Sunlight room intelligent temperature control and illumination adjustment method suitable for high solar radiation area

Abstract

The invention relates to a sunlight room intelligent temperature control and illumination adjustment method suitable for a high solar radiation area, and the method comprises the steps: obtaining indoor illumination intensity data and temperature distribution conditions, and collecting external environment data; according to the indoor illumination intensity data, the temperature distribution condition and the external environment data, adjusting data of the sunlight room is determined; collecting the indoor illumination intensity data, the temperature distribution condition, the external environment data and the corresponding adjustment data, and constructing a data set; training a preset deep learning model by using the data set to obtain a sunshine room adjustment model; and adjusting the temperature control and illumination of the sunshine room in real time by using the sunshine room adjusting model. Intelligent balance adjustment of illumination in the sunlight room can be achieved, the spatial light environment quality is improved, it is ensured that illumination of all areas is uniform and soft, the phenomenon of local too bright or too dark is avoided, a comfortable and bright indoor environment is created for a user, and the visual comfort and the working efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of environmental regulation, in particular to a sunlight room intelligent temperature control and light adjustment method suitable for high solar radiation areas. BACKGROUND

[0002] In high solar radiation areas, the sunlight room, as a special space closely connected with the outdoor natural environment in buildings, its internal environment is significantly affected by external solar radiation. In such an environment, the management of indoor light and temperature plays a crucial role in improving living comfort and achieving building energy saving.

[0003] The existing sunlight room light and temperature control technology mainly exists in the following forms:

[0004] Traditional shading technology: Many sunlight rooms currently use fixed sunshade curtains or blinds to adjust indoor light. This type of device usually needs to be adjusted manually and cannot automatically adjust the shading angle and shading area according to real-time light intensity and indoor demand. In high solar radiation areas, the intensity and angle of sunlight change greatly with time and season, and manual adjustment often lags behind, making it difficult to accurately control indoor light, resulting in uneven or excessive or insufficient indoor light, affecting the visual comfort of indoor personnel. For example, under strong sunlight in summer, if shading is not timely, the indoor light may be too strong, causing glare; while in winter, excessive shading may result in insufficient indoor lighting, requiring additional artificial lighting, increasing energy consumption;

[0005] Simple light reflection device: Some sunlight rooms will set up some fixed light reflection plates to reflect light to darker areas indoors. However, the position and angle of these light reflection plates are mostly pre-set and cannot be dynamically adjusted according to the real-time light distribution in the room. With the change of the sun's position and the change of indoor personnel activities, the effect of the light reflection plates may be greatly reduced, and they cannot continuously and effectively optimize the indoor light uniformity. For example, at different times of the day, the incident angle of sunlight is different, and fixed light reflection plates may only have good reflection effect at certain times, while at other times they cannot accurately reflect light to the desired areas, causing uneven indoor light;

[0006] The basic temperature control system: the common sunlight room temperature control mainly relies on simple temperature regulation methods such as air conditioning and ventilation equipment. These devices usually operate according to preset temperature values, lack intelligent perception and dynamic adjustment capabilities for the actual indoor thermal environment. They cannot comprehensively consider the influence of indoor and outdoor temperature difference, solar radiation intensity, personnel activity and other factors on indoor temperature, and the adjustment effect is often not accurate enough, which can easily cause large temperature fluctuations, energy waste and other problems. For example, when the external environment temperature changes sharply, the air conditioner may need a long time to adjust the indoor temperature to the comfortable range, and in order to maintain the temperature stable, the air conditioner needs to be frequently started and stopped, which not only has high energy consumption, but also can shorten the service life of the equipment

[0007] In view of the above defects of the prior art, it is urgent to propose a sunlight room intelligent temperature control and illumination adjustment method suitable for high solar radiation areas, which realizes intelligent adjustment of sunshade devices, light reflection devices and temperature control equipment through a deep learning model, so as to solve the problems of inaccurate adjustment, inability to dynamically adapt to environmental changes and low energy utilization efficiency in the prior art. SUMMARY

[0008] The purpose of the present application is to propose a sunlight room intelligent temperature control and illumination adjustment method suitable for high solar radiation areas to solve the problems existing in the prior art.

[0009] To achieve the above purpose, the present application provides the following scheme:

[0010] A sunlight room intelligent temperature control and illumination adjustment method suitable for high solar radiation areas, comprising:

[0011] Obtain the light intensity data and temperature distribution of the indoor, and collect the external environment data; wherein the external environment data includes: external light intensity, weather condition, air temperature, wind speed;

[0012] According to the light intensity data, temperature distribution and external environment data of the indoor, determine the adjustment data of the sunlight room;

[0013] Collect the light intensity data, temperature distribution and external environment data of the indoor and the corresponding adjustment data, and construct a data set;

[0014] Train the preset deep learning model using the data set, and obtain a sunlight room adjustment model;

[0015] Use the sunlight room adjustment model to adjust the temperature control and illumination of the sunlight room in real time.

