Sunlight angle dynamic tracking and energy optimization control method for photovoltaic integrated curtain wall

Abstract

The invention discloses a sunlight angle dynamic tracking and energy optimization control method for a photovoltaic integrated curtain wall. The method comprises the steps of erection of the photovoltaic integrated curtain wall, deployment of a monitoring module, preprocessing of monitoring data, construction of a neural network model, energy optimization control, model feedback optimization and weight coefficient dynamic adaptation based on seasons. An outer glass layer, a rotatable photovoltaic panel array, a dimming glass layer and a cavity ventilation module are adopted to form a photovoltaic integrated curtain wall, so that the attractiveness of a building is realized, the maintenance burden is reduced, natural lighting is optimized, and the illumination energy consumption is reduced; based on climate characteristics of different seasons, collaborative optimization is carried out on energy consumption of air conditioners, illumination and heating through a neural network model, photovoltaic power generation benefits are maximized while comprehensive energy consumption of a building is remarkably reduced, and meanwhile the long-term dynamic adaptability of the system is improved.

Description

Technical Field

[0001] This invention belongs to the field of building curtain wall technology, and relates to a method for dynamic tracking and energy optimization control of solar angle for photovoltaic integrated curtain walls. Background Technology

[0002] As an important component of a building's external envelope, the thermal insulation performance of building curtain walls directly affects building energy consumption. Photovoltaic integrated curtain walls, by combining the building's facade with photovoltaic power generation, have become a key pathway to reducing building energy consumption.

[0003] Existing photovoltaic integrated curtain walls mostly generate electricity by integrating photovoltaic modules on the exterior surface of buildings. However, in practical applications, the following technical bottlenecks still exist: 1) Fixed photovoltaic panel angles limit power generation efficiency: The photovoltaic modules of traditional photovoltaic integrated curtain walls are mostly installed at fixed angles, which cannot dynamically adapt to the real-time changes in solar altitude angle and azimuth angle, and the photoelectric conversion efficiency needs to be improved; 2) Conflict with building aesthetics: Photovoltaic integrated curtain walls are exposed to the outdoor environment for a long time, and the surface is prone to dust and rain stains, affecting the aesthetics of the building facade; 3) Conflict with the building's natural lighting function: Some photovoltaic integrated curtain walls adopt a high coverage design to ensure power generation, sacrificing the building's natural lighting needs, resulting in increased indoor lighting energy consumption; 4) Conflict with the building's thermal insulation function: There is a contradiction between the thermal insulation performance of photovoltaic panels and the building's thermal insulation requirements. In summer, the heat absorbed by photovoltaic panels under strong sunlight is easily conducted into the room, resulting in an increase in air conditioning load. In winter, the blocking of sunlight into the room will lead to an increase in heating energy consumption.

[0004] Among existing patents, the "A Ventilated Glass Curtain Wall System with Built-in Power Generation Phase Change Integrated Shading Louvers" (CN222314327U) achieves multifunctionality and energy efficiency improvement of building curtain walls by integrating photovoltaic power generation, phase change material cooling, and passive ventilation mechanisms. However, it still has shortcomings such as lacking a mechanism that relies on preset parameters rather than real-time dynamic monitoring and control, only roughly reducing heat load through ventilation modes rather than considering comprehensive energy efficiency balance, and lacking learning and adaptability, which makes the system unable to adapt and adjust itself. This limits the performance of the patent in seasonal changes and complex dynamic environments.

[0005] In summary, existing photovoltaic integrated curtain walls have technical shortcomings in dynamic angle tracking, pollution protection, multi-system collaborative control, and functional integration, making it difficult to balance power generation efficiency, building energy consumption optimization, and indoor environmental comfort. Therefore, there is an urgent need for a photovoltaic integrated curtain wall control method that can achieve dynamic adaptation to sunlight angle, collaborative operation of multiple energy consumption systems, functional integration, and learning and adaptability. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method for dynamic tracking and energy optimization control of solar integrated curtain walls. By real-time linkage adjustment of the photovoltaic panel tilt angle, light transmittance, and active heat exchange system, the method minimizes the building's overall energy consumption and maximizes photovoltaic production capacity.

