A multi-zone temperature and humidity control method for large space air conditioning environment

Abstract

The application belongs to the technical field of air conditioning system operation optimization, and discloses a multi-region temperature and humidity control method for large-space air conditioning environment, which comprises the following steps: firstly, collecting historical operation data; secondly, predicting total load demand required for ensuring constant temperature and humidity in the large space; thirdly, predicting load demand required by different sub-regions under the non-physical boundary of the large space; and finally, solving the best operation mode meeting the temperature and humidity process demand of each region in each optimization period. The application can solve the best operation mode of the constant temperature and humidity purification air conditioning system in the large space.

Description

Technical Field

[0001] This invention relates to the field of air conditioning system operation optimization technology, and in particular to a method for controlling temperature and humidity in multiple zones of a large space air-conditioned environment. Background Technology

[0002] With the continuous advancement of global industrialization, the scale of industrial production is gradually expanding and the frequency of use of environmental protection equipment is also gradually increasing. Industrial air conditioning systems play a key role in modern industry, especially in industries with high environmental requirements, such as photovoltaic manufacturing, pharmaceutical production, large-scale precision experimental research and high-tech manufacturing. These industrial sites usually need to ensure that the air conditioning environmental parameters are uniform and remain within a very small fluctuation range for a long time.

[0003] However, traditional air conditioning control technology relies on human experience, and its shortcomings, such as long adjustment cycles and control lag, have gradually become apparent. It is difficult to effectively grasp the changing trends of temperature and humidity and to accurately adjust strategies, thus failing to achieve real-time and precise control of environmental parameters within the target allowable process requirements. Summary of the Invention

[0004] The purpose of this invention is to provide a method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment. By determining an optimization cycle and based on the operating data of the previous optimization cycle and the meteorological data of the next cycle, the operating mode for the next optimization cycle is automatically set (the operating mode is the controllable relevant parameters of the air-conditioning system within the optimization cycle). Then, it is determined whether the operating mode can meet the temperature and humidity process requirements, and the temperature and humidity change range of the mode in the future time domain is calculated, thereby finding the optimal operating mode that meets the process requirements in each optimization cycle.

[0005] The technical solution adopted by this invention to solve its technical problem is as follows: A method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment includes the following steps: Collect historical operational data; Predict the total load required to maintain constant temperature and humidity in a large space; Predict the load demand required by different sub-regions under non-physical boundaries of a large space; Find the optimal operating mode that meets the temperature and humidity process requirements of each region within each optimization cycle.

[0006] In some embodiments, predicting the total load requirement for maintaining constant temperature and humidity in a large space includes the following steps: Determine the main control temperature and humidity values ​​to characterize the total load demand; Forecast total load demand.

[0007] In some embodiments, determining the master temperature and humidity values ​​to characterize the total load demand includes the following steps: A master temperature and humidity value is reconstructed by measuring the temperature and humidity values ​​at all global measurement points to reflect the actual total load demand of the space. PCA method was used to analyze temperature and humidity data from multiple measurement points in a large space to identify the influence weight of each temperature and humidity measurement point on the overall load change of the space. Based on the PCA principal component analysis results, the weighted average of each set of real-time monitoring data is used to calculate the main control temperature and humidity values ​​required for this invention.

[0008] In some embodiments, the predicted total load demand refers to: Based on the master control parameters, a master control temperature and humidity prediction model is constructed using TCN and BiLSTM time-series prediction modules to predict the total load demand of a large space in the future time domain.

[0009] In some embodiments, predicting the load demand required by different sub-regions under the non-physical boundaries of a large space includes the following steps: Determine the temperature and humidity influence characteristics of each sub-region and the mapping of measurement points; Predict temperature and humidity changes in each sub-region.

[0010] In some embodiments, determining the temperature and humidity influence characteristics of each sub-region and the mapping of measurement points includes the following steps: Clearly define the thermal and humidity requirements of each sub-region; Based on airflow organization and measurement point layout, non-physical boundary sub-regions are divided, a one-to-one mapping relationship between sub-regions and measurement points is established, and core characteristic parameters affecting the temperature and humidity of each sub-region are screened. Analyze the response patterns of each sub-region to air conditioning control.

[0011] In some embodiments, predicting the temperature and humidity changes in each sub-region includes the following steps: Based on system operating parameters and meteorological data, a multi-point temperature and humidity prediction model is constructed using a CNN-Transformer hybrid neural network to predict the temperature and humidity changes of each sub-region in the future time domain. Calculate the temperature and humidity deviations of each sub-region, quantify their cooling, heating, and humidity load requirements, and classify the control priorities according to the degree of deviation.

[0012] In some embodiments, the process of determining the optimal operating mode that meets the temperature and humidity requirements of each region within each optimization cycle includes the following steps: Read the current system operation and environment data; Predict the changes in main control temperature and humidity values ​​during the optimization cycle; Predict the changes in temperature and humidity values ​​at each measuring point during the optimization period; Determine the specific circumstances of the total load demand in the large space; Determine the temperature and humidity values ​​of each sub-area in a large space.