[0016] Optionally, obtaining the light intensity data and temperature distribution of the indoor comprises:

[0017] Based on the multi-point light sensor network and temperature sensor network pre-deployed in the sunlight room, light sensing data and temperature sensing data are collected;

[0018] The light sensing data and temperature sensing data are pre-processed to obtain the light intensity data and temperature distribution of the indoor.

[0019] Optionally, the pre-processing of the light sensing data and temperature sensing data comprises:

[0020] The light sensing data is smoothed by using a time series analysis algorithm to calculate the average light value of each node, and the light data is bound to the location by combining the indoor floor plan, and a light distribution map is generated by using the Kriging interpolation method.

[0021] The temperature sensing data is smoothed by using a time series analysis algorithm to calculate the average temperature value of each node, and the temperature data is bound to the location by combining the indoor floor plan, and a temperature distribution map is generated by using the Kriging interpolation method.

[0022] An edge detection method based on gradient calculation is used to perform convolution operation on the light distribution map to quantify the light-dark difference, mark the light-dark transition area, and determine the local light spot position and intensity range.

[0023] The temperature distribution map is processed by using the finite difference method to calculate the temperature gradient, capture the temperature difference change in the room, and identify the area with rapid temperature change.

[0024] Optionally, the adjustment data of the sunlight room comprises: the angle of the sunshade device, the position and angle of the light-reflecting device, and the operating parameters of the temperature control equipment.

[0025] Optionally, the angle of the sunshade device is determined by:

[0026] According to the latitude of the local area, the orientation of the sunlight room, and the variation of the solar altitude angle in different seasons, the basic angle range of the sunshade device is set.

[0027] When the external light intensity exceeds the preset threshold, the sunshade angle is gradually increased in the basic angle range of the sunshade device according to the preset proportional relationship, and at the same time, the sunshade angle is further increased when the indoor temperature exceeds the preset upper limit of the comfortable temperature in combination with the indoor light intensity data.

[0028] Optionally, the position and angle of the light-reflecting device are determined by:

[0029] According to the indoor light intensity data, the position of the indoor light uneven area and the light intensity difference are analyzed; for the area with weak light, the required supplementary light intensity is calculated, and the position and angle of the light-reflecting device are adjusted by using the principles of light refraction and reflection.

[0030] Optionally, determining the operating parameter of the temperature control device comprises:

[0031] When the indoor temperature deviates from the comfortable temperature range, the operating parameter of the temperature control device is adjusted according to the size and change trend of the temperature deviation, and in addition, in combination with the personnel activity detection sensor, when an increase in the number of people in the sunlight interval is detected, the air conditioning refrigerating capacity or the ventilation capacity is increased in advance.

[0032] Optionally, training the preset deep learning model using the data set comprises:

[0033] The data set is cleaned, normalized, and has abnormal values and noise data removed, and the time stamps are unified;

[0034] The data with unified time stamps is input into the preset deep learning model, historical illumination, temperature distribution data and environmental conditions are used as input features, and corresponding device adjustment parameters are used as output labels, the model is trained to learn the complex mapping relationship between the input and the output;

[0035] In the model training process, the error between the predicted value and the actual value is minimized by adjusting the hyperparameters and optimization algorithms of the model, and the prediction accuracy and generalization ability of the model are gradually improved.

[0036] Optionally, the preset deep learning model comprises a graph attention network, a time convolution network, and a residual fully connected network; and a differentiable physical layer is added after the residual fully connected network to force the output to satisfy the energy conservation law by using the glass transmission-absorption-re-radiation equation as a soft constraint;

[0037] Inputting the data with unified time stamps into the preset deep learning model comprises:

[0038] Taking the sensor as a graph node, the illumination intensity data, temperature distribution conditions and external environment data of each node in the window are spliced to form a node feature matrix;

[0039] Based on the node feature matrix, the adjacency weight is calculated according to the Euclidean distance between nodes, the glass transmission attenuation coefficient and the air thermal diffusion coefficient, and a dynamic adjacency matrix is generated, i.e. a dynamically updated undirected graph is formed;

[0040] The graph attention network is used to spatially encode the undirected graph to generate a spatial feature representation of each sensor node data at a time step;

[0041] The time convolution network is used to time-encode the spatial feature representation to obtain a spatio-temporal fusion feature vector that fuses spatial and temporal information;

[0042] The spatio-temporal fusion feature vector is globally averaged pooled to obtain a graph-level feature vector;

[0043] The image-level feature vector is passed through the residual fully connected network to form a high-dimensional embedding;

[0044] Based on the differentiable physical layer, the high-dimensional embedding is used to decode the glass transmittance, absorptance, and re-radiation coefficient, calculate the theoretical energy balance according to the steady-state radiation conduction equation, and calculate the difference between the predicted energy and the theoretical energy as the energy conservation residual, based on which the image-level feature vector is corrected;

[0045] The corrected image-level feature vector is then passed through a fully connected network to directly map to three continuous values: the angle of the shading device, the angle of the light-reflecting device, and the operating parameters of the temperature control equipment.