[0007] A method for dynamic tracking and energy optimization control of solar integrated curtain walls, comprising the following steps:

[0008] S1. Installation of photovoltaic integrated curtain wall:

[0009] The photovoltaic integrated curtain wall comprises, from the outside to the inside, an outer glass layer, a rotatable photovoltaic panel array, and a dimming glass layer, and is equipped with a cavity ventilation module. The outer glass layer uses ultra-clear tempered glass as the substrate, with a modified polypropylene coating of low surface energy to ensure that the surface of the outer glass layer is not easily contaminated by various liquids and dust. The rotatable photovoltaic panel array includes a louvered support frame, several photovoltaic panels, and several rotation drive devices. The louvered support frame consists of parallel longitudinal fixing frames and transverse support beams. The longitudinal fixing frames have shaft holes for installing the photovoltaic panel rotation shafts, and both ends are fixed to the main building structure through embedded parts. The length of the transverse support beams is appropriate to the width of the curtain wall. The photovoltaic panels are rigidly connected to the longitudinal fixing frame via bolts. The photovoltaic panels are monocrystalline silicon solar cell modules, with their center ends inserted into the shaft holes of the longitudinal fixing frame via rotating shafts, allowing them to rotate freely around the shaft. Air circulation channels are reserved between adjacent photovoltaic panels. A rotation drive device is installed on the side of the longitudinal fixing frame to drive the photovoltaic panels to rotate around the shaft. The dimming glass layer is made of electrochromic dimming glass, with its overall dimensions adapted to the curtain wall frame, used to adjust light transmittance and color according to the applied voltage. The cavity ventilation module is located in the cavity formed between the outer glass layer and the dimming glass layer, including a ventilation cavity, a bottom air inlet, and a top air outlet.

[0010] S2. Deployment of the monitoring module:

[0011] The monitoring module includes a solar tracking sensor, a temperature and humidity sensor, a illuminance sensor, a photovoltaic panel temperature sensor, a photovoltaic panel current sensor, and an energy consumption monitoring unit. The solar tracking sensor is a dual-axis digital solar angle sensor, installed at the top center of the outer side of the curtain wall, used to collect solar altitude angle and solar azimuth angle data. The temperature and humidity sensor and the illuminance sensor are arranged inside the building, used to collect indoor temperature, indoor humidity, and indoor illuminance data, respectively. The photovoltaic panel temperature sensor and the photovoltaic panel power generation sensor are both installed on the back of the photovoltaic panel, used to collect photovoltaic panel temperature and photovoltaic panel power generation data, respectively. The energy consumption monitoring unit uses a smart meter to measure air conditioning energy consumption, lighting energy consumption, and heating energy consumption data by area.

[0012] S3. Preprocessing of monitoring data:

[0013] The raw data collected by the monitoring module is preprocessed to construct a multi-dimensional feature vector, which provides good input features for the neural network model. The multi-dimensional feature vector contains 13 parameters: solar altitude angle, solar azimuth angle, indoor temperature, indoor humidity, indoor illuminance, photovoltaic panel temperature, photovoltaic panel power generation, air conditioning energy consumption, lighting energy consumption, heating energy consumption, indoor area of ​​the building, time of day, and season.

[0014] S4. Construction of Neural Network Model:

[0015] The neural network model architecture includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the multidimensional feature vector, and it is used to map the feature vector to the hidden layer for feature extraction. The hidden layer includes four backbone feature extraction layers and one fully connected layer. The backbone feature extraction layers use gated spatiotemporal convolutional units and capture temporal dependencies using the ReLU activation function. The fully connected layer uses the LeakyReLU activation function to integrate the high-dimensional features output by the backbone feature extraction layers, enhancing the fitting ability to nonlinear relationships. A dropout mechanism is introduced in the connections between hidden layers to suppress overfitting during model training. The output layer contains three neurons and is used to output target control values. The target control values ​​include the photovoltaic panel rotation angle, the exhaust volume of the cavity ventilation module, and the light transmittance of the dimming glass layer.

[0016] The training data for the neural network model comes from a historical monitoring database accumulated over a long period of operation. A weighted multi-objective loss function is used during training, as shown in the following expression:

[0017] L total =β1·E air +β2·E light +β3·E heat -β4·E pv +β5·L MSE

[0018] Among them, L total Let L be the total loss function. MSE E represents the mean square error between the predicted and actual values. air E light E heat These are the predicted energy consumption for air conditioning, lighting, and heating, respectively. pv The predicted photovoltaic power generation; β1, β2, β3, β4 and β5 are all weighting coefficients not less than 0, and β1+β2+β3+β4+β5=1;

[0019] S5, Energy Optimization Control:

[0020] A 10-minute basic control cycle is set. Within each basic control cycle, the monitoring module acquires monitoring data and, after data preprocessing in step S3, combines the current time of day, season, and indoor area of ​​the building to obtain the multi-dimensional feature vector, which is then input into the neural network model. The target control value output by the neural network model is converted into a specific control signal and the control action is executed, thereby achieving energy optimization control of the photovoltaic integrated curtain wall. Specifically, after receiving the control signal corresponding to the rotation angle of the photovoltaic panel, the rotating drive device drives the photovoltaic panel to rotate around the rotation axis to the target angle and synchronously feeds back the actual angle value to ensure positioning accuracy. After receiving the control signal corresponding to the exhaust volume of the cavity ventilation module, the cavity ventilation module adjusts the fan speed of the top exhaust port and the opening of the air valve of the bottom air inlet to form a dynamic airflow channel with controllable exhaust volume within the cavity. After receiving the control signal corresponding to the light transmittance of the dimming glass layer, the dimming glass layer adjusts the applied voltage value to achieve real-time adjustment of the light transmittance.

[0021] S6. Model Feedback Optimization:

[0022] After executing control actions in each basic control cycle, the monitoring module continuously monitors the actual values ​​of air conditioning energy consumption, lighting energy consumption, heating energy consumption, and photovoltaic power generation. Then, the multidimensional feature vector, the target control value, and the actual value are added to the historical monitoring database as new data. The neural network model is fine-tuned weekly using incremental training with the new data, thereby improving the model's prediction accuracy and generalization ability in complex dynamic environments.

[0023] S7. Dynamic adaptation of weight coefficients based on season:

[0024] To address the differentiated impact of seasonal climate variations on building energy consumption, this study dynamically adapts the values ​​of weight coefficients β1, β2, β3, β4, and β5 in the loss function to achieve a neural network model that responds to the core optimization objectives for different seasons. Specifically, during summer, β1 is increased and β4 is decreased, meaning the penalty for air conditioning energy consumption is strengthened and the reward for photovoltaic (PV) power generation is weakened in the neural network model. This ensures that the target control value prioritizes the effect of shading and cooling while avoiding excessive pursuit of PV power generation that could lead to overheating of the PV panels. During winter daytime periods, β2 and β3 are increased, meaning the penalty for lighting and heating energy consumption is strengthened in the neural network model, ensuring that the target control value prioritizes the effect of shading and cooling while avoiding excessive pursuit of PV power generation that could lead to overheating of the PV panels. By optimizing the rotation angle of the photovoltaic panels and the transmittance of the dimming glass layer, solar radiation heat is used for indoor heating, thereby reducing heating energy consumption. During winter nights, only β3 is increased, meaning that only the penalty for heating energy consumption is strengthened in the neural network model. This allows the target control value to prioritize optimizing the exhaust volume of the cavity ventilation module, forming a sealed air insulation layer in the cavity to reduce indoor heat loss, thereby reducing heating energy consumption. During spring and autumn, β2 is increased, meaning that only the penalty for lighting energy consumption is strengthened in the neural network model. This allows the target control value to prioritize optimizing the rotation angle of the photovoltaic panels and the transmittance of the dimming glass layer, ensuring the utilization of natural light and thus reducing lighting energy consumption.

[0025] Preferably, in step S3, the raw data collected by the monitoring module is preprocessed, specifically including the following steps:

[0026] S301. Data noise reduction processing: The moving average filtering method is used to process continuous data, including solar altitude angle, solar azimuth angle, indoor temperature, indoor humidity and photovoltaic panel temperature; the median filtering method is used to process data that are susceptible to transient interference, including indoor illuminance and photovoltaic panel power generation.

[0027] S302. Abnormal Data Identification and Correction: Abnormal data is identified based on the 3σ principle, that is, when any monitoring data deviates from the mean of its data group by more than 3 times the standard deviation, it is judged as abnormal data; the identified abnormal data is removed, and the data is repaired by using the effective data at adjacent time points through linear interpolation.

[0028] S303. Data normalization processing: Map the values ​​of all monitored parameters to the [0,1] interval. The mapped parameter values ​​Y norm Satisfy the following expression:

[0029]

[0030] Where Y is the original parameter value, Y max and Y min These are the maximum and minimum values ​​of the parameter recorded in the historical database and the current work cycle, respectively.

[0031] S304. Construction of multidimensional feature vectors: Unify all normalized monitoring data to the same time reference to form multidimensional feature vectors.

[0032] Preferably, in step S7, the specific value ranges of the weighting coefficients β1, β2, β3, β4 and β5 are determined based on an orthogonal experiment using historical data from the past three years: during summer, the value range of β1 is [0.3, 0.4] and the value range of β4 is [0.08, 0.12]; during winter daytime, the value range of β2 is [0.22, 0.28] and the value range of β3 is [0.25, 0.35]; during winter nighttime, the value range of β3 is [0.35, 0.45]; and during spring and autumn, the value range of β2 is [0.27, 0.33].

