Intelligent curtain wall energy-saving control method and system based on deep learning

Abstract

The invention relates to an intelligent curtain wall energy-saving control method and system based on deep learning. The method comprises the following steps: acquiring temperature data, illumination data and energy consumption data of each hierarchical region of a curtain wall system to obtain a sensor data matrix; a curtain wall energy consumption residual value is calculated, time sequence segmentation and K-means clustering are carried out, and an energy consumption feature vector and an energy consumption mode feature database are obtained; setting a state space, an action space and a reward function, and training the initial deep reinforcement learning model; self-adaptive droop coefficient adjustment is carried out on the curtain wall control area, the shading coefficient, the opening degree of a ventilation opening and the temperature of a hollow layer are adjusted, and energy-saving control parameters of all the areas are obtained; and executing multi-objective optimization control, and carrying out online updating on the initial deep reinforcement learning model to obtain a target deep reinforcement learning model. According to the invention, a regional and hierarchical control strategy is adopted, the influence of the building orientation and the floor height is fully considered, and refined control management is realized.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to an intelligent curtain wall energy-saving control method and system based on deep learning. Background Art

[0002] As an important part of the building's exterior envelope, the energy consumption control of the curtain wall system directly affects the overall energy efficiency of the building. Traditional curtain wall control methods mainly rely on simple threshold control or fixed strategy control, which cannot be dynamically adjusted according to environmental changes and user needs, resulting in high building energy consumption and difficulty in ensuring the comfort of the indoor environment.

[0003] Although the existing intelligent curtain wall control system has introduced some intelligent algorithms, there are still many problems in practical applications: the control system is difficult to effectively process multi-source heterogeneous data such as temperature, light and energy consumption in the curtain wall system, and cannot accurately capture the system operation characteristics; secondly, the traditional control strategy cannot balance the multi-objective optimization problem of energy saving effect and indoor comfort, and often loses one while focusing on the other; thirdly, the control system lacks adaptive learning capabilities and is difficult to cope with dynamic changes during the use of the building. Summary of the invention

[0004] The main purpose of the present invention is to provide an intelligent curtain wall energy-saving control method and system based on deep learning. The present invention adopts a regional and hierarchical control strategy, fully considers the influence of building orientation and floor height, and realizes refined control management.

[0005] To achieve the above object, the present invention provides an intelligent curtain wall energy-saving control method based on deep learning, comprising the following steps: Collect temperature data, lighting data and energy consumption data of each level area of ​​the curtain wall system to obtain the sensor data matrix; Calculating the curtain wall energy consumption residual value according to the sensor data matrix, and performing time series segmentation and K-means clustering on the curtain wall energy consumption residual value to obtain an energy consumption feature vector and an energy consumption pattern feature database; Based on the energy consumption feature vector and the energy consumption pattern feature database, a state space, an action space and a reward function are set to train an initial deep reinforcement learning model; Based on the initial deep reinforcement learning model, adaptive droop coefficient adjustment is performed on the curtain wall control area, and the shading coefficient, vent opening and hollow layer temperature are adjusted to obtain energy-saving control parameters for each area; Multi-objective optimization control is performed according to the energy-saving control parameters of each area, and the initial deep reinforcement learning model is updated online to obtain a target deep reinforcement learning model.

[0006] The present invention also provides an intelligent curtain wall energy-saving control system based on deep learning, comprising: The acquisition module is used to collect temperature data, light data and energy consumption data of each level area of ​​the curtain wall system to obtain a sensor data matrix; A calculation module, used for calculating the curtain wall energy consumption residual value according to the sensor data matrix, and performing time series segmentation and K-means clustering on the curtain wall energy consumption residual value to obtain an energy consumption feature vector and an energy consumption pattern feature database; A training module, used to set a state space, an action space and a reward function based on the energy consumption feature vector and the energy consumption pattern feature database, and train an initial deep reinforcement learning model; An adjustment module is used to perform adaptive droop coefficient adjustment on the curtain wall control area based on the initial deep reinforcement learning model, and adjust the shading coefficient, vent opening and hollow layer temperature to obtain energy-saving control parameters for each area; An updating module is used to perform multi-objective optimization control according to the energy-saving control parameters of each area, and to update the initial deep reinforcement learning model online to obtain a target deep reinforcement learning model.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention effectively solves the processing problem of multi-source heterogeneous data such as temperature, light and energy consumption by establishing a multi-level data acquisition network and a standardized processing flow, thereby improving data quality and availability. A control strategy based on deep reinforcement learning is adopted, and there is no need to establish an accurate mathematical model. Through online learning and experience accumulation, intelligent control of the curtain wall system is realized. The introduction of an adaptive droop coefficient adjustment mechanism enables the control system to dynamically adjust the control parameters according to the characteristics of different regions, thereby improving the accuracy and flexibility of the control. Through a multi-objective optimization algorithm, balanced control of energy-saving effects and indoor comfort is achieved, which not only ensures building energy efficiency, but also meets user experience requirements. The online update mechanism based on the incremental learning method enables the control system to have the ability of continuous learning and optimization, and can continuously adapt to dynamic changes during the use of the building. A regional and hierarchical control strategy is adopted, which fully considers the influence of building orientation and floor height, and realizes more refined control management. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the steps of an intelligent curtain wall energy-saving control method based on deep learning in one embodiment of the present invention; Figure 2 It is a structural block diagram of an intelligent curtain wall energy-saving control system based on deep learning in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 , this embodiment provides an intelligent curtain wall energy-saving control method based on deep learning, comprising the following steps: S1, collect temperature data, light data and energy consumption data of each level area of ​​the curtain wall system to obtain a sensor data matrix; Among them, the curtain wall system is divided into hierarchical areas. By comprehensively considering the orientation, floor height and environmental characteristics of the curtain wall, monitoring points are set on the inner surface, outer surface and hollow layer of the curtain wall to form a systematic monitoring point layout plan. The layout plan not only needs to cover areas of different levels, but also ensure that the monitoring points are evenly distributed and can fully capture the key environmental parameters of the curtain wall system during operation. Based on the monitoring point layout plan, appropriate sensors are installed at each monitoring point to build a sensor network layout. Temperature sensors, light sensors and energy consumption sensors are configured for each monitoring point to ensure that each sensor can accurately capture the corresponding physical quantities, such as temperature, light intensity and energy consumption data. After the sensor network layout is completed, the temperature data of each level of the system is collected by calling the data acquisition module. The collection range of temperature data covers outdoor temperature, indoor temperature and curtain wall surface temperature to ensure that the thermodynamic characteristics of different areas are effectively recorded. After the collection is completed, in order to eliminate the data errors caused by different sensor accuracy, range and installation environment, the collected temperature data is standardized to obtain standardized temperature data. At the same time, under the same sensor network layout, the light data is collected through the data acquisition module. The content of the light data includes outdoor light intensity, indoor light intensity and glass daylight coefficient. These parameters can reflect the lighting performance and light-to-heat conversion efficiency of the curtain wall system. Similarly, the collected light data is standardized to form standardized light data. On this basis, the energy consumption data is collected using energy consumption sensors deployed in the system, including important energy consumption parameters related to curtain wall operation such as shading coefficient, regional energy consumption and air conditioning load. In order to ensure the applicability of the energy consumption data in subsequent analysis, it is standardized to obtain standardized energy consumption data. The standardized temperature data, light data and energy consumption data are input into the data preprocessing module to process the entire data set. The preprocessing module performs outlier detection on the data and removes abnormal data points caused by sensor failure or external interference to improve the accuracy of the data. The cleaned data set is stored as the preprocessing data set. After the preprocessing data set is generated, it is structured according to the time series and region number to construct a three-dimensional data array, where the time dimension records the data changes at different time points, the space dimension identifies the physical location of the sensor, and the parameter dimension stores the standardized values ​​of three types of data: temperature, light, and energy consumption. In order to meet the input requirements of the deep learning model, the three-dimensional data array is processed by tensor transformation to generate a sensor data matrix.