[0046] The beneficial effects of the present application are:

[0047] The intelligent balanced adjustment of sunlight in the sunlight room is realized, the quality of the space light environment is improved, the uniform and soft illumination of each region is ensured, and the phenomenon of local over-brightness or over-darkness is avoided, thus creating a comfortable and bright indoor environment for the user and improving visual comfort and work efficiency.

[0048] The indoor temperature is accurately controlled, the shading, ventilation, refrigeration or heating equipment is automatically adjusted according to the external environmental changes and the actual temperature conditions in the room, the indoor temperature is maintained within the range of human comfort, the discomfort caused by excessively high or low temperature is effectively avoided, the comfort and satisfaction of the user are improved, and the quality of the living and working environment is improved.

[0049] The energy utilization efficiency is improved, and unnecessary energy consumption is reduced through intelligent adjustment of light and temperature.

[0050] Through the introduction of the deep learning model, automatic optimization and intelligent decision-making of the device adjustment parameters are realized. The deep learning model can continuously learn and adapt to new environmental conditions, usage habits and equipment characteristics, automatically adjust the parameter adjustment strategy, improve the intelligent level and adaptive ability of the system, and further improve the precision and effect of light and temperature control adjustment, providing more personalized and efficient environmental adjustment services for the sunlight room. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0052] Figure 1 It is a flowchart of a sunlight room intelligent temperature control and light adjustment method suitable for high solar radiation areas. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be apparently and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work belong to the protection scope of the present application.

[0054] In order to make the above objectives, characteristics and advantages of the present application more apparent and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0055] As shown in the figure, the embodiment proposes a sunlight room intelligent temperature control and light adjustment method suitable for high solar radiation areas, including: Figure 1 Obtaining the light intensity data and temperature distribution of the indoor, and collecting the external environment data; wherein, the external environment data includes: external light intensity, weather condition, air temperature, wind speed;

[0056] According to the light intensity data, temperature distribution of the indoor and external environment data, determining the adjustment data of the sunlight room;

[0057] Collecting the light intensity data, temperature distribution of the indoor and external environment data and corresponding adjustment data, and constructing a data set;

[0058] Training the preset deep learning model by using the data set, and obtaining the sunlight room adjustment model;

[0059] Adjusting the temperature control and light of the sunlight room in real time by using the sunlight room adjustment model.

[0060] Further, obtaining the light intensity data and temperature distribution of the indoor includes:

[0061] Based on the pre-deployed multi-point light sensor network and temperature sensor network in the sunlight room, collecting light sensing data and temperature sensing data;

[0062] Pretreating the light sensing data and temperature sensing data, and obtaining the light intensity data and temperature distribution of the indoor.

[0063] The pretreatment of the light sensing data and temperature sensing data includes:

[0064] Using a time series analysis algorithm to smooth the light sensing data, calculating the average light value of each node, binding the light data with the position by combining the indoor plan, and generating a light distribution map by Kriging interpolation method;

[0065] ​

[0066] Smooth the temperature sensor data using time series analysis algorithm, calculate the average temperature value of each node, bind the temperature data with the location combined with the indoor floor plan, generate the temperature distribution map through Kriging interpolation method;

[0067] Apply edge detection method based on gradient calculation to the light distribution map for convolution operation, quantify the light-dark difference, mark the light-dark transition area, determine the local light spot position and intensity range;

[0068] Use finite difference method to process the temperature distribution map, calculate the temperature gradient, capture the indoor temperature difference, identify the area with sharp temperature change.

[0069] Specifically, in this embodiment, the multi-point data acquisition network is constructed: high-precision multi-point light sensor network and temperature sensor network are deployed inside the sunlight room, the light sensor measurement range is 0-2000 lux, the accuracy is ±5 lux, the temperature sensor measurement range is -20℃-60℃, the accuracy is ±0.5℃, one sensor node is set every 4 square meters, the data acquisition frequency is 5 seconds, the data is transmitted to the central server in real time through the wireless ZigBee protocol, forming the initial data stream of light and temperature, providing comprehensive and accurate basic data for subsequent environmental state analysis and adjustment.

