (MSE) value between the measurement and forecasted data from the National Weather Service was 0.49 ?C, but the largest error was approximately 2 ?C. However, the high prediction accuracy from the ANNs was confirmed during the MPC implementation phase. The MSE value between the prediction from the ANN model and the measured data was 0.06 ?C for the water tank temperature in fig.8 and 0.13 kW for the chiller power consumption in fig.9.

Discussion

The application of an AI-enhanced Model Predictive Control (MPC) framework for managing a building's energy system, incorporating Thermal Energy Storage (TES), was examined in this study. The feasibility for cooling applications was substantiated through experiments conducted on a scaled-down mockup system. Comparative analysis between the AI-driven MPC and a Rule-Based Control (RBC) approach, which focuses on optimizing TES operations, underscored the efficacy of the AI-based MPC. In summary, this study's findings emphasize the superior performance and potential advantages of employing AI techniques within MPC for optimizing TES operations in building energy systems.

The experiment aimed to validate the efficacy of the AI-based Model Predictive Control (MPC) strategy by subjecting it to three distinct cooling load profiles: high, medium, and low levels during daytime operations.

The conclusions of this case study can be summarized as follows:

• Using an AI-driven MPC controller, the study optimized the charging and discharging rates of the Thermal Energy Storage (TES) system, comparing its performance with the Rule-Based Control (RBC) strategy.
• The results highlighted the efficiency of the AI-based MPC in optimizing TES operations, showcasing adaptability in adjusting TES charging and discharging rates based on varying cooling load profiles, a capability lacking in the RBC strategy.
• Consequently, the study concludes that the AI-based MPC offers feasible real-time optimal control, presenting a promising approach for efficiently managing TES systems in practical applications.

Study limitation

The experimental analysis of the proposed AI-based MPC strategy must be verified in actual full-scale buildings with multizone conditioning. Also, this study defined the RBC strategy not to consider operation mode 3 not to conduct the cooling operation by TES discharging and chiller operation simultaneously.

The results of the AI-based MPC strategy can be compared by the different control logic of the RBC strategy with more complexity in further studies. Further, an accelerated experiment that shortens the day from 24 hours to 4 hours was conducted. In designing the MPC controller, it is important to carefully set the parameters of the prediction time horizon and the control timestep. However, such design parameters were not fully investigated due to the restrictions of the accelerated experiment.

One strong advantage of the accelerated experiment is that the system can be tested for many more iterations, and the characteristic behavior of the system during different operation modes can be captured within a short time duration.

Also, operational disturbances that affect the AI-based MPC strategy implementation for building energy management, such as occupant-behavior-related disturbances and price-related disturbances, were not fully considered because this study assumed these conditions to be perfectly known in advance.

Conclusion

Thermal Energy Storage Systems (TESS) represent a critical advancement in energy management, offering a dynamic solution to store and distribute thermal energy for various applications. The incorporation of Artificial Intelligence (AI) techniques into TESS introduces a transformative paradigm shift, leveraging predictive analytics, machine learning, and adaptive control mechanisms.

Through advanced algorithms such as neural networks, genetic algorithms, particle swarm optimization, and adaptive neuro-fuzzy logic, AI empowers TESS to forecast energy demands accurately, optimize charging and discharging cycles, and fine-tune system configurations. This synergy enhances precision in energy utilization, reduces operational costs, and fortifies grid stability by enabling responsive adaptations to fluctuating energy requirements.

The experimental validation of AI-based Model Predictive Control (MPC) strategies underscores the potency of AI in TESS optimization. This validation highlights the efficacy of AI-driven methodologies in real-world scenarios, paving the way for widespread adoption and implementation of AI-integrated TESS solutions.

In essence, the convergence of AI and TESS marks a monumental leap toward sustainable energy practices, promising not only enhanced efficiency and cost-effectiveness but also resilience and adaptability in meeting evolving energy needs. This synthesis is poised to redefine the landscape of energy storage and distribution, steering us toward a more sustainable and eco-conscious future.

Acknowledgement

I want to acknowledge that this article is a synthesis of the insights and methodologies presented in the two referenced articles below. I have combined and adapted the ideas from these sources to discuss the integration of AI techniques into Thermal Energy Storage Systems. My aim was to consolidate and present these concepts in a coherent manner for a broader understanding of AI's role in optimizing TESS.

References

• Doyun Lee, Ryozo Ooka, Yuki Matsuda, Shintaro Ikeda, Wonjun Choi. Experimental analysis of artificial intelligence-based model predictive control for thermal energy storage under different cooling load conditions, Sustainable Cities and Society, Volume 79, 2022, 103700, ISSN 2210-6707, https://doi.org/10.1016/j.scs.2022.103700.
• A.G. Olabi, Aasim Ahmed Abdelghafar, Hussein M. Maghrabie, Enas Taha Sayed, Hegazy Rezk, Muaz Al Radi, Khaled Obaideen, Mohammad Ali Abdelkareem. Application of artificial intelligence for prediction, optimization, and control of thermal energy storage systems, Thermal Science and Engineering Progress, Volume 39, 2023, 101730, ISSN 2451-9049, https://doi.org/10.1016/j.tsep.2023.101730.