[0014] S2, calculate the curtain wall energy consumption residual value according to the sensor data matrix, and perform time series segmentation and K-means clustering on the curtain wall energy consumption residual value to obtain the energy consumption feature vector and energy consumption pattern feature database; Specifically, principal component analysis is performed on the sensor data matrix to extract the principal components of energy consumption characteristics in the data. Principal component analysis removes redundant information from the data, maps high-dimensional sensor data to a lower-dimensional feature space, and retains the core features that can best explain energy consumption changes in the curtain wall system. An energy consumption benchmark model is established based on the principal components of energy consumption characteristics. The model adopts a three-layer neural network structure to predict curtain wall energy consumption by fitting actual energy consumption data. The input of the neural network is the principal components of energy consumption characteristics, and the output is the predicted energy consumption value, while the middle hidden layer captures the complex energy consumption change relationship through a nonlinear activation function. In order to optimize the fitting performance of the neural network, the back propagation algorithm is used to continuously adjust the network parameters, and the model output is ensured to be close to the actual energy consumption by minimizing the loss function. The fully trained energy consumption benchmark model can accurately predict the curtain wall energy consumption under given conditions. The actual energy consumption data in the sensor data matrix is ​​calculated by difference with the energy consumption value predicted by the energy consumption benchmark model to obtain the difference calculation result. The difference calculation result is processed by sliding average to obtain a more representative curtain wall energy consumption residual value. The sliding average processing eliminates the influence of sudden abnormal changes, so that the residual value can more stably reflect the energy consumption deviation of the curtain wall system. The residual value of curtain wall energy consumption is segmented based on its time series characteristics, and the local energy consumption deviation information is extracted by time period. In order to reveal the potential periodicity and frequency domain characteristics of time series data, the fast Fourier transform is used to perform frequency domain analysis on each segment of data. The frequency domain characteristics reveal the dominant frequency and change pattern of energy consumption deviation in different time periods. Due to the high dimension of frequency domain features, the most distinguishing time series feature sequence is extracted through dimensionality reduction processing, thereby reducing the computational complexity and improving the accuracy of clustering. The time series feature sequence is input into the improved K-means clustering model, and the optimal number of cluster centers is dynamically determined by the silhouette coefficient and Davies-Bouldin index to ensure that the clustering results can reflect the actual structural characteristics of the data. In the clustering process, the K-means algorithm optimizes the compactness of data points within the cluster and the separation of data points between clusters through multiple iterations to obtain the energy consumption pattern clustering results, reflecting the typical energy consumption pattern of the curtain wall system under different operating conditions. Based on the energy consumption pattern clustering results, the statistical features such as the mean vector, variance vector, peak coefficient and skewness coefficient of each cluster are calculated. These features describe the central trend and dispersion of each energy consumption pattern, and reflect the symmetry and extreme value characteristics of its data distribution. Through the feature fusion method, the above statistical features are combined into a complete set of energy consumption feature vectors. The energy consumption feature vectors are normalized and each feature is scaled to a uniform numerical range to obtain a compressed feature vector. The compressed feature vector is associated with the timestamp and spatial location information to form an energy consumption pattern feature database.

[0015] S3, based on the energy consumption feature vector and energy consumption pattern feature database, sets the state space, action space and reward function, and trains the initial deep reinforcement learning model; It should be noted that the energy consumption feature vector and the energy consumption pattern feature database are combined to construct a state vector by integrating indoor temperature parameters, outdoor temperature parameters, light intensity parameters, energy consumption parameters and time parameters into a unified feature set, and then organized into a state space representation matrix in matrix form. According to the state space representation matrix, the controllable variables of the curtain wall are extracted, and the first value range of the shading coefficient, the second value range of the vent opening and the third value range of the hollow layer temperature are set respectively to generate the action space parameter set. These parameters represent the adjustable options in the curtain wall energy-saving control process, and their design needs to comprehensively consider the physical limitations of the actual hardware equipment and the operating constraints of the curtain wall system. Based on the indoor temperature parameters, light intensity parameters and energy consumption parameters in the state space representation matrix, the reward function is defined as an objective function, which aims to balance energy consumption optimization, indoor comfort and lighting conditions. The reward function expression is R=-w1×energy consumption value-w2×|actual room temperature-set room temperature|-w3×|actual illuminance-set illuminance|, where w1, w2, and w3 are weight coefficients, which measure the importance of energy consumption, room temperature deviation and lighting deviation respectively. The reward function penalizes the behavior that deviates from the ideal state in the form of negative values, which prompts the reinforcement learning model to meet the temperature and illumination requirements of the indoor environment while optimizing energy consumption. The deep Q network is constructed using the state space representation matrix and the action space parameter set. The network adopts the Q learning method in deep reinforcement learning. Its architecture includes an input layer, three hidden layers and an output layer. The number of neurons in the hidden layer is set to 256, 128 and 64 respectively. The hierarchical design can capture the complex features in the high-dimensional state space and ensure computational efficiency. The input of the deep Q network is the state vector, and the output is the Q value of the corresponding action. The Q value represents the long-term cumulative reward that can be obtained by selecting an action in a specific state. During the training process of the deep Q network, key training parameters such as the size of the experience replay buffer, the discount factor γ, the exploration factor ϵ and the learning rate α are set. Among them, the experience replay buffer is used to store past state transition samples in order to break the temporal correlation between samples; the discount factor γ controls the importance of long-term rewards; the exploration factor ϵ balances the exploration and utilization of the model; and the learning rate α determines the update amplitude of the model parameters. By inputting the state space representation matrix into the deep Q network, the network calculates the Q value of each action. Combined with the reward function model, the instant reward value is calculated and the target Q value is calculated using the temporal difference algorithm to define the loss function. The loss function measures the difference between the Q value predicted by the deep Q network and the target Q value. The Adam optimization algorithm is used to minimize the loss function and the network parameters are adjusted through back propagation. The continuous optimization of the network parameters enables the deep Q network to gradually approach the optimal Q value function and then learn the optimal control strategy. During the training process, the model gradually improves the effectiveness of its strategy by repeatedly calculating the Q value and updating the network weights.After each training phase, the trained deep Q network is applied to the validation data set for performance evaluation, the average reward value is calculated, and the performance indicators on the validation set are monitored. When the average reward value reaches a convergence state and the validation performance remains stable, the initial deep reinforcement learning model is considered to have been successfully trained. Through this model, the curtain wall control system can achieve intelligent adjustment based on deep learning, optimize energy consumption and improve the comfort of the indoor environment.