[0070] Light distribution analysis: smooth the light sensor data using time series analysis algorithm, use sliding window average method, window size is 10 data points, calculate the average light value of each node. Combined with the indoor floor plan, bind the light data with the location, generate the light distribution map through Kriging interpolation method, resolution is 0.5m x 0.5m, color gradient from dark blue (0 lux) to bright yellow (2000 lux) represents the light intensity distribution, to determine the key area of uneven light distribution.

[0071] Temperature distribution analysis: use similar method as light analysis to process the temperature sensor data, generate the temperature distribution map, resolution is 0.5m x 0.5m, color gradient from dark blue (-20℃) to bright red (60℃) represents the temperature distribution, identify the temperature abnormal area, analyze the influence of indoor and outdoor temperature difference on heat transfer.

[0072] Light-dark difference and temperature difference capture algorithm: use edge detection method based on gradient calculation (such as Sobel operator) to perform convolution operation on the light distribution map, quantify the light-dark difference, mark the light-dark transition area, determine the local light spot position and intensity range. At the same time, use finite difference method to process the temperature distribution map, calculate the temperature gradient, capture the indoor temperature difference, identify the area with sharp temperature change, comprehensively analyze the relationship between light distribution and temperature distribution and their influence on indoor environment.

[0073] External environment monitoring: Real-time collection of environmental data such as external light intensity, weather conditions, air temperature, wind speed, etc. Combined with the building orientation, glass light transmittance and other characteristics of the sunlight room, through the pre-established light penetration model and heat transfer model, the interference degree of external environmental changes on indoor light balance and heat balance is calculated. For example, according to the external light intensity, glass light transmittance and orientation coefficient, the influence value of the indoor external light is calculated, and according to the indoor and outdoor air temperature, wind speed and thermal performance of the building envelope structure, the heat transfer rate is calculated to evaluate the interference of the external environment on the indoor temperature;

[0074] When the interference degree exceeds the preset threshold (light interference threshold, temperature interference threshold is set respectively), the corresponding adjustment mechanism is triggered to provide basis for the adjustment of the shading device, the light-reflecting device and the temperature control equipment.

[0075] Further, the adjustment data of the sunlight room includes: shading device angle, light-reflecting device position and angle, and temperature control equipment operating parameters.

[0076] Further, determining the shading device angle includes:

[0077] According to the local latitude, the orientation of the sunlight room and the variation law of the solar altitude angle in different seasons, the basic angle range of the shading device is set;

[0078] When the external light intensity exceeds the preset threshold, the shading angle is gradually increased in the basic angle range of the shading device according to the preset proportional relationship. At the same time, combined with the indoor light intensity data, when the indoor temperature exceeds the preset upper limit of the comfortable temperature, the shading device angle is further increased.

[0079] In this embodiment, according to the local latitude, the orientation of the sunlight room and the variation law of the solar altitude angle in different seasons, the basic angle range of the shading device is set. For example, in the area north of the Tropic of Cancer, the solar altitude angle is lower in winter, and the basic angle can be set to 0°-30°, and the solar altitude angle is higher in summer, and the basic angle can be set to 60°-90°, to preliminarily adjust the shading effect of the shading device. Real-time monitoring of external light intensity, when the external light intensity exceeds the preset threshold (such as 1000 lux), the shading angle is gradually increased according to the preset proportional relationship (such as shading angle = basic angle + (external light intensity - threshold) x 0.01° / lux) to reduce the strong light entering the room. At the same time, combined with the indoor temperature sensor data, when the indoor temperature exceeds the upper limit of the comfortable temperature (such as 28℃), the shading angle is further increased by 5°-10°, which gives priority to reducing the heat entering to ensure that the indoor temperature is within the comfortable range.

[0080] Further, determining the light-reflecting device position and angle includes:

[0081] According to the light intensity data in the room, the position of the unevenly illuminated area and the difference in light intensity are analyzed; for the area with weak light, the required supplementary light intensity is calculated, and the position and angle of the light-reflecting device are adjusted using the principles of light refraction and reflection.

[0082] In this embodiment, according to the layout and functional zoning of the sunlight room, the approximate installation position of the light-reflecting device is manually determined in the initial stage, such as setting the light-reflecting device near the window to reflect light to the darker central area in the room. Through optical simulation software, combined with the architectural structure and window position of the sunlight room, the light reflection path is simulated to preliminarily determine the installation angle of the light-reflecting device, so that the light can cover the main activity area in the room. During the normal operation of the sunlight room, according to the light distribution data collected by the light sensor network in real time, the position of the unevenly illuminated area and the difference in light intensity are analyzed. For the area with weak light, the required supplementary light intensity is calculated, and the position and angle of the light-reflecting device are adjusted using the principles of light refraction and reflection, so that the light can accurately project to these low-light areas. For example, if the light intensity of a certain area is less than 70% of the overall average light intensity, the light-reflecting device is adjusted 5°-10° in the direction of the area, and the position of the light-reflecting device is appropriately moved, so that the light can better cover the area and improve the uniformity of light. At the same time, considering the ergonomics requirements, the light after reflection should not directly shine into the eyes or cause glare.