[0033] Compared with existing technologies, the beneficial effects of this invention are as follows: Addressing the technical shortcomings of existing photovoltaic integrated curtain walls in dynamic angle tracking, pollution protection, multi-system collaborative control, and functional integration, this invention proposes a method for dynamic tracking and energy optimization control of solar radiation angles for photovoltaic integrated curtain walls. This method includes the installation of the photovoltaic integrated curtain wall, the deployment of monitoring modules, preprocessing of monitoring data, construction of a neural network model, energy optimization control, model feedback optimization, and dynamic adaptation of weight coefficients based on the season. Specifically, it includes the following beneficial effects:

[0034] 1) Enhance building aesthetics and reduce maintenance burden: The outer glass layer uses ultra-clear tempered glass with a low surface energy homopolymer coating, which significantly reduces the adhesion of pollutants on the building facade; the photovoltaic panels are arranged in a closed cavity between the outer glass layer and the dimming glass layer, isolating them from the influence of the external environment; at the same time, the controllable airflow in the cavity and the rotation of the photovoltaic panels themselves can effectively promote the shedding of surface dust, thereby maintaining the cleanliness of the photovoltaic panels and avoiding the degradation of their heat insulation and power generation capabilities;

[0035] 2) Optimize natural lighting and reduce lighting energy consumption: By adjusting the light transmittance of the dimming glass layer and the angle of the photovoltaic panel in real time, the amount of natural light entering the room can be accurately controlled. While effectively utilizing natural light to meet indoor lighting needs, the power generation coverage efficiency of the photovoltaic panel can be maximized.

[0036] 3) Comprehensive reduction of building energy consumption: By combining three key measures—adjusting the angle of photovoltaic panels, controlling the airflow of the cavity ventilation module, and adjusting the light transmittance of the dimming glass—a neural network model is used to optimize the energy consumption of air conditioning, lighting, and heating. This significantly reduces the overall energy consumption of the building while maximizing the benefits of photovoltaic power generation.

[0037] 4) Enhance the long-term dynamic adaptability of the system: Establish a model feedback optimization mechanism, and use newly added operational monitoring data to incrementally update and train the neural network model every week, so that the system can continuously learn to adapt to complex dynamic factors such as seasonal changes, weather fluctuations, and changes in building energy consumption patterns, and continuously optimize the control strategy. This mechanism overcomes the problem of continuous performance decline of traditional preset parameter systems in complex environments.

[0038] 5) Achieve optimal energy efficiency throughout the year: Based on the climate characteristics of different seasons, by setting differentiated optimization target weights, the system prioritizes enhancing the sunshade and cooling efficiency in summer, focuses on achieving the lighting and heat preservation target in winter, and strives to improve the utilization rate of natural light in the transition season; thereby ensuring that the system can efficiently respond to the differentiated energy consumption needs of buildings at all times throughout the year and achieve optimal energy efficiency throughout the entire cycle. Attached Figure Description

[0039] Figure 1 This is a flowchart of the solar angle dynamic tracking and energy optimization control method for photovoltaic integrated curtain walls according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the photovoltaic integrated curtain wall according to an embodiment of the present invention;

[0041] Reference numerals: 1-Outer glass layer, 2-Rotable photovoltaic panel array, 21-Louvre-type bracket, 22-Photovoltaic panel, 23-Rotation drive device, 3-Dimming glass layer, 41-Ventilation cavity, 42-Bottom air inlet, 43-Top air outlet. Detailed Implementation

[0042] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for illustration and explanation only and are not intended to limit the present invention.

[0043] This application discloses, as follows: Figure 1-2 The method for dynamic tracking and energy optimization control of solar integrated curtain wall solar radiation angle, as shown, includes the following steps:

[0044] S1. Installation of the photovoltaic integrated curtain wall: The photovoltaic integrated curtain wall, from the outside to the inside, includes an outer glass layer 1, a rotatable photovoltaic panel array 2, and a dimming glass layer 3, and is equipped with a cavity ventilation module; the outer glass layer 1 uses ultra-clear tempered glass as the substrate, and the surface is coated with a modified polypropylene coating with low surface energy to ensure that the surface of the outer glass layer 1 is not easily contaminated by various liquids and dust; the rotatable photovoltaic panel array 2 includes a louvered bracket 21, several photovoltaic panels 22, and several rotation drive devices 23; the louvered bracket 21 is composed of parallel longitudinal fixed frames and transverse support beams, the longitudinal fixed frames are provided with shaft holes for installing the photovoltaic panel rotation shafts, and both ends are fixed to the main building structure through embedded parts, the length of the transverse support beams is the same as the length of the photovoltaic panel rotation shafts. The curtain wall width is adapted and rigidly connected to the longitudinal fixed frame by bolts; the photovoltaic panel 22 adopts a monocrystalline silicon cell module, and the center positions of both ends are inserted into the shaft holes of the longitudinal fixed frame through a rotating shaft so that it can rotate freely around the axis, and an air circulation channel is reserved between adjacent photovoltaic panels 22; the rotation drive device 23 is installed on the side of the longitudinal fixed frame to drive the photovoltaic panel 22 to rotate around the axis; the dimming glass layer 3 adopts electrochromic dimming glass, and its overall size is adapted to the curtain wall frame to adjust the light transmittance and color according to the applied voltage; the cavity ventilation module is located in the cavity formed between the outer glass layer 1 and the dimming glass layer 3, including a ventilation cavity 41, a bottom air inlet 42 and a top air outlet 43.

[0045] S2. Deployment of the monitoring module: The monitoring module includes a solar tracking sensor, a temperature and humidity sensor, a illuminance sensor, a photovoltaic panel temperature sensor, a photovoltaic panel current sensor, and an energy consumption monitoring unit. The solar tracking sensor is a dual-axis digital solar angle sensor, installed at the top center of the outer side of the curtain wall, used to collect solar altitude angle and solar azimuth angle data. The temperature and humidity sensor and the illuminance sensor are arranged inside the building, used to collect indoor temperature, indoor humidity, and indoor illuminance data, respectively. The photovoltaic panel temperature sensor and the photovoltaic panel power generation sensor are both installed on the back of the photovoltaic panel 22, used to collect photovoltaic panel temperature and photovoltaic panel power generation data, respectively. The energy consumption monitoring unit uses a smart meter to measure air conditioning energy consumption, lighting energy consumption, and heating energy consumption data by area.

[0046] S3. Preprocessing of monitoring data: The raw data collected by the monitoring module is preprocessed to construct a multi-dimensional feature vector, providing good input features for the neural network model. The multi-dimensional feature vector contains 13 parameters: solar altitude angle, solar azimuth angle, indoor temperature, indoor humidity, indoor illuminance, photovoltaic panel temperature, photovoltaic panel power generation, air conditioning energy consumption, lighting energy consumption, heating energy consumption, indoor area of ​​the building, time of day, and season. In specific implementation, the preprocessing of the raw data collected by the monitoring module includes the following steps:

[0047] S301. Data Noise Reduction Processing: Continuous data, including solar altitude angle, solar azimuth angle, indoor temperature, indoor humidity, and photovoltaic panel temperature, are processed using a moving average filtering method. In practice, the sliding window size of the moving average filtering method is dynamically adjusted according to the data sampling frequency. For slowly changing data such as solar altitude angle and solar azimuth angle, a sliding window of 30 data points is used; for moderately changing data such as indoor temperature, indoor humidity, and photovoltaic panel temperature, a sliding window of 10 data points is used, thereby balancing the real-time performance and smoothness of the data. Median filtering is used to process data susceptible to transient interference, including indoor illuminance and photovoltaic panel power generation. In practice, the initial window size of the median filtering method is 5 data points, and if continuous transient interference is detected, it is automatically expanded to 15 data points to enhance robustness.

[0048] S302. Abnormal Data Identification and Correction: Abnormal data is identified based on the 3σ principle, that is, when any monitoring data deviates from the mean of its data group by more than 3 times the standard deviation, it is judged as abnormal data; the identified abnormal data is removed, and the data is repaired by using the effective data at adjacent time points through linear interpolation.

[0049] S303. Data normalization processing: Map the values ​​of all monitored parameters to the [0,1] interval. The mapped parameter values ​​Y norm Satisfy the following expression:

[0050]

[0051] Where Y is the original parameter value, Y max and Y min These are the maximum and minimum values ​​of the parameter recorded in the historical database and the current work cycle, respectively.

[0052] S304. Construction of multidimensional feature vectors: Unify all normalized monitoring data to the same time reference to form multidimensional feature vectors.