[0016] S4, based on the initial deep reinforcement learning model, performs adaptive droop coefficient adjustment on the curtain wall control area, and adjusts the shading coefficient, vent opening and hollow layer temperature to obtain energy-saving control parameters for each area; Specifically, the Q-value output result of the initial deep reinforcement learning model is mapped to the actual building space. According to the orientation characteristics of the curtain wall, the control area is divided into four directions: east, south, west and north. At the same time, it is divided into low, middle and high areas according to the floor height to construct a control unit area matrix. Each unit of the matrix represents a specific spatial position. A droop control model is constructed based on the control unit area matrix. By establishing a linear mapping relationship between the actual energy consumption and the standard energy consumption of each control unit, the corresponding droop characteristic coefficient set is calculated. The droop characteristic coefficient describes the energy consumption deviation characteristics of each control unit area. In order to adapt to changes in different time periods and environmental conditions, the droop characteristic coefficient set is adaptively adjusted to form an adaptive droop coefficient matrix. The matrix can reflect the energy-saving needs of each control unit area in real time and serve as the core parameter input for subsequent calculations. Based on the adaptive droop coefficient matrix, the shading control parameters are calculated. Combined with the sun's altitude angle and azimuth , by establishing a calculation formula for the angle of the sunshade louver , determine the shading angle of the curtain wall system under different solar incidence conditions. represents the solar incidence angle; Represents the solar azimuth angle, which describes the relationship between sunlight and the direction of the curtain wall. Based on the shading angle, combined with the indoor illumination demand value, a shading control matrix is ​​generated to ensure that the heat gain caused by solar radiation can be reduced while meeting the comfort requirements of indoor illumination by dynamically adjusting the angle of the shading blinds. A ventilation control model is established based on the shading control matrix to adjust the ventilation effect of the curtain wall area. The ventilation volume is calculated based on the indoor and outdoor temperature difference and wind speed. By establishing the relationship between the wind flow rate and the vent opening, a ventilation opening matrix is ​​generated. The ventilation opening matrix is ​​input into the heat transfer model, and the temperature distribution of the hollow layer is calculated by solving the heat transfer differential equation to generate a temperature control matrix. In order to achieve overall optimization, the shading control matrix, the ventilation opening matrix and the temperature control matrix are combined to construct a multi-objective optimization function. The form of the optimization function is ,in The energy consumption target per unit area is used to reflect energy-saving efficiency; Indicates the human comfort index value, which is used to evaluate the comfort of indoor temperature and lighting; Represents the control stability index value, which is used to measure the stability of the adjustment process. To balance the factors, the weight distribution among the three objectives is controlled to ensure that the optimization results take into account energy consumption, comfort and stability. The above multi-objective optimization function is input into the particle swarm optimization algorithm, and the energy-saving control parameters of each area are obtained through the iterative optimization calculation of the particle swarm algorithm. The particle swarm optimization algorithm gradually approaches the global optimal solution by simulating the collaborative behavior of the particle group in the search space, thereby generating the optimal shading coefficient, ventilation opening and hollow layer temperature adjustment parameters for each control unit area.

[0017] S5, performing multi-objective optimization control according to the energy-saving control parameters of each area, and updating the initial deep reinforcement learning model online to obtain the target deep reinforcement learning model.

[0018] Among them, the energy-saving control parameters of each area are imported into the control execution unit of the curtain wall, and the control area is subdivided into specific execution units according to the orientation and floor height of the curtain wall. According to the specific needs of each area, the adjustment of the sunshade blind angle, the adjustment of the ventilation opening and the modification of the temperature setting value are executed to generate a regional control execution matrix. This matrix directly corresponds to the specific control actions of each area, ensuring that the control strategy can be distributed and implemented in a targeted manner in space and time. The time interval for data collection is set according to the regional control execution matrix, and the sensor network installed in the curtain wall system is used to collect real-time environmental parameter data. The temperature change information of indoor and outdoor is collected by the temperature sensor, and the fluctuation of indoor and outdoor light intensity is captured by the light sensor. At the same time, the actual energy consumption data of different areas is monitored by the energy consumption sensor. These data are integrated into an environmental parameter set to describe the current system operation status and external environmental conditions. In order to ensure the consistency and availability of the data, the environmental parameter set is normalized and the data is converted into a standardized parameter matrix. The standardized parameter matrix is ​​input into the initial deep reinforcement learning model. The model extracts the characteristic state of the current system through state space mapping and selects the optimal adjustment action in the action space. At the same time, based on the preset reward function, the instant reward value of the current action is calculated, so as to output the optimized control vectors of each area. These vectors contain specific shading adjustment, ventilation adjustment and temperature adjustment, which represent the optimal adjustment scheme of the model for the current operating state. In order to enable the optimized control vector to be applied to actual control, it is decoded and mapped. Through decoding, the normalized control quantity is restored to the actual shading louver angle value, ventilation opening value and temperature setting value. The decoded control quantity is allocated according to the specific control area to generate the execution control instruction. After the execution control instruction is generated, the actuator of the drive system is operated, including adjusting the angle of the shading louver motor to change the shading state, adjusting the opening of the ventilation window to optimize the ventilation volume, and controlling the temperature of the hollow layer to adjust the heat exchange. These actions are executed independently in different areas, and the execution status data is collected by sensors to obtain real-time system response. These response data are not only used to verify the execution effect of the control instruction, but also provide feedback information for the online update of the model. Based on the system response data, the energy-saving index and comfort index are calculated. The energy-saving index evaluates the overall energy efficiency of the curtain wall system, and the comfort index measures the impact of the temperature and light of the indoor environment on the human body. Through the incremental learning method, these indicators are fed back to the initial deep reinforcement learning model to gradually update the model parameters. The incremental learning method can adapt the model to the changing environmental conditions and operation requirements by combining the new data optimization strategy without destroying the original model performance. Through this process, the initial deep reinforcement learning model gradually evolves into the target deep reinforcement learning model. The target model has higher control accuracy and adaptability, and can achieve long-term stable operation in complex practical application scenarios, thereby improving the energy saving effect and user comfort of the curtain wall system.

[0019] Energy consumption data is extracted from the system response data, and the 24-hour operation status of each control area is statistically analyzed in time series. The cumulative energy consumption per unit area, the peak-to-valley energy consumption ratio, and the standard deviation of energy consumption are calculated. These statistics reflect the energy consumption intensity, fluctuation amplitude, and uniformity of energy consumption distribution of the curtain wall system. By weighted averaging these indicators, an energy-saving performance index matrix is ​​generated. At the same time, indoor environmental parameters are extracted from the system response data to comprehensively evaluate the comfort performance. By combining physical parameters such as indoor temperature, humidity, air flow velocity, and radiation temperature, the predicted mean voting value (PMV) is calculated, which is a comprehensive indicator reflecting human thermal comfort. The PMV value measures comfort by quantifying the deviation between the environment and the human body's thermal balance. Visual comfort is evaluated by calculating the illumination uniformity and glare value, which respectively describe the uniformity of light distribution and the potential stimulation of the light environment to the human eye. Based on these calculation results, a comfort performance index matrix is ​​generated to quantify the comfort effect of the curtain wall system under operation in different areas. According to the energy-saving performance index matrix and the comfort performance index matrix, a model performance evaluation function is constructed to comprehensively evaluate the actual performance of the initial deep reinforcement learning model. The performance evaluation function reasonably distributes the weights of energy saving and comfort indicators to generate a global performance score. When the output value of the model performance evaluation function is lower than the preset threshold, it means that the control strategy of the current model cannot meet the system optimization goal, triggering the model update mechanism and generating an update trigger instruction. After the update mechanism is started, the system response data within 24 consecutive hours is sampled to construct model training data containing state transition sequence, action sequence and reward sequence. These data reflect the dynamic behavior of the control system in actual operation and are an important basis for enhancing the adaptability of the model. The sampled data is subjected to online feature extraction to mine the dynamic feature vectors. These feature vectors capture the transition mode of the system between different states, the influence range of the action, and the regularity of the reward change. The extracted dynamic feature vectors are input into the initial deep reinforcement learning model to perform incremental training. The incremental training method gradually updates the model parameters so that it can absorb the operating laws reflected by the new data while maintaining the stability of the original model performance. In this process, the loss function of the model is dynamically adjusted to minimize the deviation between the new and old strategies while ensuring the rapid adaptation of the model to new environmental conditions. After multiple rounds of training and optimization, the target deep reinforcement learning model is finally formed. This model has higher control accuracy and flexibility, and can continuously optimize the energy-saving performance and user comfort of the curtain wall system in a changing environment.

