Energy-saving double-layer curtain wall heat recovery adjusting method and system

Abstract

The application provides a heat recovery adjusting method and system of an energy-saving double-layer curtain wall, which comprises collecting real-time parameters such as indoor and outdoor air temperature, solar radiation intensity, inner and outer glass surface temperature and cavity pressure difference; calculating a standardized time gradient vector and an environmental fluctuation rate scalar based on a preset time window; inputting the environmental parameters and the time gradient vector into a preset network model to generate a state space vector relative to four types of benchmark states; interpolating an instantaneous weight matrix by using the vector, and calculating the tendency score of each air flow direction mode; when the highest score candidate mode is different from the current mode, combining a dead zone value adjusted according to the environmental fluctuation rate and a minimum confirmation duration to continuously confirm the same candidate mode, and switching the ventilation component after meeting the condition, otherwise, keeping the original mode.

Description

Technical Field

[0001] This application belongs to the field of regulation, and in particular relates to a heat recovery regulation method and system for an energy-saving double-layer curtain wall. Background Technology

[0002] The heat recovery efficiency and building energy-saving effect of double-layer curtain walls are closely related to the airflow organization within the curtain wall cavity and the scheduling strategy of ventilation components. In actual operation, building control systems typically need to collect various environmental parameters in real time, such as indoor air temperature, outdoor air temperature, solar radiation intensity, glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity, to determine the current thermal state of the curtain wall system. As outdoor climate conditions, solar radiation intensity, and indoor heat load demands constantly change, the system needs to adjust the opening and closing status of dampers, louvers, or exhaust equipment in a timely manner to change the airflow direction and heat exchange path within the curtain wall cavity, thereby improving heat utilization efficiency and reducing building operating energy consumption.

[0003] However, in the actual control process of double-layer curtain walls, various environmental parameters often exhibit continuous fluctuations and short-term disturbances. Conventional control methods mostly rely on pre-set fixed thresholds for mode determination. When parameters such as temperature, radiation, or pressure difference are near the switching critical region, short-term fluctuations can easily trigger repeated actions of ventilation components, thereby increasing mechanical wear and ineffective energy consumption, and affecting the stability of system operation. At the same time, existing control methods usually focus more on the parameter magnitude at the current moment, and do not adequately consider the changing trend, fluctuation amplitude, and persistence of environmental conditions over a period of time, making it difficult to accurately distinguish between real operating condition changes and transient disturbances. Therefore, how to integrate the environmental time-series evolution characteristics, continuously identify the thermal state of the curtain wall, and establish an adaptive switching determination mechanism based on the degree of environmental fluctuation to achieve smooth adjustment of airflow direction has become a technical problem that needs to be solved in the energy-saving control of double-layer curtain walls. Summary of the Invention

[0004] To address the problems of existing technologies that rely on fixed thresholds for ventilation mode switching, making it difficult to determine the actual thermal state in conjunction with environmental temporal fluctuations, and easily causing frequent component movements and insufficient adjustment stability, this invention provides a heat recovery adjustment method and system for energy-saving double-layer curtain walls.

[0005] In a first aspect, the present invention provides a heat recovery regulation method for an energy-saving double-layer curtain wall, comprising: The real-time operating parameters of the double-layer curtain wall are obtained, including indoor air temperature, outdoor air temperature, solar radiation intensity, inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity; a time gradient vector is calculated based on the indoor air temperature, outdoor air temperature, and solar radiation intensity within a preset time window, and the time gradient vector is integrated into an environmental volatility scalar. Indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector are input into a preset network model to calculate the continuous membership degree of the current system state to the four reference states, generating a state space vector. The state space vector is used as interpolation coefficients to interpolate the prototype weight matrix representing the reference states, synthesizing an instantaneous weight matrix. The instantaneous weight matrix is ​​used to perform weighted calculations on the inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and time gradient vector to obtain the tendency score of the preset airflow direction pattern. If the candidate mode with the highest score is different from the current mode, determine whether the score difference between the same candidate mode and the current mode is greater than the dead zone value, and whether the same candidate mode continuously meets the switching conditions and maintains the minimum confirmation time; the dead zone value and the minimum confirmation time are determined based on the environmental volatility scalar; if the conditions are met, the curtain wall ventilation component is switched to the same candidate mode, otherwise the current mode is maintained.

[0006] On the other hand, the present invention also provides an energy-saving double-layer curtain wall heat recovery regulation system, comprising the following modules: The calculation module is used to obtain real-time operating parameters of the double-layer curtain wall, including indoor air temperature, outdoor air temperature, solar radiation intensity, inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity; it calculates a time gradient vector based on the indoor air temperature, outdoor air temperature, and solar radiation intensity within a preset time window, and integrates the time gradient vector into an environmental volatility scalar. The weighting module is used to input indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector into a preset network model, calculate the continuous membership degree of the current system state to the four reference states, and generate a state space vector; using the state space vector as interpolation coefficients, interpolation is performed on the prototype weight matrix representing the reference states to synthesize an instantaneous weight matrix; the instantaneous weight matrix is ​​used to perform weighted calculations on the inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and time gradient vector to obtain the tendency score of the preset airflow direction pattern; The determination module is used to determine whether the score difference between the same candidate mode and the current mode is greater than the dead zone value if the candidate mode with the highest score is different from the current mode, and whether the same candidate mode continues to meet the switching conditions and maintains the minimum confirmation time; the dead zone value and the minimum confirmation time are determined based on the environmental volatility scalar; if the conditions are met, the curtain wall ventilation component is switched to the same candidate mode, otherwise the current mode is maintained.