[0013] In some embodiments, determining the specific situation of the total load demand condition in the large space refers to: If it is the first case, the range of adjustment of the operating parameters is too small to ensure that the main control temperature and humidity values ​​are close to the center value of the process requirements; if it is the second case, the range of adjustment of the operating parameters is too large to ensure that the main control temperature and humidity values ​​recover to the range of the process requirements more quickly.

[0014] In some embodiments, determining the temperature and humidity values ​​of each sub-region of the large space refers to: For sub-regions that are within the process requirements, the first case is adopted, where the allocated load only needs to offset local thermal and moisture disturbances and ensure that it remains at the center value of the process requirements. For sub-regions that are not within the process requirements, the second case is adopted, where the allocated load is given priority, including the load that offsets heat and moisture generation and ensures that it returns to the process requirements zone within 1 to 2 optimization cycles.

[0015] The beneficial effects of this invention are: The optimization method provided by this invention can solve for the optimal operating mode of a large-space constant temperature and humidity clean air conditioning system that meets the process requirements of each sub-region within the set control optimization time domain. Furthermore, the operation optimization method provided by this invention splits a large number of controllable parameters in the large-space air conditioning system into two parts for independent optimization and control, which to some extent reduces the difficulty and slowness of solving during the operation optimization process. It is suitable for large-space air conditioning systems with high process requirements, a large number of temperature and humidity measuring points, and a certain level of building automation infrastructure (without the need for manual adjustment of the main unit and related equipment). Attached Figure Description

[0016] Figure 1 The diagram shows the structure of the main control temperature and humidity prediction model provided by this invention.

[0017] Figure 2 The diagram shows the structure of the multi-point temperature and humidity prediction model provided by this invention.

[0018] Figure 3 The structure diagram of the NMPC system based on the IPSO algorithm provided by this invention.

[0019] Figure 4 The present invention provides a flowchart of the control process for an industrial constant temperature and humidity purification air conditioning system based on the NMPC-IPSO algorithm.

[0020] Figure 5 This invention provides a diagram showing the variation of the main control parameter control curve. Figure 6 The diagram shows the variation of the DC water valve opening degree 1~4 provided by this invention; Figure 7The diagram shows the variation of the DC water valve opening degree control curve from 5 to 8 provided by this invention. Figure 8 The diagram showing the variation of the (d) DC water valve opening degree control curve from 9 to 12 provided by this invention; Figure 9 The main control temperature and humidity change curves before and after optimization are provided by this invention; Figure 10 The temperature variation curves 1-3 before and after optimization provided by this invention; Figure 11 The temperature variation curves 4-6 before and after optimization provided by this invention; Figure 12 The relative humidity variation curves before and after optimization provided by this invention are shown in Figure 1-3. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] In one exemplary implementation, this embodiment provides a method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment. This method may include the following steps: S1. Collect historical operation data; S2. Predict the total load requirement for maintaining constant temperature and humidity in a large space. S3. Predict the load demand required by different sub-regions under the non-physical boundaries of a large space. S4. Solve for the optimal operating mode that meets the temperature and humidity process requirements of each region within each optimization cycle.

[0023] This method is suitable for large-space air conditioning systems with high process requirements, a large number of temperature and humidity measuring points, and a certain level of building automation infrastructure (no need for manual adjustment of the main unit and related equipment).

[0024] In one exemplary implementation, step S2 in the above method of this embodiment specifically comprises: S21. Determine the main control temperature and humidity values ​​to characterize the total load demand: First, the purpose of this method is to reconstruct a master temperature and humidity value by using the temperature and humidity values ​​of all global measuring points to reflect the actual total load demand of the space; Secondly, the PCA method was used to analyze the temperature and humidity data from multiple measurement points in a large space in order to identify the influence weight of each temperature and humidity measurement point on the overall load change of the space. Finally, based on the PCA principal component analysis results, the weighted average of each set of real-time monitoring data was calculated to obtain the main control temperature and humidity values ​​required by the present invention. S22. Forecast total load demand: Since the operation of a large space air conditioning system has certain dynamic characteristics, it is difficult to respond to the upcoming load changes in a timely manner by relying solely on the current observation values ​​for regulation. Therefore, this invention relies on the main control parameters (supply air temperature and humidity setpoints, supply air volume, and total cooling capacity of local cold plates) to construct a main control temperature and humidity prediction model through TCN and BiLSTM time-series prediction modules to predict the total load demand of the large space in the future time domain.

[0025] In one exemplary implementation, step S3 in the above method of this embodiment specifically comprises: S31. Determine the temperature and humidity influence characteristics of each sub-region and the mapping of measuring points: First, the purpose of this method is to solve the problem that a single global parameter cannot reflect local temperature and humidity differences, and to clarify the thermal and humidity demand characteristics of each sub-region; Secondly, based on airflow organization and measurement point layout, non-physical boundary sub-regions are divided, a one-to-one mapping relationship between sub-regions and measurement points is established, and core characteristic parameters affecting the temperature and humidity of each sub-region are screened. Finally, the response patterns of each sub-region to air conditioning control are analyzed to provide a basis for multi-point forecasting.