[0020] In one example, temperature data, light data, and energy consumption data are collected for each level of the curtain wall system to obtain a sensor data matrix, including: Divide the curtain wall system into hierarchical areas, set monitoring points on the inner surface, outer surface and hollow layer according to the curtain wall orientation and floor height, and obtain a monitoring point layout plan; Based on the monitoring point layout plan, temperature sensors, light sensors and energy consumption sensors are installed and configured at each monitoring point to obtain the sensor network layout; According to the sensor network layout, the data acquisition module is called to collect temperature data, which includes outdoor temperature, indoor temperature and curtain wall surface temperature, and the temperature data is standardized to obtain standardized temperature data; Based on the sensor network layout, the data acquisition module is called to collect light data, which includes outdoor light intensity, indoor light intensity and glass daylight factor, and the light data is standardized to obtain standardized light data; Energy consumption data is collected through energy consumption sensors in the sensor network layout. The energy consumption data includes shading coefficient, regional energy consumption and air conditioning load, and the energy consumption data is standardized to obtain standardized energy consumption data; The standardized temperature data, the standardized light data and the standardized energy consumption data are input into the data preprocessing module to detect and delete outliers, thereby obtaining a preprocessed data set; The preprocessed data set is constructed into a three-dimensional data array according to the time series and area number. The three-dimensional data array includes time dimension, space dimension and parameter dimension, and the sensor data matrix is ​​obtained through tensor transformation.

[0021] In this example, the curtain wall system is divided into hierarchical areas according to the specific structure of the building and the distribution characteristics of the curtain wall, and grouped according to the orientation of the curtain wall (such as east, south, west, north) and the height of the floor (such as low area, middle area, high area). In each area, monitoring points are set on the inner surface, outer surface and hollow layer according to the physical structure of the curtain wall to capture the characteristics of heat transfer, light changes and energy consumption data, forming a complete monitoring point layout plan. According to the layout plan of the monitoring points, sensors are installed and configured for each monitoring point, including temperature sensors, light sensors and energy consumption sensors. The temperature sensor is used to monitor the indoor temperature ( )、Outdoor temperature( ) and curtain wall surface temperature ( The light sensor records the outdoor light intensity ( ), indoor light intensity ( ) and glass daylight factor ( ), where the glass daylighting factor is given by the formula Calculation is used to quantify the efficiency of light energy passing through glass. Energy consumption sensors are used to collect shading coefficients ( )、Regional energy consumption( ) and air conditioning load ( ), these data reflect the adjustment efficiency of the shading system, the overall energy consumption level and the cooling or heating load. After the sensor network layout is completed, the data acquisition module collects and processes various parameters. Collect temperature data, including , and Since different sensors have different ranges or resolutions, the temperature data is standardized. The standardization formula is: ; in, represents the original temperature data, is the mean of this type of data, is the standard deviation, is the standardized temperature data. After standardization, the scale difference in the temperature data is eliminated, making it suitable for further analysis. Based on the sensor network layout, light data is collected, including , and Similarly, the illumination data is standardized, and the standardization formula is: ; in, represents the original lighting data, and are the mean and standard deviation of the illumination data. The standardized illumination data can reflect the relative illumination characteristics of different areas. The energy consumption data collected by the energy consumption sensor includes , and , also needs to be standardized. The formula is similar to the standardization method of temperature and light: ; in, is the original energy consumption data, and are their mean and standard deviation respectively. The standardized temperature data, light data and energy consumption data are input into the data preprocessing module to detect and clean up the data set for outliers. By setting upper and lower thresholds (for example, based on the 95% confidence interval of historical data), outliers that may be caused by sensor failure or environmental interference are eliminated to obtain a preprocessed data set. The preprocessed data set is constructed into a three-dimensional data array according to the time series and area number. The three-dimensional data array includes time dimensions (such as time points within 24 hours per day), spatial dimensions (such as the locations of monitoring points on different floors and orientations), and parameter dimensions (such as , , ). Through tensor transformation operations, the three-dimensional data array is converted into a unified sensor data matrix to describe the operating status of the entire curtain wall system. For example, the sensor data matrix at a specific moment is expressed as: ; in, , and Respectively, in time Below, located at Tier Through the above steps, the sensor data matrix is ​​finally obtained.

[0022] In one example, the curtain wall energy consumption residual value is calculated according to the sensor data matrix, and the curtain wall energy consumption residual value is time-series segmented and K-means clustered to obtain the energy consumption feature vector and energy consumption pattern feature database, including: Perform principal component analysis on the sensor data matrix to obtain the principal components of energy consumption characteristics; An energy consumption benchmark model is established based on the main components of energy consumption characteristics, a three-layer neural network structure is used to fit the energy consumption data, and the network parameters are optimized through the back propagation algorithm to obtain the energy consumption prediction value; The actual energy consumption data in the sensor data matrix is ​​subjected to difference calculation with the predicted energy consumption value to obtain the difference calculation result, and the difference calculation result is subjected to sliding average processing to obtain the residual value of curtain wall energy consumption; The residual value of curtain wall energy consumption is segmented, the frequency domain features of each segment of data are extracted by fast Fourier transform, and the frequency domain features are reduced in dimension to obtain the time series feature sequence; The time series feature sequence is input into the improved K-means clustering model, the silhouette coefficient and Davies-Bouldin index are used to determine the optimal number of cluster centers, and iterative clustering calculation is performed to obtain the energy consumption pattern clustering results; According to the clustering results of energy consumption patterns, the mean vector, variance vector, peak coefficient and skewness coefficient of each cluster are calculated, and the energy consumption feature vector is obtained through feature fusion; The energy consumption feature vector is normalized to obtain a compressed feature vector, and the compressed feature vector is associated with the timestamp and spatial location information to obtain an energy consumption pattern feature database.

[0023] In this example, the sensor data matrix is ​​used to perform principal component analysis on energy consumption related features to reduce the complexity of high-dimensional data while retaining the key information in the data that can best explain the energy consumption changes. Contains multiple variable dimensions, such as standardized temperature data , Lighting Data and energy consumption data Assumptions is a Matrix, where is the length of the time series, is the number of parameters, by calculating the covariance matrix And decompose its eigenvalues ​​and eigenvectors to obtain the principal component matrix . Each column of the principal component matrix is ​​a linear combination, for example: ; in, It is principal components, is the linear combination coefficient, representing the The variable The contribution weight of the principal component is calculated. An energy consumption benchmark model is established based on the principal component of energy consumption characteristics. The model adopts a three-layer neural network structure to fit the nonlinear relationship between actual energy consumption data and principal component characteristics. The number of nodes in the input layer of the network is equal to the number of principal components, and the number of neurons in the hidden layer is set to and The output layer is the energy consumption prediction value The prediction process of the neural network is expressed as: ; in, and is the weight matrix, and is the bias vector, is the activation function (such as ReLU), is a linear transformation of the output layer. By minimizing the loss function , and use the back-propagation algorithm to update the parameters and , and obtain a neural network model that can accurately predict energy consumption. Actual energy consumption data Perform difference calculation to obtain difference sequence : ; In order to eliminate the interference of short-term fluctuations on the analysis, the sliding average processing is performed on the difference series to generate a smooth curtain wall energy consumption residual value. : ; in, is the size of the sliding window, is the time point. The energy consumption residual value is segmented according to the time series, and a fast Fourier transform is performed on each segment of data to extract the frequency domain features. The fast Fourier transform converts the time domain signal into the frequency domain signal, and its formula is: ; in, is the frequency domain signal, is the frequency, is the data length. The frequency domain features reveal the distribution of energy consumption residuals on different frequency components. Through dimensionality reduction techniques (such as principal component analysis or feature screening), the most distinguishing time series feature sequences are extracted. The time series feature sequences are input into the improved K-means clustering model, and the silhouette coefficient is used to and Davies-Bouldin index Determine the optimal number of cluster centers The silhouette coefficient calculation formula is: ; in, is the average distance from the sample to other points in the cluster, is the average distance from the sample to the nearest neighbor cluster. The Davies-Bouldin index is: ; in, and It is a cluster and The diameter of is the distance between cluster centers. Through iterative clustering calculation, the energy consumption pattern clustering result is obtained. According to the clustering result, the mean vector of each cluster is calculated. , variance vector , peak factor and the skewness coefficient : ; in, is the data sample within the cluster, is the number of samples. Through feature fusion, these statistical indicators are combined into a complete energy consumption feature vector. The energy consumption feature vector is normalized to eliminate the scale effect. The normalization formula is: ; The normalized compressed feature vector is associated with the timestamp and spatial location information to form an energy consumption pattern feature database. For example, for a cluster center, its feature vector is recorded. and time and location The database is used for energy consumption pattern analysis and optimization control decisions.