[0007] This invention collects curtain wall operating parameters such as indoor and outdoor temperatures, solar radiation intensity, glass surface temperature, and cavity pressure difference. It then calculates an environmental volatility scalar using a standardized time gradient vector within a preset time window to characterize environmental change trends. A preset network model calculates the continuous membership degree of the current state to each baseline state, generating a state space vector. Based on this, an instantaneous weight matrix is ​​synthesized through interpolation. This matrix is ​​then used to weight key operating parameters, yielding the tendency score for each airflow direction mode. Furthermore, by combining a dead zone value determined based on environmental volatility and a minimum confirmation time, the same candidate mode is continuously confirmed before switching, thereby helping to reduce frequent actions caused by short-term disturbances and improving the stability and energy-saving control effect of double-layer curtain wall heat recovery regulation. Attached Figure Description

[0008] Figure 1 A flowchart illustrating the heat recovery regulation method for energy-saving double-layer curtain walls; Figure 2 This is a schematic diagram showing the relative proportion coefficients of the four baseline states. Detailed Implementation

[0009] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] Example 1 In Embodiment 1 of the present invention, a heat recovery regulation method for an energy-saving double-layer curtain wall is provided, such as... Figure 1 As shown, it includes: S1, Obtain real-time operating parameters of the double-layer curtain wall, including indoor air temperature, outdoor air temperature, solar radiation intensity, inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity.

[0011] Physical data is read by hardware sensors distributed in various areas of the curtain wall.

[0012] The indoor and outdoor air temperatures are collected using PT100 platinum resistance temperature sensors, while the inner and outer glass surface temperatures are collected using patch thermocouples.

[0013] These analog electrical signals are processed into digital signals by an ADS1115 analog-to-digital converter chip and then transmitted to the Raspberry Pi microprocessor via the I²C bus.

[0014] Solar radiation intensity is collected using a photoelectric solar radiation meter, and the pressure difference between the inside and outside of the curtain wall cavity is collected using a digital differential pressure sensor of model SDP810.

[0015] The sensor communication interface reads data from the underlying hardware sensor registers and stores the six collected parameters in the order of sampling timestamps to form a real-time operating parameter sequence. For short-term data loss during real-time communication, the previous valid sample value is retained for compensation.

[0016] S2, calculate the time gradient vector based on the indoor air temperature, outdoor air temperature and solar radiation intensity within a preset time window, and integrate the time gradient vector into an environmental volatility scalar.

[0017] The preset time window size is 15 to 30 minutes.

[0018] Based on the data collected within the preset time window, least squares linear fitting is performed on indoor air temperature, outdoor air temperature and solar radiation intensity respectively, and the fitting slope of each is extracted, or the rate of change is calculated by dividing the difference between the first and last sample values ​​by the length of the time window, so as to obtain the original time gradient vector composed of these three rates of change.

[0019] The original time gradient vector is standardized according to a preset rate of change scale corresponding to each component to obtain a dimensionless standardized time gradient vector. The square root of the sum of squares of each element in the standardized time gradient vector is calculated. This square root result is converted into a floating-point scalar value and used as the environmental volatility scalar. Unless otherwise specified, the time gradient vectors subsequently used in network model input and matrix weighting operations refer to the standardized time gradient vector.

[0020] In some embodiments, the step of calculating a time gradient vector based on indoor air temperature, outdoor air temperature, and solar radiation intensity within a preset time window, and integrating the time gradient vector into an environmental volatility scalar, includes: Acquire continuous sampled values ​​of indoor air temperature, outdoor air temperature, and solar radiation intensity at multiple sampling times within a preset time window; Calculate the rate of change of each parameter within a preset time window to form a multidimensional original time gradient vector; The original time gradient vector is standardized and its magnitude is calculated, which is then used as an environmental volatility scalar representing the overall rate of change of the system.

[0021] Set the data sampling frequency, for example, record data once every 1 minute, and set a preset time window of 15 to 30 minutes, preferably 15 minutes. Continuously collect continuous sampling values ​​of indoor air temperature, outdoor air temperature, and solar radiation intensity parameters within the time window.

[0022] The rate of change of each parameter is calculated by performing linear fitting using the least squares method or by dividing the difference between the first and last values ​​by the time span. For example, the rate of change of indoor temperature is found to be 0.1 degrees Celsius per minute, with a preferred reference range of -0.5 to 0.5 degrees Celsius per minute; the rate of change of outdoor temperature is -0.2 degrees Celsius per minute, with a preferred reference range of -1.0 to 1.0 degrees Celsius per minute; and the rate of change of solar radiation is 15 watts per square meter per minute, with a preferred reference range of -50 to 50 watts per square meter per minute. These three sets of values ​​together constitute the original three-dimensional time gradient vector, with components of 0.1, -0.2, and 15, respectively.

[0023] The vector is standardized by dividing it by a preset maximum rate of change threshold, such as an indoor temperature threshold of 1.0, an outdoor temperature threshold of 2.0, and a solar radiation threshold of 100, to obtain a dimensionless standardized time gradient vector with components of 0.1, -0.1, and 0.15, respectively.

[0024] The Euclidean norm of the standardized time gradient vector is calculated by taking the square root of the sum of squares, resulting in 0.206. This calculated value of 0.206 is used as a scalar measure of environmental volatility, which characterizes the combined variation of indoor and outdoor temperature and solar radiation within a time window.

[0025] S3 inputs indoor air temperature, outdoor air temperature, solar radiation intensity and time gradient vector into the preset network model, calculates the continuous membership degree of the current system state to the four reference states, and generates the state space vector.

[0026] The preset network model adopts a radial basis function network, and the four baseline states defined include winter heating demand state, winter overheating risk state, summer cooling demand state, and transitional season operation state.

[0027] The acquired indoor air temperature, outdoor air temperature, and solar radiation intensity are standardized according to a preset scale and then concatenated with a three-dimensional standardized time gradient vector to form a six-dimensional input feature vector.

[0028] A preset radial basis function network is invoked to calculate the distance between the six-dimensional input feature vector and the Gaussian function center points corresponding to the four baseline states, and the distance is converted into an initial membership value between 0 and 1 through a Gaussian distribution model.

[0029] The four initial membership values ​​are normalized so that the sum of the four normalized membership values ​​is 1. The four normalized membership values ​​are then arranged in the order of the preset baseline states to generate a one-dimensional state space vector with a length of four.