[0026] S32. Predict temperature and humidity changes in each sub-region: Since the thermal and humidity disturbances in different areas of a large space vary significantly, relying solely on global total load forecasting cannot achieve refined zonal control. Therefore, this invention relies on system operating parameters and meteorological data to construct a multi-point temperature and humidity prediction model using a CNN-Transformer hybrid neural network to predict the temperature and humidity changes of each sub-region in the future time domain. Finally, the temperature and humidity deviations of each sub-region are calculated based on the process requirements, the cooling, heating and humidity load requirements are quantified, and the control priorities are divided according to the degree of deviation.

[0027] In one exemplary implementation, step S4 of the method described above in this embodiment specifically includes: S41. Read the current system operation and environment data; S42. Predict the changes in the main control temperature and humidity values ​​during the optimization cycle; S43. Predict the changes in temperature and humidity values ​​at each measuring point during the optimization period. S44. Determine the specific situation of the total load demand in the large space; if it is the first situation (situation 1), the adjustment range of the operating parameters is too small to ensure that the main control temperature and humidity values ​​are close to the center value of the process demand; if it is the second situation (situation 2), the adjustment range of the operating parameters is too large to ensure that the main control temperature and humidity values ​​recover to the process demand range more quickly. S45. Determine the temperature and humidity values ​​of each sub-region in the large space. For sub-regions that are within the process requirements, the first case (case 1) is adopted, and the allocated load only needs to offset local thermal disturbance and humidity disturbance and ensure that it is maintained at the central value of the process requirements. For sub-regions that are not within the process requirements, the second case (case 2) is adopted, and the allocated load will be given priority, mainly including the load to offset heat and humidity generation and ensure that it returns to the process requirements zone within 1 to 2 optimization cycles.

[0028] In another example implementation, this embodiment describes a constant temperature and humidity cleanroom air conditioning system for a large factory building, requiring a regulated air conditioning area of ​​3075.2 m². 2 The system employs a semi-centralized air conditioning system with MAU-DC-FFU. Currently, the system operates by having management personnel control relevant parameters based on experience-based methods, observing fluctuations at the main monitoring point, to ensure temperature and humidity in each area.

[0029] In this embodiment, the method for optimizing the operation of a large-space constant temperature and humidity purification air conditioning system provided in the above embodiments is also adopted, as shown in the attached figure. Figure 4 As shown.

[0030] For step S1 of this embodiment of the invention, the operating data of the constant temperature and humidity purification air conditioning system of a large space factory from June 16, 2022 to August 12, 2024 for a period of 788 days has been collected; the collection time interval is 1 minute, and the main parameters include: temperature monitoring values ​​1~6, relative humidity monitoring values ​​1~3, outdoor temperature, outdoor humidity, supply air temperature, supply air humidity, supply air volume, total cooling capacity of local cold plates, and opening degree of water valves of each local cold plate.

[0031] In one example implementation, step S2 of this embodiment specifically includes: S21. Determine the main control temperature and humidity values ​​to characterize the total load demand.

[0032] First, the purpose of this method is to reconstruct a master temperature and humidity value by using the temperature and humidity values ​​of all global measuring points to reflect the actual total load demand of the space; Secondly, the PCA method was used to analyze the temperature and humidity data from multiple measurement points in a large space in order to identify the influence weight of each temperature and humidity measurement point on the overall load change of the space. Furthermore, the calculation principle and influence weight results of the PCA method are shown below: ; In the formula: It is the raw data. It is the mean of each variable; ; In the formula: It is the covariance matrix. This is the centralized data, where n is the number of samples; ; In the formula: It is an eigenvalue. It is an eigenvector, representing the direction of the principal component; ; In the formula: These are the eigenvectors corresponding to the selected principal components; ; In the formula: Z is the data after dimensionality reduction. It is an eigenvector matrix (composed of selected principal components).

[0033] In this embodiment, the principal component analysis results for the temperature measurement points shown in Table 1 and the principal component analysis results for the relative humidity measurement points shown in Table 2 can be obtained.

[0034] Table 1. Principal component analysis results at temperature measurement points.

[0035] Table 2. Principal component analysis results of relative humidity measurement points.

[0036] Finally, based on the PCA principal component analysis results, the weighted average of each set of real-time monitoring data was calculated to obtain the main control temperature and humidity values ​​required by the present invention. Furthermore, the calculation method for the main control temperature and humidity values ​​is as follows: ; In the formula: Main control temperature or humidity, These are the actual values ​​from each temperature or humidity measuring point. These are the principal component coefficient values ​​corresponding to each humidity measurement point.

[0037] S22. Predict total load demand.

[0038] Based on the actual survey results, the main control parameters of this air conditioning system are the supply air temperature and humidity setpoint, supply air volume, and total cooling capacity of the local cold plate. This embodiment relies on the main control parameters to construct a main control temperature and humidity prediction model through TCN and BiLSTM time series prediction modules (as shown in the attached figure). Figure 1This is used to predict the total load demand of a large space in the next 5 minutes.

