Neural network prediction and adaptive control collaborative phase change energy storage energy efficiency optimization method

Abstract

A neural network prediction and adaptive control collaborative phase change energy storage energy efficiency optimization method, especially an intelligent ultra-low energy consumption control method combining modular phase change material (PCM) energy storage units and sensor networks. First, the PCM energy storage units and their monitoring systems are planned and deployed, and real-time data such as indoor and outdoor temperature and humidity, solar radiation intensity, etc. are collected through sensors. Second, a building thermal dynamic and phase change process prediction model based on GRU neural network is constructed to predict indoor temperature changes and building thermal load. Third, a multi-modal adaptive control strategy based on the prediction results is used to regulate different PCM units. Finally, a complete system collaboration and feedback optimization closed loop is established, and energy efficiency visualization management and adaptive learning are realized to continuously improve system energy efficiency and user comfort.

Description

Technical Field

[0001] This invention relates to the field of building energy conservation technology, and in particular to a method for optimizing energy efficiency of phase change energy storage through neural network prediction and adaptive control. Background Technology

[0002] With the continuous growth of global energy consumption and the increasing severity of environmental problems, energy conservation and emission reduction in the construction industry have become a key aspect of achieving sustainable development goals. Air conditioning systems are one of the main sources of energy consumption in buildings, especially in residential environments where their use accounts for a significant proportion. Therefore, how to effectively reduce the energy consumption of air conditioning systems while ensuring the comfort of residents has become an important research topic.

[0003] In recent years, phase change materials (PCMs) have been widely used in building energy conservation due to their ability to absorb or release large amounts of latent heat during phase change. PCMs store or release heat by undergoing a phase change within a specific temperature range, thereby effectively regulating indoor temperature and reducing reliance on traditional heating and cooling systems. However, the practical application of PCMs still faces several challenges:

[0004] Lack of precise control: Existing PCM application solutions often fail to fully consider the impact of changes in the external environment on PCM performance, resulting in their inability to perform precise energy management according to actual needs.

[0005] Insufficient predictive capability: Traditional PCM systems lack the ability to accurately predict future indoor and outdoor temperature and humidity, solar radiation and other environmental factors, making it difficult to achieve forward-looking energy dispatch.

[0006] Low level of intelligence: Most PCM systems lack intelligent monitoring and control systems, and cannot adjust the working status of PCM units in real time to adapt to dynamically changing environmental conditions.

[0007] To overcome these challenges, researchers have begun exploring the use of machine learning, particularly deep learning techniques such as Gated Recurrent Unit (GRU) neural networks, to improve the predictive accuracy of building thermal dynamics (PCM) models. GRU is a recurrent neural network specifically designed for processing sequential data, making it particularly suitable for time series forecasting tasks. By training GRU models, future indoor temperature changes and building heat loads can be predicted based on historical data, providing a scientific basis for the optimized operation of PCM systems. Summary of the Invention

[0008] To address the above problems, this invention proposes a method for optimizing the energy efficiency of phase-change energy storage through a combination of neural network prediction and adaptive control. The specific steps are as follows:

[0009] Step 1: Planning and Deploying Modular Phase Change Energy Storage Units and Monitoring Systems

[0010] In conjunction with the building structure, phase change materials are encapsulated into phase change wall panels as independent modular units, which are installed on walls, ceilings, under floors, and in air ducts. A sensor network is deployed, including indoor and outdoor temperature and humidity sensors, solar radiation sensors, and temperature probes embedded inside the phase change units. All sensors collect and upload data in real time with a fixed step size of 15 minutes, providing a complete and synchronous data foundation for subsequent prediction and control.

[0011] Step 2: Construct a neural network-driven prediction model for building thermal dynamics and phase change processes.

[0012] By using historical monitoring data to train a GRU neural network model, it can not only predict indoor temperature changes and building heat load, but also simulate the energy storage / release rate and phase change completion time of each phase change unit under different external conditions.

[0013] Step 3: Execution of prediction-based multimodal adaptive control strategy

[0014] Based on the predictions of the GRU neural network, the adaptive controller performs multimodal control: for phase change units installed in walls and ceilings, the effective input is indirectly controlled by controlling the airflow velocity on their surface, thus adjusting the heat exchange intensity; for phase change units installed in air ducts, the airflow is controlled by adjusting the opening and closing and angle of the air valves, thereby directly determining whether the unit is put into use; for phase change units combined with floor heating / radiant cooling systems, the water flow rate through different energy storage module loops is controlled by adjusting the valves and water pumps in the water circuit, thus achieving energy delivery in different areas and components.