[0030] In some embodiments, the step of inputting indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector into a preset network model to calculate the continuous membership degree of the current system state to the four baseline states includes: Indoor air temperature, outdoor air temperature, and solar radiation intensity are standardized according to preset scales and then concatenated with a three-dimensional standardized time gradient vector to form a six-dimensional input feature vector. The preset radial basis function network is invoked to calculate the distance between the input feature vector and the center point of the Gaussian function corresponding to the four baseline states: winter heating demand, winter overheating risk, summer cooling demand, and transitional season operation status. The distance is transformed into a membership value between 0 and 1 using a Gaussian distribution model, and then normalized to generate a continuous state space vector consisting of four membership values.

[0031] The preset network model is a radial basis function network (RBF network), which includes an input layer, hidden layers, and an output layer. The input layer receives a six-dimensional standardized input feature vector, which is composed of standardized indoor air temperature, standardized outdoor air temperature, standardized solar radiation intensity, and a three-dimensional standardized time gradient vector.

[0032] The hidden layer consists of several radial basis function nodes, each corresponding to a Gaussian function center point, representing different baseline states. The coordinates of the center point have the same dimension as the input feature vector. The hidden layer calculates the Euclidean distance between the input feature vector and each center point, and uses a Gaussian activation function for nonlinear mapping. The formula for the Gaussian activation function is: ,in σ is the Euclidean distance in the standardized feature space, and σ is a broadening parameter that matches the scale of the standardized feature space.

[0033] The output layer receives the initial membership degrees of each reference state from the hidden layer and outputs four initial membership degree values ​​corresponding to the four reference states. Real-time sensor data is extracted, such as indoor temperature 22 degrees Celsius, outdoor temperature 5 degrees Celsius, solar radiation intensity 800 watts per square meter, and the calculated three-dimensional normalized time gradient vector, with components of 0.1, -0.1, and 0.15 respectively.

[0034] Using a maximum-minimum normalization method, indoor air temperature, outdoor air temperature, and solar radiation intensity are mapped to a dimensionless interval of 0 to 1. For example, 22 degrees Celsius is mapped to 0.64, 5 degrees Celsius to 0.30, and 800 watts per square meter to 0.80. The time gradient vector is normalized into dimensionless rate-of-change components that can be positive or negative according to the maximum rate of change threshold corresponding to each component. The processed scalar and gradient vector are concatenated in a fixed order to form a six-dimensional input feature vector, with components of 0.64, 0.30, 0.80, 0.1, -0.1, and 0.15, respectively.

[0035] The radial basis function network preloaded in memory is invoked. This network has four baseline state center point coordinates preset offline; for example, the center point coordinates representing the risk of overheating in winter might be 0.6, 0.1, 0.9, 0.5, 0.5, and 0.5. The Euclidean distances between the input feature vector and each center point are calculated, assuming the calculated distances are 0.8, 0.3, 1.5, and 1.1, respectively.

[0036] Substituting the aforementioned distances into the set activation function, with a broadening parameter of 1.0, the initial membership degrees are calculated to be 0.726, 0.956, 0.324, and 0.546, respectively. These four values, ranging from 0 to 1, quantify the similarity between the current actual curtain wall thermal state and the four typical baseline states.

[0037] In some embodiments, generating the state space vector includes: The total global membership value of the system is obtained by summing the membership values ​​of the four baseline states output. Divide the membership value of each baseline state by the total global membership value of the system to obtain the proportion coefficients of each item; The proportion coefficients are arranged in a fixed order according to a preset baseline state to generate a four-dimensional continuous state space vector.

[0038] The system receives four initial membership values ​​from the radial basis function network output, for example, 0.726, 0.956, 0.324, and 0.546. These four values ​​are linearly summed to calculate the total global membership value of the system, S = 2.552. This total value represents the comprehensive result of projecting the current environmental state onto each reference state in the multidimensional feature space.

[0039] Normalization is performed by dividing each initial membership value by the calculated total global membership value S, thus obtaining their respective relative proportion coefficients. Specifically, this is calculated as the proportion coefficient. ≈0.284, percentage coefficient ≈0.375, percentage coefficient ≈0.127, percentage coefficient ≈0.214.

[0040] The four calculated proportion coefficients are combined according to a fixed baseline order, such as winter heating, winter overheating, summer cooling, and transitional seasons, to generate a four-dimensional continuous state space vector with components of 0.284, 0.375, 0.127, and 0.214, respectively. The sum of all dimensions of this vector equals 1. The relative proportion coefficients of the four baseline states obtained from the above example are as follows: Figure 2 As shown.

[0041] During the offline calibration phase, historical operating condition samples are extracted in advance from the building energy management system, curtain wall sensor records, and ventilation component control logs. Each historical operating condition sample includes indoor air temperature, outdoor air temperature, solar radiation intensity, inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, the actual operating mode of the ventilation component at the corresponding time, fan energy consumption or air conditioning load, and the number of ventilation component switching times within the corresponding statistical period. The fan energy consumption or air conditioning load and the number of ventilation component switching times are normalized to obtain the energy consumption index E and the switching cost index A, and the comprehensive cost is calculated according to J=λ1E+λ2A, where λ1 and λ2 are preset non-negative weights and their sum is 1.

[0042] Based on indoor air temperature, outdoor air temperature, solar radiation intensity, and standardized temporal gradient vectors calculated using the same time window from historical operating condition samples, a first six-dimensional standardized calibration feature for state division is constructed. This first six-dimensional standardized calibration feature is then clustered to obtain the coordinates of the center points corresponding to four baseline states: winter heating demand state, winter overheating risk state, summer cooling demand state, and transitional season operating state. These center point coordinates are used for subsequent membership degree calculations of the radial basis function network.