[0083] Further, determining the operating parameters of the temperature control equipment includes:

[0084] When the indoor temperature deviates from the comfortable temperature range, the operating parameters of the temperature control equipment are adjusted according to the size and change trend of the temperature deviation. In addition, combined with the personnel activity detection sensor, when an increase in the number of people in the sunlight room is detected, the air conditioning cooling capacity or ventilation volume is increased in advance.

[0085] In this embodiment, according to the local climate conditions and the habits of the user, the comfortable temperature range in the sunlight room is set, such as 24-28°C in summer and 18-22°C in winter. In the initial stage, the temperature control equipment such as air conditioner, ventilation equipment, etc. operates according to the preset schedule and temperature threshold. For example, in summer, when the indoor temperature exceeds 28°C during the day, the air conditioner starts the cooling mode, the temperature is set to 26°C, and the ventilation equipment runs at low speed to assist heat dissipation; in winter, when the indoor temperature is lower than 18°C at night, the auxiliary heating equipment is started, the temperature is set to 20°C, and the ventilation equipment reduces the air exchange frequency to maintain the indoor temperature. Real-time monitoring of indoor temperature and humidity changes, when the indoor temperature deviates from the comfortable temperature range, according to the size of the temperature deviation and the trend of change, adjust the operating parameters of the temperature control equipment. If the indoor temperature rises sharply in a short time (such as more than 3°C per hour), increase the air conditioning cooling capacity by 20-30%, and at the same time increase the ventilation frequency; if the temperature slowly decreases (such as 1-2°C per hour), appropriately reduce the air conditioning cooling capacity or increase the power of the auxiliary heating equipment. In addition, combined with the personnel activity detection sensor, when the number of people in the sunlight room is detected to increase, the air conditioning cooling capacity or ventilation capacity is increased in advance to cope with the increase in heat caused by personnel activity, and to ensure the stability and comfort of the indoor temperature.

[0086] Further, training the preset deep learning model with the data set comprises:

[0087] cleaning and normalizing the data set, removing outliers and noise data;

[0088] input the processed data into the preset deep learning model, use historical light, temperature distribution data and environmental conditions as input features, and corresponding device adjustment parameters as output labels, train the model to learn the complex mapping relationship between input and output;

[0089] In the model training process, by adjusting the hyperparameters and optimization algorithms of the model, minimizing the error between the predicted value and the actual value, gradually improving the prediction accuracy and generalization ability of the model.

[0090] The preset deep learning model includes: graph attention network, time convolution network, residual fully connected network; and a differentiable physical layer is added after the residual fully connected network to force the output to satisfy the energy conservation with the glass transmission-absorption-re-radiation equation as a soft constraint.

[0091] inputting the data with unified timestamp into the preset deep learning model comprises:

[0092] using sensors as graph nodes, concatenating the light intensity data, temperature distribution and external environment data of each node in the window to form a node feature matrix;

[0093] Based on the node feature matrix, an adjacency weight is calculated according to an Euclidean distance between nodes, a glass transmittance attenuation coefficient and an air thermal diffusion coefficient, a dynamic adjacency matrix is generated, and a dynamically updated undirected graph is formed;

[0094] The graph attention network is used for spatial coding of the undirected graph, and a spatial feature representation of each sensor node data at a time step is generated.

[0095] The time convolution network is used for time coding of the spatial feature representation, and a spatio-temporal fusion feature vector that fuses spatial and temporal information is obtained.

[0096] Global average pooling is performed on the spatio-temporal fusion feature vector, and a graph-level feature vector is obtained.

[0097] The graph-level feature vector passes through the residual fully connected network to form a high-dimensional embedding.

[0098] Based on the differentiable physical layer, the high-dimensional embedding is used to decode the glass transmittance, absorption and re-radiation coefficient, calculate the theoretical energy budget according to the steady-state radiation conduction equation, and calculate the difference between the predicted energy and the theoretical energy as the energy conservation residual, and correct the graph-level feature vector based on the energy conservation residual.

[0099] The corrected graph-level feature vector passes through a fully connected network to be directly mapped to three continuous values: the angle of the shading device, the angle of the light-reflecting device and the operating parameters of the temperature control equipment.