[0015] Step 4: Establish a closed loop for system collaboration and feedback optimization

[0016] The neural network prediction model, adaptive controller, and all the dispersed modular phase change units are integrated into a cohesive system. The controller continuously compares the predicted temperature with the actual temperature and the predicted phase change process with the actual process, updates the control gain and strategy in real time, and dynamically adjusts the actions of each actuator, air valve, water pump, and valve to ensure that the system always maintains the optimal balance between energy consumption and comfort.

[0017] Step 5: Implement visualized energy efficiency management and adaptive learning

[0018] A central management platform is built to display the working status of all phase change units, indoor comfort, system energy consumption and energy efficiency indicators in real time. Based on long-term operating data, the system continuously optimizes the accuracy of the neural network prediction model through machine learning algorithms and adaptively adjusts the control parameters, so that the entire phase change energy storage system can continuously improve itself according to usage habits and seasonal changes, and achieve continuous energy efficiency improvement.

[0019] The present invention provides a method for optimizing the energy efficiency of phase-change energy storage through a combination of neural network prediction and adaptive control. The beneficial effects of this invention are as follows:

[0020] 1. This invention employs an advanced GRU neural network for accurate prediction of indoor temperature and heat load. The PCM energy storage system can more accurately determine when to store or release heat, thereby effectively reducing the operating time and energy consumption of the air conditioning system. This not only reduces the overall energy consumption of the building but also helps achieve energy conservation and emission reduction goals.

[0021] 2. The adaptive control strategy of this invention can dynamically adjust the operating state of the PCM unit according to indoor and outdoor environmental conditions, ensuring that the indoor temperature is always maintained within a comfortable range. Compared with traditional PCM application solutions, this intelligent adjustment method better meets users' needs for the living environment and improves the quality of life.

[0022] 3. The PCM energy storage system proposed in this invention can serve as an effective supplement to renewable energy sources such as solar energy, storing excess energy during the day and releasing it for use at night or on cloudy days. This approach helps balance the grid load and increases the utilization rate of clean energy. Attached Figure Description

[0023] Figure 1 This is a flowchart of the present invention;

[0024] Figure 2 This is a schematic diagram of the GRU model of the present invention. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0026] This invention proposes a method for optimizing the energy efficiency of phase change energy storage through a synergistic approach of neural network prediction and adaptive control. By utilizing a GRU neural network, it achieves accurate prediction of indoor temperature and heat load, optimizing the energy storage and release processes of the PCM unit. This system can significantly improve building energy efficiency, reduce operating costs, and enhance living comfort. It possesses self-learning capabilities, continuously optimizing performance based on historical data, and is applicable to energy-saving retrofits of buildings under various climatic conditions, providing an innovative solution for the development of green buildings. The invention flowchart is shown below. Figure 1 As shown, the steps of the present invention will be described in detail below.

[0027] Step 1: Planning and Deploying Modular Phase Change Energy Storage Units and Monitoring Systems

[0028] In conjunction with the building structure, phase change materials are encapsulated into phase change wall panels as independent modular units, which are installed on walls, ceilings, under floors, and in air ducts. A sensor network is deployed, including indoor and outdoor temperature and humidity sensors, solar radiation sensors, and temperature probes embedded inside the phase change units. All sensors collect and upload data in real time with a fixed step size of 15 minutes, providing a complete and synchronous data foundation for subsequent prediction and control.

[0029] Step 2: Construct a neural network-driven prediction model for building thermal dynamics and phase change processes.

[0030] A GRU neural network model was trained using historical monitoring data, enabling it not only to predict indoor temperature changes and building heat load, but also to simulate the energy storage / release rate and phase change completion time of each phase change unit under different external conditions. A schematic diagram of the GRU model is shown below. Figure 2 As shown.

[0031] Step 2.1 Define input features

[0032] Input feature: Outdoor temperature Outdoor humidity Solar radiation intensity Indoor temperature Phase change unit temperature

[0033] Output characteristics: Building heat load .

[0034] The input data consists of data collected over the previous 120 minutes, with a collection step size of 15 minutes.

[0035] Step 2.2 GRU Gating Mechanism and Hidden State Calculation

[0036] Let the input of the GRU be the eigenvector at time t. The hidden state at the previous moment was ,but:

[0037] Step 2.2.1 Reset the door The output range is [0,1]. The closer the value is to 0, the more the previous state is ignored. :

[0038]

[0039] in, To reset the gate weight matrix; To reset the gate bias vector; It is a sigmoid activation function; for and The splicing operation.

[0040] Step 2.2.2 Update the door The output range is [0,1]. The closer the value is to 1, the more likely it is to retain the previous state. :

[0041]

[0042] in, , These are the weight matrix and bias vector of the update gate, respectively.