[0024] In one example, based on the energy consumption feature vector and the energy consumption pattern feature database, the state space, action space and reward function are set, and the initial deep reinforcement learning model is trained, including: The energy consumption feature vector and the energy consumption pattern feature database are combined to construct a state vector including indoor temperature parameters, outdoor temperature parameters, light intensity parameters, energy consumption parameters and time parameters, and obtain a state space representation matrix; The curtain wall control variables are extracted according to the state space representation matrix, and the first value range of the shading coefficient, the second value range of the vent opening, and the third value range of the hollow layer temperature are set to obtain the action space parameter set; According to the indoor temperature parameter in the state space representation matrix, the light intensity parameter in the state space representation matrix and the energy consumption parameter in the state space representation matrix, a reward function R=-w1×energy consumption value-w2×|actual room temperature-set room temperature|-w3×|actual illuminance-set illuminance| is constructed to obtain a reward function model, where R represents the reward function, and w1, w2, and w3 represent weight coefficients; A deep Q network is constructed based on the state space representation matrix and the action space parameter set. The deep Q network includes an input layer, three hidden layers, and an output layer. The number of neurons in the hidden layers is 256, 128, and 64, respectively. Set the experience replay buffer size, discount factor γ, exploration factor ε and learning rate α for the deep Q network to obtain the deep Q network training parameters; Input the state space representation matrix into the deep Q network, calculate the Q value through the deep Q network, use the reward function model to calculate the immediate reward, calculate the target Q value based on the temporal difference algorithm, and obtain the loss function value; Back propagation operation is performed according to the loss function value, and the Adam optimization algorithm is used to update the network parameters to obtain the trained deep Q network. The performance of the trained deep Q network is evaluated on the validation data set. When the average reward value converges and the performance indicators on the validation set are stable, the initial deep reinforcement learning model is obtained.

[0025] In this example, the energy consumption feature vector and the energy consumption pattern feature database are combined to obtain the indoor temperature parameter. , Outdoor temperature parameters , Light intensity parameters , Energy consumption parameters and time parameters Integrate into a unified set of state descriptions to construct a state vector For a specific moment , the state vector is expressed as: ; Combine the state vectors at all times to form a state space representation matrix , where each row represents the state at a moment, and the columns represent different parameter dimensions. After the state space is defined, the control variables of the curtain wall system, including the shading coefficient, are extracted according to the state space representation matrix. , Ventilation opening and hollow layer temperature These variables control the behavior of the shading system, ventilation system, and thermal system, respectively. In order to ensure that the value range of the control variables is reasonable, a specific range is set for each variable, the first value range of the shading coefficient, the second value range of the vent opening, and the third value range of the hollow layer temperature are set to obtain the action space parameter set ,Right now: ; Based on the state space representation matrix and energy consumption parameters , Indoor temperature parameters , Light intensity parameters , build a reward function model The reward function penalizes undesirable energy consumption, temperature deviation, and illumination deviation in the form of negative values. The specific expression is: ; in, , and are weight coefficients, which respectively represent the relative importance of energy consumption, room temperature deviation and light deviation to the overall optimization goal; and are the set room temperature and light target values ​​respectively. Based on the state space representation matrix and the action space parameter set , construct a deep Q network, whose structure includes an input layer, three hidden layers and an output layer. The number of nodes in the input layer is equal to the dimension of the state vector, for example, , , , , and time There are six nodes in total. The hidden layer is set to three layers, with the number of neurons being 256, 128, and 64 respectively. The layer-by-layer reduction design helps capture high-dimensional features in the state space and compress them effectively. The number of nodes in the output layer is equal to the dimension of the action space, and the output value of each node corresponds to the Q value of an action. In order to train the deep Q network, the experience replay buffer size is set , Discount Factor , Exploration Factor and learning rate . Among them, the buffer size Used to store past state transition samples; discount factor Determine the weight of future rewards; exploration factor Used to control the balance between random exploration and greedy selection; learning rate Determines the step size of the model parameter update. During the training process, the state space representation matrix Input the deep Q network, the network is based on the current state Calculate the Q-values ​​of all possible actions . Instant Rewards It is calculated through the reward function model, and the target Q value is calculated according to the temporal difference algorithm: ; Based on the predicted Q value and the target Q value, the loss function is defined as: ; in, is the sample batch size. The back propagation algorithm is used to optimize the loss function, and the Adam optimization algorithm is used to adjust the parameters of the deep Q network, including weights and biases. The model parameters are continuously updated during the training process, and the loss function is finally converged. The performance of the trained deep Q network is evaluated on the validation data set, and the average reward value is calculated. and other performance indicators (such as mean square error) to judge the effect of the model. When convergence and the indicators on the validation set are stable, the initial deep reinforcement learning model is obtained.

[0026] In one example, based on the initial deep reinforcement learning model, adaptive droop coefficient adjustment is performed on the curtain wall control area, and the shading coefficient, vent opening and hollow layer temperature are adjusted to obtain energy-saving control parameters for each area, including: The Q-value output results of the initial deep reinforcement learning model are mapped according to the building space attributes. The control area is divided into the east control area, the south control area, the west control area and the north control area according to the curtain wall orientation. The control area is divided into the low control area, the middle control area and the high control area according to the floor height to obtain the control unit area matrix. A droop control model is established based on the control unit area matrix, a linear mapping relationship is established between the actual energy consumption of each control unit and the standard energy consumption, a droop characteristic coefficient set is obtained, and the droop characteristic coefficient set is adaptively adjusted to obtain an adaptive droop coefficient matrix; The shading coefficient is calculated based on the adaptive droop coefficient matrix, and the calculation formula of the shading shutter angle αs is established according to the solar altitude angle θs and the azimuth angle φs: αs=arctan(tanθs×cosφs), where θs represents the solar incidence angle, and φs represents the solar azimuth angle. Combined with the indoor illumination demand value, the shading control matrix is ​​obtained. A ventilation control model is established according to the shading control matrix, and the ventilation volume is calculated based on the indoor and outdoor temperature difference and wind speed to obtain the ventilation opening matrix. The ventilation opening matrix is ​​input into the heat transfer model, and the temperature distribution of the hollow layer is calculated by solving the heat transfer differential equation to obtain the temperature control matrix. The shading control matrix, ventilation opening matrix and temperature control matrix are combined to construct a multi-objective optimization function F=λ1×E1+λ2×E2+λ3×E3, where E1 represents the target value of energy consumption per unit area, E2 represents the human comfort index value, E3 represents the control stability index value, and λ1, λ2, and λ3 represent balance factors. The multi-objective optimization function is input into the particle swarm optimization algorithm, and the energy-saving control parameters of each area are obtained through iterative optimization calculation.