[0053] S4. Construction of the Neural Network Model: The neural network model architecture includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the multidimensional feature vector, used to map the feature vector to the hidden layer for feature extraction. The hidden layer includes four backbone feature extraction layers and one fully connected layer. The backbone feature extraction layer uses gated spatiotemporal convolutional units, integrating the spatial feature extraction capability of the convolutional layer with the temporal dependency capture capability of the gating mechanism, and capturing temporal dependencies through the ReLU activation function. The fully connected layer uses the LeakyReLU activation function to integrate the high-dimensional features output by the backbone feature extraction layer, enhancing the fitting ability to nonlinear relationships. A dropout mechanism with a probability of 0.2 is introduced in the connections between hidden layers to randomly discard some neuron connections, used to suppress overfitting during model training. The output layer contains three neurons, used to output target control values. The target control values ​​include the photovoltaic panel rotation angle, the exhaust volume of the cavity ventilation module, and the light transmittance of the dimming glass layer.

[0054] The training data for the neural network model comes from a historical monitoring database accumulated over a long period of operation. This database must contain at least one year of continuous operational data, covering all four seasons, different weather conditions, and peak or off-peak periods of building energy consumption. A weighted multi-objective loss function is used during training, as shown in the following expression:

[0055] L total =β1·E air +β2·E light +β3·E heat -β4·E pv +β5·L MSE (2)

[0056] Among them, L total Let L be the total loss function. MSE E represents the mean square error between the predicted and actual values. air E light E heat These are the predicted energy consumption for air conditioning, lighting, and heating, respectively. pv β1, β2, β3, β4 and β5 are all weighting coefficients not less than 0, and β1+β2+β3+β4+β5=1.

[0057] S5. Energy Optimization Control: A 10-minute basic control cycle is set. Within each basic control cycle, the monitoring module acquires monitoring data and, after the data preprocessing process in step S3, combines the current time of day, season, and indoor area of ​​the building to obtain the multi-dimensional feature vector, which is then input into the neural network model. The target control value output by the neural network model is converted into a specific control signal and the control action is executed, thereby achieving energy optimization control of the photovoltaic integrated curtain wall. Specifically, after receiving the control signal corresponding to the rotation angle of the photovoltaic panel, the rotation drive device 23 drives the photovoltaic panel 22 to rotate around the rotation axis to the target angle and synchronously feeds back the actual angle value to ensure positioning accuracy. After receiving the control signal corresponding to the exhaust volume of the cavity ventilation module, the cavity ventilation module adjusts the fan speed of the top exhaust port 43 and the opening of the air valve of the bottom air inlet 42, thereby forming a dynamic airflow channel with controllable exhaust volume within the cavity. After receiving the control signal corresponding to the light transmittance of the dimming glass layer 3, the applied voltage value is adjusted to achieve real-time adjustment of the light transmittance.

[0058] S6. Model Feedback Optimization: After executing control actions in each basic control cycle, the monitoring module continuously monitors the actual values ​​of air conditioning energy consumption, lighting energy consumption, heating energy consumption, and photovoltaic power generation. Then, the multidimensional feature vector, the target control value, and the actual value are added to the historical monitoring database as new data. The neural network model is fine-tuned weekly using incremental training with the new data, thereby improving the model's prediction accuracy and generalization ability in complex dynamic environments.

[0059] S7. Season-Based Dynamic Adaptation of Weighting Coefficients: Addressing the differentiated impact of seasonal climate variations on building energy consumption, this system dynamically adapts the values ​​of weighting coefficients β1, β2, β3, β4, and β5 in the loss function to enable the neural network model to respond to the core optimization objectives for different seasons. Specifically: During summer, β1 is increased and β4 is decreased, meaning the penalty for air conditioning energy consumption is strengthened and the reward for photovoltaic (PV) power generation is weakened in the neural network model, prioritizing the effect of shading and cooling while avoiding excessive pursuit of PV power generation leading to PV overheating; During winter daytime, β2 and β3 are increased, meaning the penalty for lighting and heating energy consumption is strengthened in the neural network model, prioritizing the optimization of PV panel rotation angle and light transmittance of the dimming glass layer to utilize solar radiation for indoor heating, thereby reducing heating energy consumption; During winter nighttime, only β3 is increased, meaning only the penalty for heating energy consumption is strengthened in the neural network model, prioritizing the optimization of PV panel rotation angle and light transmittance of the dimming glass layer to utilize solar radiation for indoor heating, thus reducing heating energy consumption; During winter nighttime, only β3 is increased, meaning the penalty for heating energy consumption is strengthened only in the neural network model, prioritizing the optimization of PV panel rotation angle and light transmittance of the dimming glass layer to reduce heating energy consumption. The method of reducing the exhaust volume of the cavity ventilation module forms a sealed air insulation layer within the cavity to reduce indoor heat loss, thereby reducing heating energy consumption; during spring and autumn, β2 is increased, that is, the penalty for lighting energy consumption is strengthened only in the neural network model, so that the target control value prioritizes the utilization of natural light by optimizing the rotation angle of the photovoltaic panel and the light transmittance of the dimming glass layer, thereby reducing lighting energy consumption; in specific implementation, the specific value range of the weighting coefficients β1, β2, β3, β4 and β5 is based on... Orthogonal experiments based on historical data from the past three years determined that: during summer, the range of β1 is [0.3, 0.4] and the range of β4 is [0.08, 0.12]; during winter daytime, the range of β2 is [0.22, 0.28] and the range of β3 is [0.25, 0.35]; during winter nighttime, the range of β3 is [0.35, 0.45]; and during spring and autumn, the range of β2 is [0.27, 0.33].