[0100] Specifically, in this embodiment, during the early stage of the sunlight interval operation, light sensor data, temperature sensor data, and corresponding data such as the angle of the shading device, the position and angle of the light-reflecting device, and the operating parameters of the temperature control equipment are continuously collected to establish a detailed data set containing time stamps, environmental conditions, device states and indoor environmental indicators. The collected data is cleaned, normalized, and abnormal values and noise data are removed, and the data is divided into a training set, a validation set and a test set

[0101] The deep learning model preset in this embodiment adopts a graph attention network GAT (Graph Attention Network), a time convolution network TCN (Temporal Convolutional Network) and a residual fully connected network (Residual MLP), and a differentiable physical layer (Radiative-Conductive Physics Layer) is added after the residual fully connected network to force the output to satisfy the energy conservation with the glass transmittance-absorption-re-radiation equation as a soft constraint.

[0102] The light intensity data and temperature distribution are decomposed by three layers of wavelet packet, and then reconstructed to time domain to obtain the smoothed light channel and temperature channel. Then the light and temperature channels of the same node at the same time are connected to form a light-temperature coupling vector. The external environment vector is directly spliced behind the light-temperature coupling vector to form the original feature of the node.

[0103] Each sensor is taken as a graph node. The strength of the edge between nodes is determined by the product of three parts: the square decay factor of the straight-line distance between sensors, the current glass transmittance attenuation coefficient, and the current air thermal diffusion coefficient. The edge strength is recalculated every sixty seconds to obtain new adjacency relationships, i.e., a dynamically updated undirected graph.

[0104] The graph attention network GAT is used to perform a linear transformation of the node features at each time step in the undirected graph with shared weights, mapping the original dimension to an intermediate dimension. A learnable attention mechanism is used to calculate the attention coefficient between any two adjacent nodes, and the neighbor node features are weighted and summed according to the attention coefficient to update the current node features. Multi-head attention is used to perform multiple parallel operations, and the results are spliced in the feature dimension to obtain a node-level spatial feature sequence.

[0105] The point-level spatial feature sequence is input into the time convolution network TCN in chronological order. The dilation rate of the convolution kernel is no longer fixed, but is dynamically adjusted according to the solar elevation angle: the dilation rate increases when the solar elevation angle changes rapidly, and decreases when the solar elevation angle changes slowly. After convolution, weight normalization, ReLU activation, and Dropout are performed, and two layers are stacked with residual connections, finally obtaining the spatio-temporal fusion features of each node.

[0106] The spatio-temporal fusion features of all nodes are averaged at the same time step to obtain a graph-level feature vector. The graph-level feature vector is input into the differentiable radiation conduction physical layer: the glass transmittance, absorption rate, and re-radiation coefficient are decoded from the vector; the theoretical energy balance is calculated according to the steady-state radiation conduction equation; the difference between the predicted energy and the theoretical energy is calculated as the energy conservation residual; the residual affects the graph-level feature vector in the opposite direction, so that it meets the energy conservation constraint while maintaining data-driven accuracy.

[0107] The corrected graph-level feature vector is then input into a fully connected network to directly map to three continuous values: the angle of the shading device, the angle of the reflective device, and the operating parameters of the temperature control equipment.

[0108] During training, the light temperature and environmental features are taken as input, and the actual recorded shading angle, reflective angle, and temperature control amount are taken as labels. The total loss is the sum of two parts:

[0109] Data loss: the mean square error between the three output values and the labels;

[0110] Physical loss: square of the energy conservation residual;

[0111] Using Adam optimizer, the learning rate, weight decay coefficient, and Dropout rate are determined by grid search. The sum of the two losses is monitored on the validation set, and the hyperparameters corresponding to the minimum value are selected. Finally, a highly robust sunlight adjustment model is obtained.

[0112] In the preset deep learning model, the original sensor data and the external environment vector are parallelly spliced into each node at the input layer to form a unified node feature, and then jointly enter the first layer network; the first layer is a graph attention network GAT, which outputs a node-level spatial feature; this feature sequence directly enters the second layer time convolution network TCN to complete the time dimension modeling; the connection between the two levels is serial: the outlet of GAT is the inlet of TCN, without bypass branch; the node-level spatio-temporal fusion feature output by TCN is globally averaged pooled, compressing the "node dimension" to "graph dimension" to obtain a single graph-level feature vector; this step realizes the structural conversion from node representation to overall representation. The graph-level feature vector passes through two layers of residual jump full connection network in turn to form a high-dimensional embedding, and the residual connection is only within the layer. The output of the residual full connection network is immediately sent to the differentiable physics layer, which is structurally regarded as an additional "soft constraint module". Its input comes from the residual full connection, and its output is the same dimension as the residual full connection, and it transmits gradients to all previous modules during backpropagation, realizing end-to-end constraint. The graph-level embedding corrected by the physics layer is finally connected to a single-layer full connection output network to produce three continuous values of sun-shading angle, light-reflecting angle, and temperature control amount.