[0043] Step 2.2.3 Candidate Hidden State By combining the output of the reset gate with the current input, a candidate state containing the current timing information is generated:

[0044]

[0045] in, The candidate state weight matrix; This is the candidate state bias vector; is the hyperbolic tangent activation function, with an output range of [-1, 1]. This is an element-wise product operation.

[0046] Step 2.2.4 Final Hidden State

[0047] By updating the gate balance of the previous state With candidate state To obtain the hidden state at the current moment:

[0048]

[0049] This hidden state It contains all the time series information from the previous 120 minutes. By inputting it into a fully connected layer, the predicted target can be output.

[0050] Step 2.3 GRU Output Layer Design

[0051] The output layer uses a linear activation function to store the hidden states. Mapped to a single prediction target The formula is:

[0052]

[0053] in, This is the predicted load value; This is the output layer weight matrix; This is the output layer bias vector.

[0054] Step 3: Execution of prediction-based multimodal adaptive control strategy

[0055] Based on the predictions of the GRU neural network, the adaptive controller performs multimodal control: for phase change units installed in walls and ceilings, the effective input is indirectly controlled by controlling the airflow velocity on their surface, thus adjusting the heat exchange intensity. For phase change units installed in air ducts, the airflow through the unit is controlled by adjusting the opening and closing and angle of the air valves, thereby directly determining whether the unit is put into use. For phase change units integrated with floor heating / radiant cooling systems, the water flow through different energy storage module loops is controlled by adjusting valves and pumps in the water circuit, achieving zoned and component-based energy delivery.

[0056] Step 3.1 Define the trigger threshold

[0057] Temperature trigger threshold Set to ±2℃, and set the load trigger threshold to ±50W.

[0058] Temperature deviation is judged as follows:

[0059]

[0060] A temperature greater than 2°C indicates overheating, requiring enhanced energy storage. A temperature below -2°C indicates supercooling, requiring increased energy release. Within the trigger threshold, it indicates that the system has entered a steady state and no further action is required. To set the temperature, The temperature was collected indoors.

[0061] The load is calculated by the output layer of the GRU network in step 3. When the phase transition capacity is insufficient, it indicates that an increase in input is needed; When the phase change capacity is excessive, it indicates that the input needs to be reduced; the controller reads the GRU prediction value every 15 minutes, sends instructions, and finally the actuator responds.

[0062] Step 3.2 Multimodal Control Strategy

[0063] Wall / ceiling unit control:

[0064] Wall / ceiling unit controls the fan speed by adjusting the airflow using a lookup table with the "ΔT - target airflow speed".

[0065]

[0066] Duct unit control:

[0067] The duct unit control determines the effective heat exchange area by adjusting the angle of the air valve, thus prioritizing the offsetting of load deviations.

[0068]

[0069] The floor unit controls the water flow rate to determine the intensity of water-side heat transfer, prioritizing the maintenance of stable indoor temperature.

[0070]

[0071] Step 4: Establish a closed loop for system collaboration and feedback optimization

[0072] The neural network prediction model, adaptive controller, and all the dispersed modular phase change units are integrated into a cohesive system. The controller continuously compares the predicted temperature with the actual temperature and the predicted phase change process with the actual process, updates the control gain and strategy in real time, and dynamically adjusts the actions of each actuator, air valve, water pump, and valve to ensure that the system always maintains the optimal balance between energy consumption and comfort.

[0073] Step 5: Implement visualized energy efficiency management and adaptive learning

[0074] A central management platform is built to display the working status of all phase change units, indoor comfort, system energy consumption and energy efficiency indicators in real time. Based on long-term operating data, the system continuously optimizes the accuracy of the neural network prediction model through machine learning algorithms and adaptively adjusts the control parameters, so that the entire phase change energy storage system can continuously improve itself according to usage habits and seasonal changes, and achieve continuous energy efficiency improvement.