[0027] In this example, the Q-value output of the initial deep reinforcement learning model is mapped according to the spatial attributes of the building, and combined with the curtain wall characteristics of the building, it is divided into east, south, west and north control areas according to the orientation, and subdivided into low, middle and high areas according to the floor height. Through these two-dimensional divisions, a control unit area matrix is ​​formed. , each element in the matrix Corresponding to Direction and On this basis, a droop control model is constructed to dynamically adjust the energy consumption level of each control unit. For each control unit, the actual energy consumption is calculated. Standard energy consumption The difference between them is established by linear mapping, and the droop characteristic coefficient is defined. for: ; in Describes the deviation of each unit from the standard energy consumption. By continuously collecting new energy consumption data, the droop characteristic coefficient Perform adaptive adjustment to generate an adaptive droop coefficient matrix , whose elements Indicates Direction, The droop characteristics of each floor. Based on the adaptive droop coefficient matrix , calculate the shading coefficient Combined with the sun's altitude angle and azimuth , derive the adjustment angle of the sunshade blinds through geometric relationships , the formula is: ; in, represents the incident angle of sunlight; It is the angle of sunlight relative to the curtain wall. Combined with the required indoor illumination value , adjust the shading coefficient To meet the illumination requirements, a sunshade control matrix is ​​finally formed. According to the shading control matrix , establish ventilation control model and calculate the opening of ventilation opening . Ventilation volume The calculation formula is: ; in, is the air volume coefficient, Indicates wind speed, is the opening of the vent. Combined with the temperature difference between indoor and outdoor , adjust the opening of the vents so that the temperature change of the indoor environment meets the comfort requirements, forming a ventilation opening matrix . The ventilation opening matrix Input the heat transfer model and calculate the temperature distribution in the hollow layer by solving the heat transfer differential equation : ; in, is the thermal diffusivity, is the spatial position, is the time. The solved temperature distribution is used to generate the temperature control matrix . The shading control matrix , ventilation opening matrix and temperature control matrix Combine and construct a multi-objective optimization function , which is of the form: ; in, It represents the target value of energy consumption per unit area, which is calculated as: ; It represents the human comfort index value, taking into account the temperature, humidity and light conditions, and is expressed as the predicted mean voting value (PMV). The calculation formula of PMV is: ; in is the metabolic rate, is the heat load. Represents the control stability index value, which is used to constrain the frequent changes of control actions. The multi-objective optimization function is optimized by particle swarm optimization algorithm. Perform iterative optimization, the initial particle position is the random value of the control parameter, and the particle update rule is: ; ; in, is the particle velocity, is the particle position, is the best historical position of the particle, is the global optimal position, , , are the inertia weight and learning factor respectively, , is a random number. Through multiple iterations of optimization, the energy-saving control parameters of each area are finally obtained, including the optimized shading coefficient, ventilation opening and hollow layer temperature value.

[0028] In one example, multi-objective optimization control is performed according to the energy-saving control parameters of each area, and the initial deep reinforcement learning model is updated online to obtain a target deep reinforcement learning model, including: The energy-saving control parameters of each area are imported into the control execution unit, and the sunshade shutter angle adjustment, ventilation opening adjustment and temperature setting value adjustment are performed according to the control area corresponding to the curtain wall orientation and floor height to obtain the regional control execution matrix; According to the regional control execution matrix, the data collection time interval is set, the indoor and outdoor temperature data are collected through the temperature sensor, the indoor and outdoor light intensity data are collected through the light sensor, and the regional energy consumption data are collected through the energy consumption sensor to obtain the environmental parameter set; Normalize the environmental parameter set to obtain a standardized parameter matrix, and input the standardized parameter matrix into the initial deep reinforcement learning model. After state space mapping, action space selection and reward function calculation, the shading adjustment amount, ventilation adjustment amount and temperature adjustment amount of each area are output to obtain the optimized control vector. Decode and map the optimized control vector, restore the normalized control quantity to the actual sunshade louver angle value, ventilation opening value and temperature setting value, and distribute them according to the control area to obtain the execution control instruction; According to the execution control instructions, the actuator is driven to operate, the sunshade shutter motor angle, ventilation window opening and hollow layer temperature are adjusted, and the execution status data is collected to obtain the system response data; The energy-saving index and comfort index are calculated based on the system response data, and the initial deep reinforcement learning model is updated online using the incremental learning method to obtain the target deep reinforcement learning model.

[0029] In this example, the energy-saving control parameters of each area are imported into the control execution unit, and the building is divided into zones according to the specific orientation of the curtain wall (such as east, south, west, north) and the floor height (low zone, middle zone, high zone). Each control zone corresponds to an execution unit, and its control parameters include the angle of the sunshade louver. , ventilation opening and temperature setting These parameters are organized into a regional control execution matrix , the elements of the matrix Indicates Direction and After the control parameters are assigned, the control matrix is ​​executed according to the area control. Set the time interval for data collection, for example, collect data every 15 minutes, and record the indoor temperature through the temperature sensor and outdoor temperature , collect indoor light intensity through light sensors and outdoor light intensity and measure regional energy consumption through energy consumption sensors These data are combined into the environmental parameter set , which is used to describe the current state of the curtain wall system. The environmental parameter set is normalized, and the normalization formula is: ; in, is the original data, and are the minimum and maximum values ​​of the parameter respectively. The standardized parameter matrix is ​​obtained by normalization , so that all data values ​​are in the range of [0,1], which is convenient for subsequent input into the deep reinforcement learning model. Input into the initial deep reinforcement learning model, the model extracts features of the current system state through state space mapping and obtains the state vector , which contains elements including , , , , and time parameters The deep reinforcement learning model selects the optimal action in the action space A and calculates the Q value of the current state. . Calculate the immediate reward through the reward function model : ; in, , , are weight coefficients, representing the impact of energy consumption, room temperature deviation, and light deviation on the total reward, and are the set room temperature and light target values ​​respectively. The model outputs the optimized control vector The optimized control vector needs to be decoded and mapped to restore the normalized control quantity to the actual value. For example, the decoding formula for the sunshade louver angle is: ; in, and are the minimum and maximum values ​​of the blinds angle, respectively. Similarly, the actual values ​​of the ventilation opening and temperature setting are also calculated by the corresponding decoding formula. After decoding, the control parameters of each area are assigned to the corresponding execution unit to generate the final execution control instruction. The execution control instruction is passed to the hardware layer to drive the sunshade blinds motor to adjust the angle. , adjust the ventilation window opening , and set the hollow layer temperature At the same time, the sensor collects the execution status data in real time, such as the actual adjusted louver angle, ventilation volume and hollow layer temperature, to generate system response data. Based on the system response data, the energy saving index and comfort index are calculated. Energy saving index Expressing energy consumption per unit area, the formula is: ; The comfort index It is expressed by the predicted mean voting value (PMV), and the specific formula is: ; in is the metabolic rate, is the human body heat load. Combined with the system response data, incremental learning is performed on the initial deep reinforcement learning model. The incremental learning method samples a new state transition sequence Construct training data and optimize the model online. During the update process, the target Q value is calculated by the temporal difference formula: ; And by minimizing the loss function The model parameters are updated and the optimization algorithm adopts the Adam optimizer to finally form the optimized target deep reinforcement learning model and realize a dynamic and adaptive energy-saving control strategy.

[0030] In one example, the energy saving index and the comfort index are calculated based on the system response data, and the initial deep reinforcement learning model is updated online using the incremental learning method to obtain the target deep reinforcement learning model, including: Perform time series statistics on the energy consumption data in the system response data, calculate the cumulative energy consumption per unit area, peak-to-valley energy consumption ratio and energy consumption standard deviation of each control area within 24 hours, and obtain the energy-saving performance indicator matrix through weighted average; Evaluate and calculate the indoor environmental parameters in the system response data, calculate the predicted mean voting value PMV based on temperature, humidity, air flow velocity and radiation temperature, calculate the visual comfort based on illumination uniformity and glare value, and obtain the comfort performance index matrix; According to the energy-saving performance index matrix and the comfort performance index matrix, a model performance evaluation function is constructed, and the performance of the initial deep reinforcement learning model is evaluated. When the output value of the model performance evaluation function is less than the preset threshold, the model update mechanism is activated to obtain an update trigger instruction; Based on the update trigger instruction, the system response data for 24 consecutive hours is sampled to construct model training data including state transition sequence, action sequence and reward sequence; Online feature extraction is performed on the model training data to obtain a dynamic feature vector, and the dynamic feature vector is input into the initial deep reinforcement learning model for incremental training to obtain the target deep reinforcement learning model.