[0113] Using the sunlight adjustment model, real-time adjustment of the sunlight temperature control and illumination includes:

[0114] On the basis of the previous adjustment strategy, combined with the sun-shading device angle predicted by the deep learning model, the sun-shading device is dynamically optimized. The model predicts the optimal sun-shading angle according to the current light, temperature distribution, and external environmental conditions. After receiving the angle instruction, the controller of the sun-shading device accurately adjusts the angle of the sun-shading device through the motor drive device, ensuring that the sun-shading effect matches the indoor light and temperature requirements. At the same time, the system monitors the running state of the sun-shading device and the changes in the indoor environment in real time, and feeds back the actual sun-shading effect to the deep learning model to further optimize the prediction accuracy of the model.

[0115] According to the previously determined position and angle adjustment method of the light-reflecting device and the optimization suggestions predicted by the deep learning model, the position and angle of the light-reflecting device are automatically adjusted. The model takes into account factors such as indoor light distribution, personnel activity area, and external light conditions to calculate the optimal position and angle of the light-reflecting device, driving the actuator of the light-reflecting device to make precise adjustments, so that the light can be evenly distributed throughout the sunlight room, improving the comfort and utilization efficiency of the light. During the adjustment process, the light sensor monitors the light changes in real time to ensure that the adjustment effect of the light-reflecting device meets the expectations, and the light feedback data is used for continuous optimization of the model.

[0116] According to the previously set temperature control strategy and the operating parameters of the temperature control equipment predicted by the deep learning model, the operation of the air conditioner, ventilation equipment and other temperature control equipment is intelligently adjusted. The model predicts the optimal operating parameters of the temperature control equipment, such as the cooling capacity, heating capacity, and wind speed of the air conditioner, and the ventilation frequency of the ventilation equipment, based on factors such as indoor temperature, humidity, number of personnel, and external weather conditions. The control system of the temperature control equipment receives these parameter instructions and automatically adjusts the operation of the equipment to achieve precise control of the indoor temperature. At the same time, the system monitors the indoor temperature changes in real time through temperature sensors, and makes fine adjustments to the temperature control equipment according to the deviation between the actual temperature and the target temperature, and inputs the temperature feedback information into the deep learning model, so that the model can continuously learn and adapt to new environmental conditions and user needs, improving the intelligent level and comfort of the temperature control system.

[0117] In this embodiment, a light and temperature joint optimization algorithm and feedback control are also used.

[0118] The light uniformity optimization algorithm and the temperature field optimization algorithm are used to evaluate the indoor light distribution and temperature distribution, respectively, and calculate the deviation values of the light distribution from the preset uniformity standard and the temperature distribution from the preset comfort standard. If the deviation value exceeds the preset range, the light distribution and temperature distribution characteristics of the deviation area are analyzed to determine the area range and device parameter configuration that need to be adjusted. Through the pre-established mapping relationship table, the specific values of the parameter fine-tuning are calculated, and the parameters of the shading device, light-reflecting device, and temperature control equipment are automatically updated and optimized.

[0119] Under the participation of the deep learning model, the feedback control module feeds back the light and temperature data after each adjustment and the device operating state to the model, which continuously learns and accumulates experience, automatically adjusts the internal parameters and weights, and optimizes the prediction accuracy of the device adjustment parameters. The entire adjustment process forms a closed-loop control logic, which monitors the adjusted light and temperature state in real time through sensors, feeds back the data to the central control system, and continuously cycles the deviation analysis, parameter fine-tuning, and adjustment operations until the indoor light distribution meets the uniformity standard and the temperature distribution meets the comfort requirements.

[0120] The above described embodiments are only to illustrate the preferred modes of the present application, and are not intended to limit the scope of the present application. Any modification and improvement made by those skilled in the art to the technical solutions of the present application without departing from the design spirit of the present application shall fall within the protection scope of the present application as defined by the claims.

Claims

1. A sunlight interval intelligent temperature control and light adjustment method suitable for high solar radiation areas, characterized in that, The method comprises the following steps: Obtain indoor light intensity data and temperature distribution conditions, and collect external environment data; wherein, the external environment data includes: external light intensity, weather conditions, air temperature, wind speed; Determine the adjustment data of the sunlight room according to the indoor light intensity data, temperature distribution conditions and external environment data; Collect the indoor light intensity data, temperature distribution conditions and external environment data and the corresponding adjustment data to construct a data set; Train a preset deep learning model using the data set to obtain a sunlight room adjustment model; Use the sunlight room adjustment model to adjust the temperature control and light of the sunlight room in real time.

2. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 1, characterized in that, Obtaining indoor light intensity data and temperature distribution conditions comprises: Based on the pre-deployed multi-point light sensor network and temperature sensor network in the sunlight room, collect light sensing data and temperature sensing data; Preprocess the light sensing data and temperature sensing data to obtain the indoor light intensity data and temperature distribution conditions.

3. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 2, characterized in that, The preprocessing of the light sensing data and temperature sensing data comprises: Smooth the light sensing data using a time series analysis algorithm, calculate the average light value of each node, bind the light data with the location by combining the indoor floor plan, and generate a light distribution map by Kriging interpolation method; Smooth the temperature sensing data using a time series analysis algorithm, calculate the average temperature value of each node, bind the temperature data with the location by combining the indoor floor plan, and generate a temperature distribution map by Kriging interpolation method; Use the edge detection method based on gradient calculation to perform convolution operation on the light distribution map, quantify the light and dark difference, mark the light and dark transition area, and determine the local light spot position and intensity range; Use the finite difference method to process the temperature distribution map, calculate the temperature gradient, capture the temperature difference change in the room, and identify the area with sharp temperature change.

4. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 1, characterized in that, The adjustment data of the sunlight room includes: the angle of the sunshade device, the position and angle of the light-reflecting device, and the operating parameters of the temperature control equipment.

5. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 4, characterized in that, Determining the angle of the sunshade device comprises: According to the local latitude, the orientation of the sunlight room and the variation of the solar altitude angle in different seasons, set the basic angle range of the sunshade device; When the external light intensity exceeds the preset threshold, gradually increase the sunshade angle in the basic angle range of the sunshade device according to the preset proportional relationship, and at the same time, combined with the indoor light intensity data, further increase the sunshade angle when the indoor temperature exceeds the preset upper limit of comfortable temperature.

6. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 4, characterized in that, Determining the position and angle of the light-reflecting device comprises: According to the indoor light intensity data, analyze the position and light intensity difference of the indoor uneven light area; for the area with weak light, calculate the required supplementary light intensity, and adjust the position and angle of the light-reflecting device by using the principles of light refraction and reflection.

7. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 4, characterized in that, Determining the operating parameters of the temperature control equipment comprises: When the indoor temperature deviates from the comfortable temperature range, adjust the operating parameters of the temperature control equipment according to the size and change trend of the temperature deviation, in addition, combined with the personnel activity detection sensor, when the number of people in the sunlight room is detected to increase, increase the air conditioning refrigerating capacity or ventilation capacity in advance.

8. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 1, characterized in that, Training a preset deep learning model using the data set comprises: The data set is cleaned, normalized, and unified timestamp, and abnormal values and noise data are removed; The data with unified timestamp is input into a preset deep learning model, historical light, temperature distribution data and environmental conditions are taken as input features, and corresponding device adjustment parameters are taken as output labels, the model is trained to learn the complex mapping relationship between input and output; During the model training process, the hyperparameters of the model and the optimization algorithm are adjusted to minimize the error between the predicted value and the actual value, gradually improving the prediction accuracy and generalization ability of the model.

9. The intelligent temperature control and light adjustment method for sunlight interval in high solar radiation areas according to claim 8, characterized in that, The preset deep learning model includes a graph attention network, a time convolution network, and a residual fully connected network; and a differentiable physical layer is added after the residual fully connected network to take the glass transmission-absorption-re-radiation equation as a soft constraint to force the output to satisfy the energy conservation; The data with unified timestamp is input into a preset deep learning model, which includes: Taking the sensor as a graph node, the light intensity data, temperature distribution and external environment data of each node in the window are spliced to form a node feature matrix; Based on the node feature matrix, the adjacency weight is calculated according to the Euclidean distance between nodes, the glass transmission attenuation coefficient and the air thermal diffusion coefficient to generate a dynamic adjacency matrix, i.e. a dynamically updated undirected graph; The graph attention network is used to spatially encode the undirected graph to generate a spatial feature representation of each sensor node data at a time step; The time convolution network is used to time-encode the spatial feature representation to obtain a spatio-temporal fusion feature vector that fuses spatial and temporal information; The spatio-temporal fusion feature vector is globally averaged pooled to obtain a graph-level feature vector; The graph-level feature vector is passed through the residual fully connected network to form a high-dimensional embedding; Based on the differentiable physical layer, the high-dimensional embedding is decoded to obtain the glass transmission rate, absorption rate and re-radiation coefficient, the theoretical energy balance is calculated according to the steady-state radiation conduction equation, and the difference between the predicted energy and the theoretical energy is calculated as the energy conservation residual. The graph-level feature vector is corrected based on the energy conservation residual; The corrected graph-level feature vector is then passed through a fully connected network to directly map to three continuous values: sunshade device angle, light-reflecting device angle, and temperature control equipment operating parameters.

Images