[0039] In this embodiment, for Figure 1 Detailed introduction of all modules of the master temperature and humidity prediction model built in the system: This invention relies on master control parameters to construct a master control temperature and humidity prediction model through TCN and BiLSTM time series prediction modules. The model mainly consists of an input layer, a TCN feature extraction module, a BiLSTM time series modeling module, and a fully connected output layer.

[0040] The system comprises several layers: an input layer for receiving multi-source time-series data, including historical master temperature and humidity sequences, air conditioning system master control parameters, and outdoor meteorological disturbance parameters; a TCN feature extraction module for extracting local variation features of master temperature and humidity and related control parameters at different time scales, which includes structures such as dilated causal convolution, weight normalization, ReLU activation function, Dropout random deactivation, and residual connections; a BiLSTM time-series modeling module for further capturing long-term temporal dependencies in the master temperature and humidity variation process, where forward LSTM is used to learn the forward temporal variation trend and backward LSTM is used to supplement the reverse temporal dependency information; and a fully connected output layer for mapping the fused temporal features to predicted master temperature and humidity values ​​for future optimization cycles.

[0041] In one example implementation, the total load demand forecasting process in this embodiment is as follows: First, data input preprocessing: The historical master temperature and humidity input sequence, system master control parameters and exogenous features such as outdoor weather are normalized and preprocessed to eliminate the influence of different dimensions before being input into the TCN feature extraction layer.

[0042] Secondly, multi-scale temporal feature extraction: The input data is sequentially processed through stacked temporal blocks: local variation features at different time scales are extracted through dilated causal convolution, nonlinear transformation is completed through weight normalization and ReLU activation, overfitting is suppressed through Dropout, and effective gradient propagation is ensured through residual connections. The final output is a TCN feature sequence that integrates multi-scale local patterns.

[0043] Then, bidirectional long-term dependency modeling: TCN feature sequences are input into the BiLSTM layer: forward LSTM captures the positive temporal trend, backward LSTM captures the negative temporal dependency, and the hidden states of the two are concatenated to obtain a deep fusion feature that simultaneously contains short-term local features and long-term bidirectional dependencies.

[0044] Then, the predicted output and load representation are performed: the fused feature input is mapped to the fully connected layer to output the master temperature and humidity prediction sequence for the future optimization cycle.

[0045] Since the main temperature and humidity are global representative indicators obtained by PCA weighting, they can characterize the overall heat and humidity load level of large industrial spaces. Therefore, this prediction sequence can characterize the changes in total load demand in the future time domain, providing a basis for judging the subsequent global control magnitude.

[0046] In one example implementation, step S3 in this embodiment specifically includes: S31. Determine the temperature and humidity influence characteristics of each sub-region and the mapping of measurement points.

[0047] First, the purpose of this method is to solve the problem that a single global parameter cannot reflect local temperature and humidity differences, and to clarify the thermal and humidity demand characteristics of each sub-region; Secondly, based on the airflow organization of the large space and the arrangement of 6 temperature measuring points and 3 humidity measuring points, the continuous space without clear physical separation is divided into corresponding non-physical boundary sub-regions. A one-to-one mapping relationship between the sub-regions and the measuring points is established, and characteristic parameters such as supply air temperature and humidity, supply air volume, opening degree of each local DC water valve, and outdoor temperature and humidity are selected. Finally, all feature parameters are normalized preprocessed to eliminate the influence of different units on the prediction model. Furthermore, the feature normalization calculation method is as follows: ; In the formula: Normalized value; It is the minimum value; This is the maximum value.

[0048] S32. Predict the temperature and humidity changes in each sub-region.

[0049] Based on actual survey results, the newly added control parameters for each sub-zone of the air conditioning system are the opening degree of each local DC water valve. This invention relies on system operating parameters and meteorological data to construct a multi-point temperature and humidity prediction model through a CNN-Transformer hybrid neural network (as attached). Figure 2 This is used to predict the temperature and humidity changes in each sub-region within the next 5 minutes.

[0050] See Figure 2 In this embodiment, following the data flow from input to output and from bottom to top, the model can be divided into four parts: input and primary feature extraction layer, Transformer encoding / decoding layer, adaptive fusion layer, and output layer. The meaning and function of all modules are as follows: (a) Input and primary feature extraction layer, which mainly includes: The input layer contains normalized historical temperature and humidity data for each measuring point, air conditioning system operating parameters (supply air temperature and humidity, supply air volume, and opening degree of each local DC water valve), outdoor temperature and humidity meteorological data, and predicted values ​​of global master control parameters. All data are input into the model after being aligned by time step.

[0051] Feature concatenation, or feature concatenation module, integrates input features from different sources and dimensions along the time dimension to form a unified temporal feature sequence, ensuring that the time steps of all features strictly correspond and providing standardized input for subsequent feature extraction.

[0052] Linear Layers 1 to n are linear transformation layers, with each sub-prediction task corresponding to an independent linear layer.