[0060] The above describes one or more embodiments of the present invention in a relatively specific and detailed manner, but it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for dynamic tracking and energy optimization control of solar radiation angle in photovoltaic integrated curtain walls, characterized in that, Includes the following steps: S1. Installation of the photovoltaic integrated curtain wall: The photovoltaic integrated curtain wall, from the outside to the inside, includes an outer glass layer, a rotatable photovoltaic panel array, and a dimming glass layer, and is equipped with a cavity ventilation module; the outer glass layer uses ultra-clear tempered glass as the substrate, and the surface is coated with a modified polypropylene coating with low surface energy to ensure that the surface of the outer glass layer is not easily contaminated by various liquids and dust; the rotatable photovoltaic panel array includes a louvered bracket, several photovoltaic panels, and several rotation drive devices; the louvered bracket consists of parallel longitudinal fixed frames and transverse support beams, the longitudinal fixed frames are provided with shaft holes for installing the photovoltaic panel rotation shafts, and both ends are fixed to the main building structure through pre-embedded parts, the length of the transverse support beams is... The photovoltaic panels are adapted to the width of the curtain wall and rigidly connected to the longitudinal fixing frame by bolts. The photovoltaic panels are monocrystalline silicon solar cell modules, with their center ends inserted into the shaft holes of the longitudinal fixing frame via rotating shafts, allowing them to rotate freely around the shaft. Air circulation channels are reserved between adjacent photovoltaic panels. A rotation drive device is installed on the side of the longitudinal fixing frame to drive the photovoltaic panels to rotate around the shaft. The dimming glass layer is made of electrochromic dimming glass, with its overall dimensions adapted to the curtain wall frame, used to adjust light transmittance and color according to the applied voltage. The cavity ventilation module is located in the cavity formed between the outer glass layer and the dimming glass layer, including a ventilation cavity, a bottom air inlet, and a top air outlet. S2. Deployment of the monitoring module: The monitoring module includes a solar tracking sensor, a temperature and humidity sensor, a light intensity sensor, a photovoltaic panel temperature sensor, a photovoltaic panel current sensor, and an energy consumption monitoring unit. The solar tracking sensor is a dual-axis digital solar angle sensor, installed at the top center of the outer side of the curtain wall, used to collect solar altitude angle and solar azimuth angle data. The temperature and humidity sensor and the light intensity sensor are arranged inside the building, used to collect indoor temperature, indoor humidity, and indoor light intensity data, respectively. The photovoltaic panel temperature sensor and the photovoltaic panel power generation sensor are both installed on the back of the photovoltaic panel, used to collect photovoltaic panel temperature and photovoltaic panel power generation data, respectively. The energy consumption monitoring unit uses a smart meter to measure air conditioning energy consumption, lighting energy consumption, and heating energy consumption data by area. S3. Preprocessing of monitoring data: The raw data collected by the monitoring module is preprocessed to construct a multi-dimensional feature vector, providing good input features for the neural network model; the multi-dimensional feature vector contains 13 parameters, namely: solar altitude angle, solar azimuth angle, indoor temperature, indoor humidity, indoor illuminance, photovoltaic panel temperature, photovoltaic panel power generation, air conditioning energy consumption, lighting energy consumption, heating energy consumption, indoor area of ​​the building, time of day and season; S4. Construction of the Neural Network Model: The architecture of the neural network model includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the multidimensional feature vector, used to map the feature vector to the hidden layer for feature extraction. The hidden layer includes four backbone feature extraction layers and one fully connected layer. The backbone feature extraction layers use gated spatiotemporal convolutional units and capture temporal dependencies using the ReLU activation function. The fully connected layer uses the LeakyReLU activation function to integrate the high-dimensional features output by the backbone feature extraction layers, enhancing the fitting ability to nonlinear relationships. A dropout mechanism is introduced in the connections between hidden layers to suppress overfitting during model training. The output layer contains three neurons and is used to output target control values. The target control values ​​include the photovoltaic panel rotation angle, the exhaust volume of the cavity ventilation module, and the light transmittance of the dimming glass