[0043] Based on the inner and outer glass surface temperatures, the pressure difference between the inside and outside of the curtain wall cavity, and the standardized time gradient vector from the same historical operating condition sample, a second six-dimensional standardized calibration feature for matrix calibration is constructed according to the feature order and standardization scale required for the weighted operation of the prototype weight matrix. For each baseline state, the historical operating condition samples belonging to that baseline state are used as a calibration sample group. Within this calibration sample group, the mode cost is determined based on the comprehensive cost of the samples corresponding to different airflow direction modes. After inverse normalization of each mode cost, it is mapped to the target score range of 0 to 10, so that airflow direction modes with lower comprehensive costs have higher target scores. Using the second six-dimensional standardized calibration feature as input and the target scores corresponding to the internal circulation mode, external circulation mode, and exhaust heat recovery mode as output, the linear mapping relationship between the two is solved using least squares regression to obtain the 3x6 prototype weight matrix corresponding to the baseline state.

[0044] S4. Using the state space vector as interpolation coefficients, interpolation is performed on the prototype weight matrix representing the baseline state to synthesize the instantaneous weight matrix.

[0045] The prototype weight matrix is ​​a fixed-coefficient matrix obtained through the aforementioned offline calibration phase. The four prototype weight matrices correspond to the winter heating demand state, the winter overheating risk state, the summer cooling demand state, and the transitional season operation state, respectively. The four independent components of the aforementioned four-dimensional state space vector are extracted and multiplied by the four prototype weight matrices pre-stored in the microprocessor memory. The new matrices obtained from the four multiplication operations are then summed. The resulting single matrix is ​​the synthesized instantaneous weight matrix.

[0046] In some embodiments, the step of using the state space vector as interpolation coefficients to perform interpolation operations on the prototype weight matrix representing the baseline state to synthesize the instantaneous weight matrix includes: Extract four prototype weight matrices corresponding to the four baseline states, and the dimensions of the prototype weight matrices are matched with the dimensions of the input vector for the subsequent weighting operation. Using the four dimensions of the continuous state space vector as corresponding scaling factors, scalar multiplication is performed on the four prototype weight matrices to obtain four weighted transition matrices. The elements at the same coordinate positions in the four transition matrices are summed by matrix addition to synthesize a unique instantaneous weight matrix.

[0047] Four pre-defined prototype weight matrices are extracted from the memory, each corresponding to one of four baseline states. For example, if the system has three candidate airflow modes—internal circulation, external circulation, and exhaust heat recovery—and the evaluation requires six feature dimensions—inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and a three-dimensional normalized time gradient vector—then each extracted prototype weight matrix will have a fixed dimension of 3 rows and 6 columns.

[0048] The value range of each weight element within the matrix is ​​calibrated according to the scale of the input features, preferably between -5.0 and 5.0 after the input features are standardized. The four-dimensional continuous state space vector obtained from the previous calculation is used, and its internal values ​​are used as corresponding scalar scaling factors. Scalar multiplication is then performed sequentially on the four prototype weight matrices: the first prototype weight matrix is ​​multiplied by 0.284, the second by 0.375, the third by 0.127, and the fourth by 0.214. This generates four identical 3x6 transition matrices.

[0049] The four transition matrices are summed by matrix addition at corresponding element positions, thereby fusing the characteristic control logic of each baseline state to synthesize a 3x6 instantaneous weight matrix. This interpolation mechanism helps to achieve smooth reconstruction of the weight matrix across operating conditions and reduces control step abrupt changes based on lookup table methods.

[0050] S5 uses the instantaneous weight matrix to perform weighted calculations on the inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and time gradient vector to obtain the tendency score of the preset airflow direction pattern.

[0051] The preset airflow direction modes include internal circulation mode, external circulation mode, and exhaust heat recovery mode.

[0052] The system pre-stores a mode execution mapping table, which records the control status of air vents, louvers, dampers, and exhaust fans corresponding to each airflow direction mode. The internal circulation mode corresponds to the opening of the indoor return air vent and the cavity return air vent, and the closing of the outdoor air inlet and the outdoor air outlet, allowing the cavity air to primarily circulate and exchange heat with the indoor side. The external circulation mode corresponds to the opening of the outdoor air inlet and the outdoor air outlet, and the closing of the indoor connecting vent, allowing the curtain wall cavity to ventilate and exchange heat with the outdoor air. The exhaust heat recovery mode corresponds to the opening of the indoor exhaust air inlet, the heat recovery channel damper, and the cavity exhaust fan, while closing or partially closing the outdoor direct bypass air vent, allowing indoor exhaust air to be discharged through the curtain wall cavity or the heat recovery channel and exchange heat with the cavity. The above mapping relationships are pre-written into the controller storage unit according to the actual curtain wall air vent numbers.

[0053] The inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and the three-dimensional standardized time gradient vector are standardized according to the same scale as in the offline training phase and then spliced ​​into a six-dimensional state feature vector.

[0054] Perform a matrix dot product operation between the synthesized instantaneous weight matrix and the six-dimensional state feature vector, and output a result vector containing three numerical elements. The first to third values ​​in the result vector correspond to the tendency scores of the inner circulation mode, outer circulation mode, and exhaust heat recovery mode, respectively.

[0055] In some embodiments, the step of using an instantaneous weighting matrix to weight the inner glass surface temperature, the outer glass surface temperature, the pressure difference between the inside and outside of the curtain wall cavity, and the time gradient vector to obtain a tendency score for a preset airflow direction pattern includes: The obtained inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity are processed with the standardized time gradient vector according to the same scale as in the offline training stage and then spliced ​​together to construct a real-time state column vector. The synthesized instantaneous weight matrix is ​​multiplied by the real-time state column vector to output a result column vector containing the scores corresponding to each preset airflow direction mode. The specific values ​​of each preset airflow direction pattern are extracted from the result column vector and used as the final tendency score of the corresponding pattern.

[0056] Read real-time physical parameters from front-end sensors, such as the inner glass surface temperature of 18.5 degrees Celsius, the outer glass surface temperature of 8.2 degrees Celsius, and the pressure difference between the inside and outside of the curtain wall cavity of 12.5 Pascals, with a typical fluctuation range of -50 to 50 Pascals.