[0053] The shared input is adapted for dimension and preliminarily mapped to transform the unified input features into the feature space corresponding to each subtask, adapting to the differences in feature scale and data distribution in different sub-regions.

[0054] CNN (Convolutional Neural Network) is a local feature extraction module shared by all tasks. It captures local fluctuation patterns and short-range correlations between parameters in time-series data through convolution operations, such as short-term trends in temperature and humidity and the instantaneous response characteristics of control actions, extracting high-dimensional local time-series features to provide a foundation for subsequent deep time-series modeling.

[0055] The features from Task 1 to Task n are the primary features specific to each prediction subtask. After extraction by the corresponding linear layer and CNN, each sub-region (measuring point) has its own initial feature sequence. Each feature corresponds to the temperature and humidity prediction task of a sub-region and serves as the input for subsequent Transformer time series modeling.

[0056] (ii) The Transformer encoding / decoding layer, which mainly includes: A shared encoder, also known as a shared Transformer encoder, is an encoding module used by all prediction tasks. In this embodiment, a global temporal model of the features of all tasks can be performed based on a self-attention mechanism to mine common temperature and humidity variation patterns in all sub-regions (such as the common effects of global meteorological disturbances and changes in air supply parameters), thereby achieving cross-task knowledge sharing and improving the prediction accuracy in areas with smaller sample sizes.

[0057] Encoder 1 to Encoder n are task-specific Transformer encoders, with each subtask corresponding to an independent encoder. Dedicated time-series modeling is performed for the thermal and humidity characteristics and local disturbance patterns of each sub-region, capturing the differentiated response characteristics of different sub-regions (e.g., areas closer to heat-generating equipment experience faster temperature rise, while corner areas have slower humidity response), preserving the unique characteristics of each region.

[0058] The shared decoder, also known as the shared Transformer decoder, is a decoding module used by all prediction tasks. It performs unified temporal decoding based on the encoded global common features, generating prediction representations shared by all tasks. It leverages global patterns to provide general support for predictions in each sub-region, enhancing information reuse among multiple tasks.

[0059] Decoder 1 to Decoder n are task-specific Transformer decoders, with one decoder for each subtask. They perform dedicated decoding based on the thermal and humidity response characteristics of their respective sub-regions, outputting a high-dimensional predictive representation of the corresponding sub-region to accurately depict the temperature and humidity change trends of each sub-region.

[0060] Vector 1 to Vector n are feature vectors, which are the high-dimensional feature vectors of the corresponding subtask output by each dedicated decoder. The vectors integrate global common laws and sub-region-specific characteristics, and contain the main prediction information of temperature and humidity changes in the sub-region. They serve as the intermediate carrier connecting the decoding module and the fusion module.

[0061] (III) Adaptive feature fusion layer, which mainly includes: The selector, or feature selector, automatically learns the optimal weight ratio of "global common features" and "region-specific features" in the prediction of each sub-region, achieving adaptive fusion of the two types of features and balancing the effects of multi-task knowledge sharing and task-specific characteristics.

[0062] Gating (Gate) refers to a gate unit, with one independent gate corresponding to each subtask. In this embodiment, it works in conjunction with the selector to control the flow of information between shared and specific features through the gating mechanism, filtering redundant information and retaining the features most effective for the current subtask, thereby improving the model's training stability and prediction accuracy.

[0063] Softmax, or normalized exponential function, normalizes the weights output by the selector in this embodiment, mapping them to a probability distribution in the range of 0 to 1. This ensures that the sum of the fused weights is 1, enabling adaptive and interpretable adjustment of the feature fusion ratio.

[0064] Fc (Fully Connected Layer) is the layer that is fully connected. In this embodiment, the fused feature vector is subjected to dimensional mapping and nonlinear transformation to convert the high-dimensional prediction representation into the output dimension corresponding to the temperature and humidity values, thus preparing for the final output result.

[0065] (iv) Output layer, which mainly includes: Output 1 to Output n are the predicted outputs, each corresponding to a temperature and humidity prediction sequence for a sub-region (measuring point). The output content is the temperature or humidity value at each time step in the future time domain, which is the final prediction result of the model.

[0066] Task 1 to Task n are the prediction tasks, each corresponding to the temperature and humidity prediction of a sub-region. In this embodiment, there are 6 temperature prediction tasks and 3 humidity prediction tasks, for a total of 9 sub-tasks, corresponding to 6 temperature measurement points and 3 humidity measurement points in the space.

[0067] In one example implementation, the calculation process for temperature and humidity prediction and zoned load demand in each sub-region is as follows: First, multi-source data input and preprocessing: normalized historical temperature and humidity data, system operating parameters, outdoor meteorological data, and predicted values ​​of global master control parameters from each measuring point are input into the model. The data are then integrated by the feature splicing module according to time steps to form a unified input feature sequence.

[0068] Secondly, in the primary local feature extraction, the input features are adapted to the dimensions by the linear layers dedicated to each task and then fed into the shared CNN layer. The local fluctuation features of the time series data are extracted and correlated with short-range parameters through convolution operations to obtain the primary feature sequence corresponding to each subtask.