[0031] In this example, the energy consumption data of each control area is extracted from the system response data, and time series statistical analysis is performed on it. The cumulative energy consumption per unit area of ​​each control area within 24 hours is calculated. , the formula is: ; in, It is the control area In time The actual energy consumption value, is the area of ​​the region. Calculate the peak-to-valley energy consumption ratio of each control area : ; This indicator reflects the volatility of energy consumption. Calculate the standard deviation of energy consumption for each control area. , which indicates the discrete degree of energy consumption: ; in, It is a region The average energy consumption, is the number of time points. After weighted average of the above three indicators, the energy-saving performance indicator matrix is ​​constructed. , whose elements Corresponding to Region No. At the same time, indoor environmental parameters are extracted from the system response data to evaluate the comfort performance index. ,humidity , air flow velocity and radiation temperature , calculate the predicted mean voting value (PMV), the formula is: ; in, is the metabolic rate, Indicates the heat load of the human body, which is calculated by comprehensive factors such as radiation heat transfer, convection heat transfer and evaporative heat dissipation. and glare value To evaluate visual comfort, the calculations are: ; ; in and are the minimum and maximum illumination, For the the brightness of the light source, is the average brightness, is the number of light sources. Based on PMV and visual comfort, a comfort performance index matrix is ​​constructed. By combining the energy-saving performance indicator matrix and comfort performance index matrix , define the model performance evaluation function: ; in, and are the target values ​​for energy saving and comfort, and is the weight. When the performance evaluation function When the value is less than the preset threshold, the model update mechanism is triggered. Based on the update trigger instruction, the system response data is sampled for 24 consecutive hours to generate a state transition sequence. ,in, Indicates the current state, including environmental parameters; is the action vector, including shading, ventilation, and temperature regulation; For immediate rewards; After constructing the action sequence and reward sequence, the input data is used for online feature extraction to extract the dynamic feature vector : ; in is a feature extraction function used to capture time correlation. Input to the initial deep reinforcement learning model for incremental training. Update the target Q value through the temporal difference algorithm: ; in Is the discount factor. Define the loss function: ; The Adam optimization algorithm is used to minimize the loss function and update the model parameters. After multiple rounds of iterative training, the model gradually learns the latest laws of environmental changes and eventually forms the target deep reinforcement learning model.

[0032] Reference Figure 2 , this embodiment provides an intelligent curtain wall energy-saving control system based on deep learning, including: The acquisition module 1 is used to collect temperature data, light data and energy consumption data of each level area of ​​the curtain wall system to obtain a sensor data matrix; Calculation module 2, used to calculate the curtain wall energy consumption residual value according to the sensor data matrix, and perform time series segmentation and K-means clustering on the curtain wall energy consumption residual value to obtain energy consumption feature vectors and energy consumption pattern feature database; Training module 3, used to set the state space, action space and reward function based on the energy consumption feature vector and the energy consumption pattern feature database, and train the initial deep reinforcement learning model; Adjustment module 4 is used to perform adaptive droop coefficient adjustment on the curtain wall control area based on the initial deep reinforcement learning model, and adjust the shading coefficient, vent opening and hollow layer temperature to obtain energy-saving control parameters of each area; The updating module 5 is used to perform multi-objective optimization control according to the energy-saving control parameters of each area, and to update the initial deep reinforcement learning model online to obtain a target deep reinforcement learning model.

[0033] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0034] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0035] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0036] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0037] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0038] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0039] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A smart curtain wall energy-saving control method based on deep learning, characterized in that: The following steps are involved: Collect temperature data, lighting data, and energy consumption data for each level of the curtain wall system to obtain a sensor data matrix; Calculating the curtain wall energy consumption residual value according to the sensor data matrix, and performing time series segmentation and K-means clustering on the curtain wall energy consumption residual value to obtain an energy consumption feature vector and an energy consumption pattern feature database; Based on the energy consumption feature vector and the energy consumption pattern feature database, setting a state space, an action space, and a reward function, and training an initial deep reinforcement learning model; Based on the initial deep reinforcement learning model, adaptive droop coefficient adjustment is performed on the curtain wall control area, and the shading coefficient, vent opening and hollow layer temperature are adjusted to obtain energy-saving control parameters for each area; Multi-objective optimization control is performed according to the energy-saving control parameters of each area, and the initial deep reinforcement learning model is updated online to obtain a target deep reinforcement learning model.

2. The deep learning-based intelligent curtain wall energy-saving control method according to claim 1 is characterized in that: The temperature data, light data and energy consumption data of each level area of ​​the curtain wall system are collected to obtain a sensor data matrix, including: Divide the curtain wall system into hierarchical areas, set monitoring points on the inner surface, outer surface and hollow layer according to the curtain wall orientation and floor height, and obtain a monitoring point layout plan; Based on the monitoring point layout plan, temperature sensors, light sensors, and energy consumption sensors are installed and configured at each monitoring point to obtain a sensor network layout; According to the sensor network layout, calling a data acquisition module to collect temperature data, the temperature data including outdoor temperature, indoor temperature and curtain wall surface temperature, and performing standardization processing on the temperature data to obtain standardized temperature data; Based on the sensor network layout, calling a data acquisition module to collect light data, the light data including outdoor light intensity, indoor light intensity and glass daylight factor, and performing standardization processing on the light data to obtain standardized light data; Collecting energy consumption data through energy consumption sensors in the sensor network layout, the energy consumption data including shading coefficient, regional energy consumption and air conditioning load, and performing standardization processing on the energy consumption data to obtain standardized energy consumption data; Inputting the standardized temperature data, the standardized illumination data, and the standardized energy consumption data into a data preprocessing module to perform outlier detection and deletion to obtain a preprocessed data set; The preprocessed data set is constructed into a three-dimensional data array according to the time series and the region number. The three-dimensional data array includes a time dimension, a space dimension and a parameter dimension, and a sensor data matrix is ​​obtained through tensor transformation.

3. The deep learning-based intelligent curtain wall energy-saving control method according to claim 2 is characterized in that: The method of calculating the curtain wall energy consumption residual value according to the sensor data matrix, and performing time series segmentation and K-means clustering on the curtain wall energy consumption residual value to obtain an energy consumption feature vector and an energy consumption pattern feature database includes: Performing principal component analysis on the sensor data matrix to obtain principal components of energy consumption characteristics; An energy consumption benchmark model is established based on the principal components of the energy consumption characteristics, a three-layer neural network structure is used to fit the energy consumption data, and the network parameters are optimized by the back propagation algorithm to obtain the energy consumption prediction value; Performing a difference calculation between the actual energy consumption data in the sensor data matrix and the energy consumption prediction value to obtain a difference calculation result, and performing a sliding average process on the difference calculation result to obtain a curtain wall energy consumption residual value; The curtain wall energy consumption residual value is segmented, the frequency domain features of each segment of data are extracted by fast Fourier transform, and the frequency domain features are subjected to dimensionality reduction processing to obtain a time series feature sequence; Inputting the time series feature sequence into the improved K-means clustering model, using the silhouette coefficient and Davies-Bouldin index to determine the optimal number of cluster centers, performing iterative clustering calculations, and obtaining energy consumption pattern clustering results; According to the energy consumption pattern clustering results, the mean vector, variance vector, peak coefficient and skewness coefficient of each cluster are calculated, and the energy consumption feature vector is obtained by feature fusion; The energy consumption feature vector is normalized to obtain a compressed feature vector, and the compressed feature vector is associated with a timestamp and spatial position information to obtain an energy consumption pattern feature database.