[0057] These three steady-state parameters are standardized according to the temperature and pressure difference scales consistent with the offline training phase, and then vertically concatenated with the three-dimensional standardized time gradient vector output by the previous algorithm according to the predefined data frame format to construct a real-time state column vector with a dimension of 6 rows and 1 column. For example, the standardized components are 0.46, 0.40, 0.63, 0.1, -0.1 and 0.15 respectively.

[0058] The system calls upon the 3x6 instantaneous weight matrix generated by four-state interpolation, multiplies the instantaneous weight matrix with the real-time state column vector to obtain the result column vector. This outputs a result column vector with dimensions of 3x1, for example, with result components of 7.54, 6.21, and 8.83 respectively.

[0059] The values ​​are extracted according to the row number index. The score of 7.54 in the first row is assigned to the inner loop mode preference score, the score of 6.21 in the second row is assigned to the outer loop mode score, and the score of 8.83 in the third row is assigned to the exhaust heat recovery mode score. This algorithm can convert multi-path associated environmental parameters into a quantitative result of the priority of specific execution modes.

[0060] S6. If the candidate mode with the highest score is different from the current mode, determine whether the score difference between the same candidate mode and the current mode is greater than the dead zone value, and whether the same candidate mode continues to meet the switching conditions and maintains the minimum confirmation time; the dead zone value and the minimum confirmation time are determined based on the environmental volatility scalar.

[0061] Find the array index corresponding to the maximum value in the tendency score result vector, and determine the candidate pattern corresponding to the highest score based on the index.

[0062] When the system identifies that the candidate pattern is inconsistent with the current operating mode of the system recorded in memory, it records the candidate pattern as a candidate pattern to be confirmed and calculates the absolute value of the difference between the score of the candidate pattern to be confirmed and the score of the current mode.

[0063] Simultaneously, a constant base dead zone value and a constant base confirmation duration are defined within the microprocessor program. A dead zone adjustment factor is calculated based on the environmental volatility scalar, and then multiplied by the base dead zone value to obtain the dead zone value. A duration adjustment factor is calculated based on the environmental volatility scalar, and then multiplied by the base confirmation duration to obtain the minimum confirmation duration.

[0064] To prevent the calculated results from exceeding the allowable range when the environmental volatility scalar is large, the controller performs amplitude limiting processing on the calculated dead zone value and minimum confirmation time, so that the dead zone value is kept within the preset dead zone range and the minimum confirmation time is kept within the preset confirmation time range.

[0065] The system determines whether the candidate mode to be confirmed remains the highest-scoring candidate mode, and whether the absolute value of the score difference between the candidate mode to be confirmed and the current mode is greater than the dead zone value calculated at the current time. If both conditions are met, the system clock records the timestamp, accumulates the number of seconds that the same candidate mode to be confirmed continuously meets the above conditions, and determines whether the number of seconds is greater than or equal to the calculated minimum confirmation duration. If the highest-scoring candidate mode changes, or the absolute value of the score difference is not greater than the dead zone value, the internal timer is cleared, and confirmation is re-performed based on the newly determined highest-scoring candidate mode.

[0066] In some embodiments, determining whether the score difference between the same candidate mode and the current mode is greater than the dead zone value, and whether the same candidate mode continuously meets the switching conditions and maintains the minimum confirmation time, includes: Calculate the difference between the score of the candidate mode with the highest score and the score of the current running mode, and record the candidate mode with the highest score as a candidate mode to be confirmed; Real-time monitoring is conducted to ensure that the candidate mode to be confirmed remains the candidate mode with the highest score, and to determine whether the difference between the candidate mode to be confirmed and the current operating mode is greater than the dead zone value configured in the current system. When the candidate mode to be confirmed remains the candidate mode with the highest score, and the difference between the candidate mode to be confirmed and the current running mode is greater than the dead zone value, the internal timer is started to accumulate the time, and it is determined whether the accumulated time has reached the currently set minimum confirmation time. If the cumulative duration reaches the minimum confirmation duration, and during this period the candidate mode to be confirmed remains the candidate mode with the highest score, and the difference between the candidate mode to be confirmed and the current operating mode is always greater than the dead zone value, then a mode switching confirmation signal is generated. If the candidate mode with the highest score changes during the cumulative timeout period, or if the difference between the candidate mode to be confirmed and the current operating mode is not greater than the dead zone value, the internal timer is cleared and the confirmation is performed again based on the newly determined candidate mode with the highest score.

[0067] After obtaining the tendency scores for all modes, the target mode with the highest current score is identified. Assume the current physical state of the curtain wall remains in the internal circulation mode with a score of 7.54, while the candidate mode with the highest score obtained through matrix operations is the exhaust heat recovery mode with a score of 8.83. Subtraction is performed, and the absolute difference between the two is ΔS = 1.29.

[0068] The difference is compared in real time with the dead zone value dynamically mapped from environmental parameters at the system's underlying level, for example, the currently calculated value is 0.872. When the difference of 1.29 is detected to be greater than the dead zone value of 0.872, and the exhaust heat recovery mode remains the candidate mode with the highest score, the internal timer is triggered to start accumulating the count from 0. At the same time, the currently issued minimum confirmation duration is loaded as the confirmation duration threshold, for example, set to 85.4 seconds, with a normal range between 60 and 300 seconds.

[0069] During the timing period, the difference is continuously calculated at a sampling frequency of, for example, 2Hz. If, within this continuous 85.4 seconds, the exhaust heat recovery mode consistently remains the candidate mode with the highest score, and its score difference relative to the current internal circulation mode remains above 0.872, without any candidate mode jump or reversal below the dead zone value, then the environmental change trend is deemed to meet the mode switching confirmation condition. At this point, the timer reaches the threshold, triggering the interrupt program to send a mode switching confirmation signal to the execution layer, driving the electric damper actuator to switch the airflow direction.