[0069] Then, the Transformer employs dual-path encoding: commonality and individuality are modeled separately. The primary features of each task are input into a dual-path encoding structure consisting of a shared encoder and a dedicated encoder. The shared encoder mines the common variation patterns of temperature and humidity across the entire space, enabling cross-task knowledge sharing; the dedicated encoder captures the differentiated thermal and humidity response characteristics of each sub-region, preserving regional individuality. After the two-path encoding is completed, a deep encoded feature that integrates global and local information is obtained.

[0070] Then, the Transformer performs dual-path decoding: generating predictive representations and inputting encoded features into a dual-path decoding structure of "shared decoder + dedicated decoder": the shared decoder generates general predictive representations based on global rules; the dedicated decoder generates dedicated predictive representations for the characteristics of each sub-region and outputs high-dimensional feature vectors for the corresponding sub-tasks.

[0071] Then, adaptive feature fusion is performed. In this embodiment, the selector combines the gating unit and Softmax normalization to dynamically calculate the fusion weight of shared features and exclusive features in each subtask. The features output by the two decoders are adaptively weighted and fused to filter redundant information and obtain the optimal prediction representation.

[0072] Finally, the output mapping and load quantization are performed. In this embodiment, the fused features undergo dimensionality transformation through a fully connected layer, ultimately outputting a minute-by-minute temperature and humidity prediction sequence for each sub-region within the future optimization cycle. Combined with the process allowable range, the temperature and humidity deviation of each sub-region can be calculated, the cooling, heating, and humidity load requirements can be quantified, and the control priority can be divided according to the degree of deviation.

[0073] It should be noted that in this embodiment, step S4 uses NMPC-IPSO to solve for the optimal operating mode. The main parameters are set as follows: control time domain H=12, i.e., one set of operating conditions every 60 minutes; particle number: 100; number of iterations: 60; and inertial weight: [value missing]. Individual learning factor is Social learning factor is The maximum mutation probability is 0.1, and the mutation strength is 0.1. See the appendix for the detailed procedure. Figures 3-4 ,in, Figure 3 This diagram shows the system architecture of a nonlinear model predictive control (NMPC) system based on the improved particle swarm optimization (IPSO) algorithm. It represents the core algorithmic framework for solving the optimal operating mode in step S4. All variables in the diagram can be categorized into two types: NMPC control framework variables and IPSO optimization solution variables. The meaning and function of each variable will be explained below in conjunction with each sub-step of S4.

[0074] S41. Read the current system operation and environment data.

[0075] This step provides the initial state and boundary conditions for the NMPC framework. Figure 3 Input class variables in: 1.x(k): The system state vector at time k, which represents the set of system states at the initial time of the current optimization cycle (time k). Specifically, it corresponds to the real-time monitoring values ​​of the 6 temperature measurement points and 3 humidity measurement points read in this step, as well as the air conditioning system operating parameters such as supply air temperature and humidity, supply air volume, total cooling capacity of the cold plate, and opening degree of each local DC water valve. It is the initial state input for subsequent prediction model and optimization solution.

[0076] 2.d(k): Exogenous disturbance vector at time k, which represents the uncontrollable external disturbance parameters in the current and future optimization cycles. Specifically, it corresponds to the meteorological forecast data such as outdoor temperature and outdoor humidity read in this step for the next optimization cycle, and is used as the boundary condition input of the prediction model to simulate the influence of the external environment on indoor temperature and humidity.

[0077] S42. Predict the changes in the main control temperature and humidity values ​​during the optimization cycle.

[0078] S43. Predict the changes in temperature and humidity values ​​at each measuring point during the optimization period.

[0079] These two steps correspond Figure 3 The core variables of the prediction model module are as follows: 3. The system output prediction sequence represents the predicted system output values ​​for the next Hp time steps based on the system state at time k. Specifically, S42 outputs the global master temperature and humidity prediction sequence, and S43 outputs the temperature and humidity prediction sequences for each sub-region measurement point; the variable "k+i|k" indicates the predicted output value at time k+i based on the information at time k.

[0080] 4. yr(k+i) (i=1,2,…,Hp): Reference trajectory sequence, which represents the process target trajectory for temperature and humidity control. It corresponds to the process center value of 23℃ temperature and 50% relative humidity in the patent, as well as the allowable range of ±0.3℃ temperature and 40%~60% relative humidity. It is the tracking target and constraint judgment benchmark for optimized control.

[0081] 5. Hp: Prediction time domain, which represents the length of future time steps covered by the model prediction. In this embodiment, the prediction time domain covers the entire 60-minute optimization cycle (predicting minute by minute, for a total of 60 time steps), which is used to predict the temperature and humidity change trend within the entire optimization cycle, supporting the judgment of operating conditions and optimization solution.