[0075] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A method for optimizing the energy efficiency of phase change energy storage through neural network prediction and adaptive control, comprising the following specific steps, characterized in that: Step 1: Planning and Deploying Modular Phase Change Energy Storage Units and Monitoring Systems In conjunction with the building structure, phase change materials are encapsulated into phase change wall panels as independent modular units, which are installed on walls, ceilings, under floors, and in air ducts. A sensor network is deployed, including indoor and outdoor temperature and humidity sensors, solar radiation sensors, and temperature probes embedded inside the phase change units. All sensors collect and upload data in real time with a fixed step size of 15 minutes, providing a complete and synchronous data foundation for subsequent prediction and control. Step 2: Construct a neural network-driven prediction model for building thermal dynamics and phase change processes. By using historical monitoring data to train a GRU neural network model, it can not only predict indoor temperature changes and building heat load, but also simulate the energy storage / release rate and phase change completion time of each phase change unit under different external conditions. Step 3: Execution of prediction-based multimodal adaptive control strategy Based on the predictions of the GRU neural network, the adaptive controller performs multimodal control: for phase change units installed in walls and ceilings, the effective input is indirectly controlled by controlling the airflow velocity on their surface, thus adjusting the heat exchange intensity; for phase change units installed in air ducts, the airflow is controlled by adjusting the opening and closing and angle of the air valves, thereby directly determining whether the unit is put into use; for phase change units combined with floor heating / radiant cooling systems, the water flow rate through different energy storage module loops is controlled by adjusting the valves and water pumps in the water circuit, thus achieving energy delivery in different areas and components. Step 4: Establish a closed loop for system collaboration and feedback optimization The neural network prediction model, adaptive controller, and all the dispersed modular phase change units are integrated into a cohesive system. The controller continuously compares the predicted temperature with the actual temperature and the predicted phase change process with the actual process, updates the control gain and strategy in real time, and dynamically adjusts the actions of each actuator, air valve, water pump, and valve to ensure that the system always maintains the optimal balance between energy consumption and comfort. Step 5: Implement visualized energy efficiency management and adaptive learning A central management platform is built to display the working status of all phase change units, indoor comfort, system energy consumption and energy efficiency indicators in real time. Based on long-term operating data, the system continuously optimizes the accuracy of the neural network prediction model through machine learning algorithms and adaptively adjusts the control parameters, so that the entire phase change energy storage system can continuously improve itself according to usage habits and seasonal changes, and achieve continuous energy efficiency improvement.

2. The method for optimizing the energy efficiency of phase change energy storage by combining neural network prediction and adaptive control according to claim 1, characterized in that: The neural network-driven prediction model for building thermal dynamics and phase change processes constructed in step 2 can be expressed as follows: Step 2.1 Define input features Input feature: Outdoor temperature Outdoor humidity Solar radiation intensity Indoor temperature Phase change unit temperature ; Output Characteristics: Building heat load ; The input data consists of data collected over the previous 120 minutes, with a data collection step size of 15 minutes. Step 2.2 GRU Gating Mechanism and Hidden State Calculation Let the input of the GRU be the eigenvector at time t. The hidden state at the previous moment was ,but: Step 2.2.1 Reset the door The output range is [0,1]. The closer the value is to 0, the more the previous state is ignored. : ； in, To reset the gate weight matrix; To reset the gate bias vector; It is a sigmoid activation function; for and splicing operation; Step 2.2.2 Update the door The output range is [0,1]. The closer the value is to 1, the more likely it is to retain the previous state. : ； in, , These are the weight matrix and bias vector of the update gate, respectively; Step 2.2.3 Candidate Hidden State By combining the output of the reset gate with the current input, a candidate state containing the current timing information is generated: ； in, The candidate state weight matrix; This is the candidate state bias vector; is the hyperbolic tangent activation function, with an output range of [-1, 1]. For element-wise product operations; Step 2.2.4 Final Hidden State The previous state is balanced by updating the gate. With candidate state To obtain the hidden state at the current moment: ； This hidden state It contains all the time series information from the previous 120 minutes. By inputting it into a fully connected layer, the predicted target can be output. Step 2.3 GRU Output Layer Design The output layer uses a linear activation function to store the hidden states. Mapped to a single prediction target The formula is: ； in, This is the predicted load value; This is the output layer weight matrix; This is the output layer bias vector.

3. The method for optimizing the energy efficiency of phase change energy storage by combining neural network prediction and adaptive control according to claim 1, characterized in that: The execution of the prediction-based multimodal adaptive control strategy in step 3 can be represented as follows: Step 3.1 Define the trigger threshold Temperature trigger threshold Set to ±2℃, and set the load trigger threshold to ±50W; Temperature deviation is judged as follows: ； A temperature greater than 2°C indicates overheating, requiring enhanced energy storage. A temperature below -2°C indicates supercooling, requiring increased energy release. Within the trigger threshold, it indicates that the system has entered a steady state and no further action is required. To set the temperature, Indoor temperature was collected. The load is calculated by the output layer of the GRU network in step 3. When the phase transition capacity is insufficient, it indicates that an increase in input is needed; When the phase change capacity is excessive, it indicates that the input needs to be reduced; the controller reads the GRU prediction value every 15 minutes, sends instructions, and finally the actuator responds; Step 3.2 Multimodal Control Strategy Wall / ceiling unit control: The wall / ceiling unit controls the airflow by adjusting the fan speed and using a table to look up "ΔT - target fan speed"; the duct unit controls the effective heat exchange area by adjusting the damper angle, prioritizing the offsetting of load deviations; and the floor unit controls the water flow rate to determine the water-side heat transfer intensity, prioritizing the maintenance of stable indoor temperature.

Images