[0070] In some embodiments, the dead zone value and minimum confirmation time are determined based on an environmental volatility scalar, including: Extract the system's underlying preset basic dead zone value and basic confirmation time; Obtain the real-time calculated environmental volatility scalar, and calculate a dead zone adjustment factor of not less than 1 and a duration adjustment factor of not less than 1 respectively according to the preset mapping function; The dead zone adjustment factor is multiplied by the base dead zone value to obtain the dead zone value, and the duration adjustment factor is multiplied by the base confirmation duration to obtain the minimum confirmation duration. Limiting processing is then applied to the dead zone value and the minimum confirmation duration respectively.

[0071] Read the engineering-calibrated basic control parameters from non-volatile memory, such as setting the basic dead zone value to 0.5 and the basic confirmation time to 60 seconds. Obtain the environmental volatility scalar R obtained from the previous time gradient analysis, for example, the current value is 0.206. This dimensionless value is used to characterize the degree of short-term meteorological changes caused by changes in indoor and outdoor temperature and solar radiation.

[0072] Based on this, a preset nonlinear adaptive mapping function is invoked, such as using a dead-zone adjustment factor function. and duration adjustment factor function Take the engineering experience coefficient. Equal to 4.0 and Substituting 10.0 into the calculation yields the dead zone adjustment factor. =1.744, and duration adjustment factor =1.424, the two-function structure ensures that the lower limit of the factor value is not less than 1.0.

[0073] After calculating the two factors mentioned above, a multiplication operation is performed, and a limiting function is used to restrict the calculation result to a preset normal range, resulting in the latest judgment thresholds: a dead zone value of 0.872 and a minimum confirmation time of 85.4 seconds. When the environmental volatility scalar continues to increase and the calculated value exceeds the upper limit of the preset range, the dead zone value is limited to 1.5 minutes, and the minimum confirmation time is limited to 300 seconds.

[0074] By using a parameter amplification mechanism, when indoor and outdoor temperatures or solar radiation intensity fluctuate significantly in the short term, the threshold for mode switching is raised, which helps to filter out invalid short-term interference signals. During periods of stable climate, various parameters automatically fall back to near their baseline values, which helps to improve the responsiveness to changes in thermal efficiency.

[0075] S7. If the condition is met, switch the curtain wall ventilation component to the same candidate mode; otherwise, maintain the current mode.

[0076] When the same candidate mode to be confirmed remains the highest-scoring mode, and the absolute value of the score difference between it and the current mode remains greater than the dead zone value, and the number of seconds is greater than or equal to the minimum confirmation time, the controller generates the corresponding mode switching control signal.

[0077] The controller reads a preset mode execution mapping table to determine the target states of each execution component corresponding to the same candidate mode that has been continuously confirmed. These execution components include indoor air vents, outdoor air vents, ventilation louvers, dampers, and cavity exhaust fans. For air vents, louvers, and dampers that need to be opened or closed, the controller converts the corresponding target state into a switching control signal and drives the corresponding component to switch to the open or closed state via a relay module or electric actuator. For cavity exhaust fans that require speed adjustment, the controller generates a speed control command based on the preset speed level or target speed value corresponding to the target mode and sends it to the fan speed controller, causing the cavity exhaust fan to operate at the target speed. This ensures that the actual ventilation path of the double-layer curtain wall is consistent with the structural configuration corresponding to the same candidate mode in the mode execution mapping table.

[0078] If any of the aforementioned conditions are not met, the controller will not send a mode switching control signal and will maintain the current open / closed state, angle state, and speed state of each actuator.

[0079] Example 2 Embodiment 2 of the present invention proposes an energy-saving double-layer curtain wall heat recovery regulation system, comprising the following modules: The calculation module is used to obtain real-time operating parameters of the double-layer curtain wall, including indoor air temperature, outdoor air temperature, solar radiation intensity, inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity; it calculates a time gradient vector based on the indoor air temperature, outdoor air temperature, and solar radiation intensity within a preset time window, and integrates the time gradient vector into an environmental volatility scalar. The weighting module is used to input indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector into a preset network model, calculate the continuous membership degree of the current system state to the four reference states, and generate a state space vector; using the state space vector as interpolation coefficients, interpolation is performed on the prototype weight matrix representing the reference states to synthesize an instantaneous weight matrix; the instantaneous weight matrix is ​​used to perform weighted calculations on the inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and time gradient vector to obtain the tendency score of the preset airflow direction pattern; The determination module is used to determine whether the score difference between the same candidate mode and the current mode is greater than the dead zone value if the candidate mode with the highest score is different from the current mode, and whether the same candidate mode continues to meet the switching conditions and maintains the minimum confirmation time; the dead zone value and the minimum confirmation time are determined based on the environmental volatility scalar; if the conditions are met, the curtain wall ventilation component is switched to the same candidate mode, otherwise the current mode is maintained.

[0080] In some embodiments, the step of calculating a time gradient vector based on indoor air temperature, outdoor air temperature, and solar radiation intensity within a preset time window, and integrating the time gradient vector into an environmental volatility scalar, includes: Acquire continuous sampled values ​​of indoor air temperature, outdoor air temperature, and solar radiation intensity at multiple sampling times within a preset time window; Calculate the rate of change of each parameter within a preset time window to form a multidimensional original time gradient vector; The original time gradient vector is standardized and its magnitude is calculated, which is then used as an environmental volatility scalar representing the overall rate of change of the system.

[0081] In some embodiments, the step of inputting indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector into a preset network model to calculate the continuous membership degree of the current system state to the four baseline states includes: Indoor air temperature, outdoor air temperature, and solar radiation intensity are standardized according to preset scales and then concatenated with a three-dimensional standardized time gradient vector to form a six-dimensional input feature vector. The preset radial basis function network is invoked to calculate the distance between the input feature vector and the center point of the Gaussian function corresponding to the four baseline states: winter heating demand, winter overheating risk, summer cooling demand, and transitional season operation status. The distance is transformed into a membership value between 0 and 1 using a Gaussian distribution model, and then normalized to generate a continuous state space vector consisting of four membership values.