[0082] 6. Hc: Control time domain, which represents the length of the control action time step in the optimization solution. That is, the control time domain H=12 set in the patent corresponds to 12 control steps (each step is 5 minutes) within a 60-minute optimization cycle. The control quantity outside the control time domain remains constant, thereby reducing the solution dimension and improving the computational efficiency.

[0083] S44. Find the optimal operating mode that meets the temperature and humidity process requirements of each region.

[0084] This step is the core of the IPSO optimization solution, corresponding to Figure 3 The optimization objectives, constraints, and internal variables of the IPSO algorithm are specifically divided into two categories: (1) Variables in the NMPC optimization problem.

[0085] 7.J: Optimization objective function (performance index function), which represents the core evaluation index of NMPC optimization solution. This function is the basis for the optimization of the IPSO algorithm. The smaller the function value, the better the control scheme.

[0086] 8. U(k) = [u(k), u(k+1), ..., u(k+H_c-1)]^T: Control sequence, representing the sequence of air conditioning control parameters to be optimized. It is solved independently in two levels: the first level is the global master control parameter sequence, including the supply air temperature and humidity setpoints, supply air volume, and total cooling capacity of the cold plate; the second level is the zone allocation parameter sequence, including the opening degree of 12 local DC water valves. This sequence is the core solution object of the IPSO algorithm.

[0087] 9. u_min, u_max: Upper and lower limit constraints of control quantities, which represent the physical operating boundaries of air conditioning equipment, that is, the minimum and maximum values ​​allowed by the control parameters. For example, the opening range of the local DC water valve is 0%~100%, and the air supply volume must be within the rated range of the equipment. These are hard constraints that must be satisfied in the optimization solution.

[0088] 10. Δu_max: Control quantity change step size constraint, which represents the maximum allowable adjustment range of the control parameter within a single control step size. It is used to avoid excessively drastic control actions that could damage the equipment and extend the equipment's service life. This corresponds to the description in the patent that "the equipment's operating range and adjustment step size are the constraints".

[0089] (2) Core variables of the IPSO algorithm.

[0090] 11.x_i: The position vector of the i-th particle, which represents the specific value of the particle in the solution space, i.e. a set of candidate control quantity sequences U(k); each particle corresponds to a set of air conditioning operation parameters to be verified, and all particles in the particle swarm together constitute the candidate solution set for optimization.

[0091] 12. v_i: The velocity vector of the i-th particle, which represents the direction and step size of the particle's iterative update in the solution space. It is calculated by the inertia term, the individual cognition term, and the social learning term, and is used to update the particle's position in each iteration.

[0092] 13.w: Inertia weight, which represents the degree to which a particle inherits its historical velocity. In this patent, it is set to decrease linearly from 0.9 to 0.4. The weight is larger in the early stage of iteration, which is beneficial for the algorithm to perform a global search; the weight decreases in the later stage of iteration, which is beneficial for the algorithm to perform fine local optimization near the optimal solution.

[0093] 14.C1: Individual learning factor (cognitive factor), which represents the learning weight of a particle moving closer to its own historical best position. In this patent, it is linearly reduced from 2.5 to 0.5 to balance the influence weight of individual experience and group experience in the optimization process.

[0094] 15.C2: Social Learning Factor (Social Factor) This means that the learning weights for particles to move closer to the global optimal position of the population are linearly increased from 0.5 to 2.5 in this embodiment. In the later stages of iteration, the guiding role of the population optimal solution is gradually enhanced to improve the convergence efficiency.

[0095] 16. Pbest_i: Individual optimal position, which means: the sequence of control quantities that makes the objective function J optimal when the i-th particle searches during the iteration process, representing the historical optimal experience accumulated by the particle itself.

[0096] 17. Gbest: Global optimal position, which represents the optimal sequence of control variables searched by the entire particle swarm during the iteration process. It represents the current best solution found by the entire swarm and is the common guiding direction for all particles to iterate and update.

[0097] 18.p_m: Maximum mutation probability, which represents the trigger probability of the mutation operation in the improved particle swarm optimization algorithm. In this embodiment, it is set to 0.1. During the iteration process, the particle positions are randomly perturbed with this probability to increase population diversity and avoid the algorithm getting trapped in local optima.

[0098] 19. σ: Mutation intensity, which represents the perturbation amplitude of the particle position by the mutation operation. In this patent, it is set to 0.1 to control the magnitude of the mutation length and balance the global exploration capability and local development capability of the algorithm.

[0099] In the iterative solution process of S44, the algorithm first initializes the position x_i and velocity v_i of all particles, then calculates the objective function J for each particle and updates Pbest_i and Gbest, and then updates the particle velocity and position through w, C1, and C2, while triggering mutation operation with probability p_m. After 60 iterations, the final output is the control quantity sequence corresponding to Gbest, which is the optimal running mode for the next optimization cycle.

[0100] Specifically, step S4 above is mainly achieved through the following steps: S41. Read the current system operation and environment data.