4. The deep learning-based intelligent curtain wall energy-saving control method according to claim 3 is characterized in that: The step of setting a state space, an action space, and a reward function based on the energy consumption feature vector and the energy consumption pattern feature database, and training an initial deep reinforcement learning model includes: Performing feature combination on the energy consumption feature vector and the energy consumption pattern feature database, constructing a state vector including indoor temperature parameters, outdoor temperature parameters, light intensity parameters, energy consumption parameters, and time parameters, and obtaining a state space representation matrix; Extracting curtain wall control variables according to the state space representation matrix, setting a first value range of the shading coefficient, a second value range of the vent opening, and a third value range of the hollow layer temperature to obtain an action space parameter set; According to the indoor temperature parameter in the state space representation matrix, the light intensity parameter in the state space representation matrix, and the energy consumption parameter in the state space representation matrix, a reward function R=-w1×energy consumption value-w2×|actual room temperature-set room temperature|-w3×|actual illuminance-set illuminance| is constructed to obtain a reward function model, where R represents the reward function, and w1, w2, and w3 represent weight coefficients; Constructing a deep Q network based on the state space representation matrix and the action space parameter set, wherein the deep Q network comprises an input layer, three hidden layers, and an output layer, wherein the number of neurons in the hidden layers is 256, 128, and 64, respectively; Setting the experience replay buffer size, discount factor γ, exploration factor ε, and learning rate α for the deep Q network to obtain deep Q network training parameters; Inputting the state space representation matrix into the deep Q network, calculating the Q value through the deep Q network, calculating the immediate reward using the reward function model, calculating the target Q value based on the temporal difference algorithm, and obtaining the loss function value; A backpropagation operation is performed according to the loss function value, and the network parameters are updated using the Adam optimization algorithm to obtain a trained deep Q network. The performance of the trained deep Q network is evaluated on a validation dataset. When the average reward value converges and the performance indicators on the validation set are stable, the initial deep reinforcement learning model is obtained.

5. The deep learning-based intelligent curtain wall energy-saving control method according to claim 4 is characterized in that: Based on the initial deep reinforcement learning model, adaptive droop coefficient adjustment is performed on the curtain wall control area, and the shading coefficient, vent opening and hollow layer temperature are adjusted to obtain energy-saving control parameters for each area, including: The Q-value output of the initial deep reinforcement learning model is mapped according to the building space attributes, and the control area is divided into east control area, south control area, west control area and north control area according to the curtain wall orientation. The control area is divided into low control area, middle control area and high control area according to the floor height to obtain the control unit area matrix; A droop control model is established based on the control unit region matrix, a linear mapping relationship is established between the actual energy consumption of each control unit and the standard energy consumption to obtain a droop characteristic coefficient set, and the droop characteristic coefficient set is adaptively adjusted to obtain an adaptive droop coefficient matrix; The shading coefficient is calculated based on the adaptive droop coefficient matrix, and the calculation formula for the shading louver angle αs is established according to the solar altitude angle θs and the azimuth angle φs: αs=arctan(tanθs×cosφs), where θs represents the solar incidence angle and φs represents the solar azimuth angle. The shading control matrix is ​​obtained by combining the indoor illumination demand value. A ventilation control model is established according to the sunshade control matrix, the ventilation volume is calculated based on the indoor and outdoor temperature difference and the wind speed, and a ventilation opening matrix is ​​obtained. The ventilation opening matrix is ​​input into a heat transfer model, and the temperature distribution of the hollow layer is calculated by solving the heat transfer differential equation to obtain a temperature control matrix. The shading control matrix, the ventilation opening matrix, and the temperature control matrix are combined to construct a multi-objective optimization function F=λ1×E1+λ2×E2+λ3×E3, where E1 represents the target energy consumption per unit area, E2 represents the human comfort index value, E3 represents the control stability index value, and λ1, λ2, and λ3 represent balance factors; The multi-objective optimization function is input into the particle swarm optimization algorithm, and the energy-saving control parameters of each area are obtained through iterative optimization calculation.

6. The deep learning-based intelligent curtain wall energy-saving control method according to claim 5 is characterized in that: The multi-objective optimization control is performed according to the energy-saving control parameters of each area, and the initial deep reinforcement learning model is updated online to obtain a target deep reinforcement learning model, including: The energy-saving control parameters of each area are imported into the control execution unit, and the sunshade louver angle adjustment, ventilation opening adjustment and temperature setting value adjustment are performed according to the control area corresponding to the curtain wall orientation and floor height to obtain a regional control execution matrix; Setting a data collection time interval according to the regional control execution matrix, collecting indoor and outdoor temperature data through a temperature sensor, collecting indoor and outdoor light intensity data through a light sensor, and collecting regional energy consumption data through an energy consumption sensor to obtain an environmental parameter set; Normalizing the environmental parameter set to obtain a standardized parameter matrix, and inputting the standardized parameter matrix into the initial deep reinforcement learning model. After state space mapping, action space selection, and reward function calculation, the shading adjustment amount, ventilation adjustment amount, and temperature adjustment amount of each area are output to obtain an optimized control vector; Decoding and mapping the optimized control vector, restoring the normalized control quantity to the actual sunshade louver angle value, ventilation opening value and temperature setting value, and allocating them according to the control area to obtain the execution control instruction; According to the execution control instruction, the actuator is driven to operate, the sunshade louver motor angle, the ventilation window opening and the hollow layer temperature are adjusted, and the execution status data is collected to obtain the system response data; The energy-saving index and the comfort index are calculated based on the system response data, and the initial deep reinforcement learning model is updated online using an incremental learning method to obtain a target deep reinforcement learning model.

7. The deep learning-based intelligent curtain wall energy-saving control method according to claim 6 is characterized in that: The energy saving index and the comfort index are calculated based on the system response data, and the initial deep reinforcement learning model is updated online using an incremental learning method to obtain a target deep reinforcement learning model, including: Perform time series statistics on the energy consumption data in the system response data, calculate the cumulative energy consumption per unit area, peak-to-valley energy consumption ratio, and energy consumption standard deviation of each control area within 24 hours, and obtain an energy-saving performance indicator matrix through weighted average; Evaluate and calculate the indoor environmental parameters in the system response data, calculate the predicted mean voting value (PMV) based on temperature, humidity, airflow velocity, and radiant temperature, and calculate the visual comfort based on illumination uniformity and glare value to obtain a comfort performance index matrix; Constructing a model performance evaluation function based on the energy-saving performance indicator matrix and the comfort performance indicator matrix, and performing a performance evaluation on the initial deep reinforcement learning model; when an output value of the model performance evaluation function is less than a preset threshold, activating a model update mechanism and obtaining an update trigger instruction; Based on the update trigger instruction, the system response data is sampled for 24 consecutive hours to construct model training data including state transition sequence, action sequence and reward sequence; Online feature extraction is performed on the model training data to obtain a dynamic feature vector, and the dynamic feature vector is input into the initial deep reinforcement learning model for incremental training to obtain a target deep reinforcement learning model.

8. An intelligent curtain wall energy-saving control system based on deep learning, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the system comprises: The acquisition module is used to collect temperature data, light data and energy consumption data of each level area of ​​the curtain wall system to obtain a sensor data matrix; A calculation module is used to calculate the curtain wall energy consumption residual value according to the sensor data matrix, and perform time series segmentation and K-means clustering on the curtain wall energy consumption residual value to obtain an energy consumption feature vector and an energy consumption pattern feature database; A training module, configured to set a state space, an action space, and a reward function based on the energy consumption feature vector and the energy consumption pattern feature database, and train an initial deep reinforcement learning model; An adjustment module is used to perform adaptive droop coefficient adjustment on the curtain wall control area based on the initial deep reinforcement learning model, and adjust the shading coefficient, vent opening and hollow layer temperature to obtain energy-saving control parameters for each area; An updating module is used to perform multi-objective optimization control according to the energy-saving control parameters of each area, and to update the initial deep reinforcement learning model online to obtain a target deep reinforcement learning model.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images