[0082] In some embodiments, generating the state space vector includes: The total global membership value of the system is obtained by summing the membership values ​​of the four baseline states output. Divide the membership value of each baseline state by the total global membership value of the system to obtain the proportion coefficients of each item; The proportion coefficients are arranged in a fixed order according to a preset baseline state to generate a four-dimensional continuous state space vector.

[0083] In some embodiments, the step of using the state space vector as interpolation coefficients to perform interpolation operations on the prototype weight matrix representing the baseline state to synthesize the instantaneous weight matrix includes: Extract four prototype weight matrices corresponding to the four baseline states, and the dimensions of the prototype weight matrices are matched with the dimensions of the input vector for the subsequent weighting operation. Using the four dimensions of the continuous state space vector as corresponding scaling factors, scalar multiplication is performed on the four prototype weight matrices to obtain four weighted transition matrices. The elements at the same coordinate positions in the four transition matrices are summed by matrix addition to synthesize a unique instantaneous weight matrix.

[0084] In some embodiments, the step of using an instantaneous weighting matrix to weight the inner glass surface temperature, the outer glass surface temperature, the pressure difference between the inside and outside of the curtain wall cavity, and the time gradient vector to obtain a tendency score for a preset airflow direction pattern includes: The obtained inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity are processed with the standardized time gradient vector according to the same scale as in the offline training stage and then spliced ​​together to construct a real-time state column vector. The synthesized instantaneous weight matrix is ​​multiplied by the real-time state column vector to output a result column vector containing the scores corresponding to each preset airflow direction mode. The specific values ​​of each preset airflow direction pattern are extracted from the result column vector and used as the final tendency score of the corresponding pattern.

[0085] In some embodiments, determining whether the score difference between the same candidate mode and the current mode is greater than the dead zone value, and whether the same candidate mode continuously meets the switching conditions and maintains the minimum confirmation time, includes: Calculate the difference between the score of the candidate mode with the highest score and the score of the current running mode, and record the candidate mode with the highest score as a candidate mode to be confirmed; Real-time monitoring is conducted to ensure that the candidate mode to be confirmed remains the candidate mode with the highest score, and to determine whether the difference between the candidate mode to be confirmed and the current operating mode is greater than the dead zone value configured in the current system. When the candidate mode to be confirmed remains the candidate mode with the highest score, and the difference between the candidate mode to be confirmed and the current running mode is greater than the dead zone value, the internal timer is started to accumulate the time, and it is determined whether the accumulated time has reached the currently set minimum confirmation time. If the cumulative duration reaches the minimum confirmation duration, and during this period the candidate mode to be confirmed remains the candidate mode with the highest score, and the difference between the candidate mode to be confirmed and the current operating mode is always greater than the dead zone value, then a mode switching confirmation signal is generated. If the candidate mode with the highest score changes during the cumulative timeout period, or if the difference between the candidate mode to be confirmed and the current operating mode is not greater than the dead zone value, the internal timer is cleared and the confirmation is performed again based on the newly determined candidate mode with the highest score.

[0086] In some embodiments, the dead zone value and minimum confirmation time are determined based on an environmental volatility scalar, including: Extract the system's underlying preset basic dead zone value and basic confirmation time; Obtain the real-time calculated environmental volatility scalar, and calculate a dead zone adjustment factor of not less than 1 and a duration adjustment factor of not less than 1 respectively according to the preset mapping function; The dead zone adjustment factor is multiplied by the base dead zone value to obtain the dead zone value, and the duration adjustment factor is multiplied by the base confirmation duration to obtain the minimum confirmation duration. Limiting processing is then applied to the dead zone value and the minimum confirmation duration respectively.

[0087] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A heat recovery regulation method for an energy-saving double-layer curtain wall, characterized in that, include: The real-time operating parameters of the double-layer curtain wall are obtained, including indoor air temperature, outdoor air temperature, solar radiation intensity, inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity. The time gradient vector is calculated based on the indoor air temperature, outdoor air temperature and solar radiation intensity within a preset time window, and the time gradient vector is integrated into an environmental volatility scalar. Indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector are input into a preset network model to calculate the continuous membership degree of the current system state to the four reference states, generating a state space vector. The state space vector is used as interpolation coefficients to interpolate the prototype weight matrix representing the reference states, synthesizing an instantaneous weight matrix. The instantaneous weight matrix is ​​used to perform weighted calculations on the inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and time gradient vector to obtain the tendency score of the preset airflow direction pattern. If the candidate mode with the highest score is different from the current mode, determine whether the score difference between the same candidate mode and the current mode is greater than the dead zone value, and whether the same candidate mode continues to meet the switching conditions and maintains the minimum confirmation time. The dead zone value and minimum confirmation time are determined based on the environmental volatility scalar; if the conditions are met, the curtain wall ventilation components are switched to the same candidate mode; otherwise, the current mode is maintained.

2. The method according to claim 1, characterized in that, The calculation of the time gradient vector based on indoor air temperature, outdoor air temperature, and solar radiation intensity within a preset time window, and the integration of the time gradient vector into an environmental volatility scalar, includes: Acquire continuous sampled values ​​of indoor air temperature, outdoor air temperature, and solar radiation intensity at multiple sampling times within a preset time window; Calculate the rate of change of each parameter within a preset time window to form a multidimensional original time gradient vector; The original time gradient vector is standardized and its magnitude is calculated, which is then used as an environmental volatility scalar representing the overall rate of change of the system.

3. The method according to claim 1, characterized in that, The process involves inputting indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector into a preset network model to calculate the continuous membership degree of the current system state to the four baseline states, including: Indoor air temperature, outdoor air temperature, and solar radiation intensity are standardized according to preset scales and then concatenated with a three-dimensional standardized time gradient vector to form a six-dimensional input feature vector. The preset radial basis function network is invoked to calculate the distance between the input feature vector and the center point of the Gaussian function corresponding to the four baseline states: winter heating demand, winter overheating risk, summer cooling demand, and transitional season operation status. The distance is transformed into a membership value between 0 and 1 using a Gaussian distribution model, and then normalized to generate a continuous state space vector consisting of four membership values.