layer. The training data for the neural network model comes from a historical monitoring database accumulated over a long period of operation. A weighted multi-objective loss function is used during training, as shown in the following expression: L total =β1·E air +β2·E light +β3·E heat -β4·E pv +β5·L MSE Among them, L total Let L be the total loss function. MSE E represents the mean square error between the predicted and actual values. air E light E heat These are the predicted energy consumption for air conditioning, lighting, and heating, respectively. pv The predicted photovoltaic power generation; β1, β2, β3, β4 and β5 are all weighting coefficients not less than 0, and β1+β2+β3+β4+β5=1; S5, Energy Optimization Control: A 10-minute basic control cycle is set. Within each basic control cycle, the monitoring module acquires monitoring data and, after data preprocessing in step S3, combines the current time of day, season, and indoor area of ​​the building to obtain the multi-dimensional feature vector, which is then input into the neural network model. The target control value output by the neural network model is converted into a specific control signal and the control action is executed. Specifically, after receiving the control signal corresponding to the rotation angle of the photovoltaic panel, the rotating drive device drives the photovoltaic panel to rotate around the rotation axis to the target angle and synchronously feeds back the actual angle value to ensure positioning accuracy. After receiving the control signal corresponding to the exhaust volume of the cavity ventilation module, the cavity ventilation module adjusts the fan speed of the top exhaust port and the opening of the air valve of the bottom air inlet to form a dynamic airflow channel with controllable exhaust volume within the cavity. After receiving the control signal corresponding to the light transmittance of the dimming glass layer, the dimming glass layer adjusts the applied voltage value to achieve real-time adjustment of the light transmittance. S6. Model Feedback Optimization: After executing control actions within each basic control cycle, the monitoring module continuously monitors the actual values ​​of air conditioning energy consumption, lighting energy consumption, heating energy consumption, and photovoltaic power generation. Subsequently, the multidimensional feature vector, the target control value, and the actual value are added as new data to the historical monitoring database. The neural network model is fine-tuned weekly using incremental training with the new data. S7. Dynamic adaptation of weight coefficients based on season: To address the differentiated impact of seasonal climate variations on building energy consumption, the neural network model responds to the core optimization objectives of different seasons by dynamically adapting the values ​​of the weight coefficients β1, β2, β3, β4, and β5 in the loss function. Specifically, this includes: increasing β1 and decreasing β4 during summer, i.e., strengthening the penalty for air conditioning energy consumption and weakening the reward for photovoltaic power generation in the neural network model; increasing β2 and β3 during winter daytime, i.e., strengthening the penalty for lighting and heating energy consumption in the neural network model; increasing only β3 during winter nighttime, i.e., strengthening the penalty for heating energy consumption only in the neural network model; and increasing β2 during spring and autumn, i.e., strengthening the penalty for lighting energy consumption only in the neural network model.

2. The method for dynamic tracking and energy optimization control of solar integrated curtain wall according to claim 1, characterized in that, In step S3, the raw data collected by the monitoring module is preprocessed, specifically including the following steps: S301. Data noise reduction processing: The moving average filtering method is used to process continuous data, including solar altitude angle, solar azimuth angle, indoor temperature, indoor humidity and photovoltaic panel temperature; the median filtering method is used to process data that are susceptible to transient interference, including indoor illuminance and photovoltaic panel power generation. S302. Abnormal Data Identification and Correction: When any monitoring data deviates from the mean of its data group by more than 3 times the standard deviation, it is judged as abnormal data; the identified abnormal data is removed, and the data is repaired by using the effective data at adjacent time points through linear interpolation. S303. Data normalization processing: Map the values ​​of all monitored parameters to the [0,1] interval. The mapped parameter values ​​Y norm Satisfy the following expression: Where Y is the original parameter value, Y max and Y min These are the maximum and minimum values ​​of the parameter recorded in the historical database and the current work cycle, respectively. S304. Construction of multidimensional feature vectors: Unify all normalized monitoring data to the same time reference to form multidimensional feature vectors.

Images