4. The method according to claim 1, characterized in that, The generation of the state space vector includes: The total global membership value of the system is obtained by summing the membership values ​​of the four baseline states output. Divide the membership value of each baseline state by the total global membership value of the system to obtain the proportion coefficients of each item; The proportion coefficients are arranged in a fixed order according to a preset baseline state to generate a four-dimensional continuous state space vector.

5. The method according to claim 4, characterized in that, The step of using the state space vector as interpolation coefficients to perform interpolation operations on the prototype weight matrix representing the baseline state to synthesize the instantaneous weight matrix includes: Extract four prototype weight matrices corresponding to the four baseline states, and the dimensions of the prototype weight matrices are matched with the dimensions of the input vector for the subsequent weighting operation. Using the four dimensions of the continuous state space vector as corresponding scaling factors, scalar multiplication is performed on the four prototype weight matrices to obtain four weighted transition matrices. The elements at the same coordinate positions in the four transition matrices are summed by matrix addition to synthesize a unique instantaneous weight matrix.

6. The method according to claim 1, characterized in that, The method utilizes an instantaneous weighting matrix to perform weighted calculations on the inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and time gradient vector to obtain a tendency score for a preset airflow direction pattern, including: The obtained inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity are processed with the standardized time gradient vector according to the same scale as in the offline training stage and then spliced ​​together to construct a real-time state column vector. The synthesized instantaneous weight matrix is ​​multiplied by the real-time state column vector to output a result column vector containing the scores corresponding to each preset airflow direction mode. The specific values ​​of each preset airflow direction pattern are extracted from the result column vector and used as the final tendency score of the corresponding pattern.

7. The method according to claim 1, characterized in that, The determination of whether the score difference between the same candidate mode and the current mode is greater than the dead zone value, and whether the same candidate mode continuously meets the switching conditions and maintains the minimum confirmation time, includes: Calculate the difference between the score of the candidate mode with the highest score and the score of the current running mode, and record the candidate mode with the highest score as a candidate mode to be confirmed; Real-time monitoring is conducted to ensure that the candidate mode to be confirmed remains the candidate mode with the highest score, and to determine whether the difference between the candidate mode to be confirmed and the current operating mode is greater than the dead zone value configured in the current system. When the candidate mode to be confirmed remains the candidate mode with the highest score, and the difference between the candidate mode to be confirmed and the current running mode is greater than the dead zone value, the internal timer is started to accumulate the time, and it is determined whether the accumulated time has reached the currently set minimum confirmation time. If the cumulative duration reaches the minimum confirmation duration, and during this period the candidate mode to be confirmed remains the candidate mode with the highest score, and the difference between the candidate mode to be confirmed and the current operating mode is always greater than the dead zone value, then a mode switching confirmation signal is generated. If the candidate mode with the highest score changes during the cumulative timeout period, or if the difference between the candidate mode to be confirmed and the current operating mode is not greater than the dead zone value, the internal timer is cleared and the confirmation is performed again based on the newly determined candidate mode with the highest score.

8. The method according to claim 2, characterized in that, The dead zone value and minimum confirmation time are determined based on an environmental volatility scalar, including: Extract the system's underlying preset basic dead zone value and basic confirmation time; Obtain the real-time calculated environmental volatility scalar, and calculate a dead zone adjustment factor of not less than 1 and a duration adjustment factor of not less than 1 respectively according to the preset mapping function; The dead zone adjustment factor is multiplied by the base dead zone value to obtain the dead zone value, and the duration adjustment factor is multiplied by the base confirmation duration to obtain the minimum confirmation duration. Limiting processing is then applied to the dead zone value and the minimum confirmation duration respectively.

9. A heat recovery and regulation system for an energy-saving double-layer curtain wall, characterized in that, include: The calculation module is used to obtain the real-time operating parameters of the double-layer curtain wall, which include indoor air temperature, outdoor air temperature, solar radiation intensity, inner glass surface temperature, outer glass surface temperature, and pressure difference between the inside and outside of the curtain wall cavity. The time gradient vector is calculated based on the indoor air temperature, outdoor air temperature and solar radiation intensity within a preset time window, and the time gradient vector is integrated into an environmental volatility scalar. The weighting module is used to input indoor air temperature, outdoor air temperature, solar radiation intensity, and time gradient vector into a preset network model, calculate the continuous membership degree of the current system state to the four reference states, and generate a state space vector; using the state space vector as interpolation coefficients, interpolation is performed on the prototype weight matrix representing the reference states to synthesize an instantaneous weight matrix; the instantaneous weight matrix is ​​used to perform weighted calculations on the inner glass surface temperature, outer glass surface temperature, pressure difference between the inside and outside of the curtain wall cavity, and time gradient vector to obtain the tendency score of the preset airflow direction pattern; The determination module is used to determine whether the score difference between the same candidate mode and the current mode is greater than the dead zone value if the candidate mode with the highest score is different from the current mode, and whether the same candidate mode continues to meet the switching conditions and maintains the minimum confirmation time. The dead zone value and minimum confirmation time are determined based on the environmental volatility scalar; if the conditions are met, the curtain wall ventilation components are switched to the same candidate mode; otherwise, the current mode is maintained.

10. The system according to claim 9, characterized in that, The calculation of the time gradient vector based on indoor air temperature, outdoor air temperature, and solar radiation intensity within a preset time window, and the integration of the time gradient vector into an environmental volatility scalar, includes: Acquire continuous sampled values ​​of indoor air temperature, outdoor air temperature, and solar radiation intensity at multiple sampling times within a preset time window; Calculate the rate of change of each parameter within a preset time window to form a multidimensional original time gradient vector; The original time gradient vector is standardized and its magnitude is calculated, which is then used as an environmental volatility scalar representing the overall rate of change of the system.

Images