[0101] First, the purpose of this method is to obtain the real-time initial state and boundary conditions required for optimization calculation, so as to provide a basis for subsequent prediction and optimization; Secondly, read the real-time monitoring values ​​of 6 temperature measuring points and 3 humidity measuring points at the current moment, as well as the air conditioning system operating parameters such as supply air temperature, supply air humidity, supply air volume, total cooling capacity of the cooling coil, and opening degree of each local DC water valve; Finally, read the outdoor temperature, outdoor humidity and other meteorological forecast data for the next optimization cycle (60 minutes).

[0102] S42. Predict the changes in the main control temperature and humidity values ​​during the optimization cycle.

[0103] First, the purpose of this method is to predict the changing trend of the total global load demand and determine the overall control direction of the system; Secondly, the current system operating parameters, historical master temperature and humidity data, and meteorological data for the next optimization cycle are input into the TCN-BiLSTM master temperature and humidity prediction model constructed by S22; Furthermore, the model outputs a minute-by-minute predicted sequence of master temperature and master humidity for the next 60 minutes; Finally, based on the predicted sequence, the total load demand of the large space is determined as follows: if the predicted values ​​of the main control temperature and humidity are all within the process allowable range (temperature 23±0.3℃, relative humidity 40%~60%), it is determined as case 1; if the main control temperature and humidity exceed the process allowable range at any time, it is determined as case 2.

[0104] S43. Predict the changes in temperature and humidity values ​​at each measuring point during the optimization period.

[0105] First, the purpose of this method is to predict the differences in local load demand in each sub-region, so as to provide a basis for refined load allocation by region; Secondly, the current system operating parameters, historical temperature and humidity data of each measuring point, meteorological data of the next optimization cycle, and the predicted values ​​of the global master control parameters obtained by S42 are input into the CNN-Transformer multi-point temperature and humidity prediction model constructed by S32. Furthermore, the model outputs a predicted temperature and humidity sequence for 6 temperature measurement points and 3 humidity measurement points every minute over the next 60 minutes; Finally, the temperature and humidity conditions of the corresponding sub-region are determined based on the predicted sequence of each measuring point: if all the predicted temperature and humidity values ​​of the sub-region are within the process allowable range, it is determined as case 1; if the temperature and humidity of the sub-region exceed the process allowable range at any time, it is determined as case 2.

[0106] S44. Find the optimal operating mode that meets the temperature and humidity process requirements of each region.

[0107] First, the purpose of this method is to find the optimal combination of control parameters that ensures the temperature and humidity of all sub-regions in the entire space meet the requirements, while satisfying the physical constraints of the equipment. Secondly, an optimization solution framework based on NMPC-IPSO is constructed, with the optimization objectives being to minimize the temperature and humidity tracking error and the control action amplitude, and the constraints being the equipment operating range and adjustment step size. Furthermore, optimization strategies are determined based on the total load condition and the sub-regional operating conditions: If the total load is Case 1, the global main control parameter adjustment range is relatively small, so that the main control temperature and humidity values ​​are maintained near the process requirement center value; if the total load is Case 2, the global main control parameter adjustment range is relatively large, so that the main control temperature and humidity values ​​are restored to the process allowable range first. This part is mainly responsible for adjusting the main control parameters, namely: supply air temperature and humidity set value, supply air volume, and total cooling capacity of the cold plate. If the sub-region is Case 1, the allocated load only needs to offset the local thermal disturbance and humidity disturbance to ensure that it is maintained near the center value of the process requirement; if the sub-region is Case 2, the allocated load will prioritize the sub-region to ensure that it returns to the process allowable range within 1 to 2 optimization cycles. This part optimizes the allocation parameters under the condition of fixing the main control optimization parameters of the previous stage, namely: the opening degree of the local cold plate water valve is 1 to 12. Finally, by improving the particle swarm optimization (IPSO) algorithm to solve for the optimal control sequence in a rolling manner, the optimal setpoint for the next optimization cycle is output.

[0108] In this embodiment, a constant temperature and humidity purification air conditioning system for a large-scale experimental research base, under the condition of strictly maintaining the original process temperature and humidity accuracy requirements (temperature 23±0.3℃, relative humidity 40%~60%), is compared with the original manual experience control method for temperature and humidity control. The results are as follows: (1) The temperature at each measuring point converges rapidly to near the target value in the initial stage and remains stable at around 23℃ for a long time. The curve is flatter, the average temperature fluctuation of the whole space is reduced, and the temperature of all measuring points is within the allowable range of the process throughout the process.

[0109] (2) The humidity at each measuring point rises rapidly to about 50% in the early stage of control, and then remains relatively stable with a decrease in average fluctuation.

[0110] Furthermore, no single measuring point achieved the standard while other measuring points deviated for an extended period during the control process. This indicates that the control strategy of this invention does not merely compensate for a single point, but rather achieves coordinated and stable temperature and humidity across multiple points in the entire space through global air treatment and terminal zone allocation. Specific changes in operating parameters and variations at each temperature and humidity measuring point can be found in the appendix. Figure 5 , Figure 6 Figure 7 and Figure 6 ,as well as Figure 9 , Figure 10 , Figure 11 and Figure 12 .

[0111] Among them, for Figures 5-8 The following is a comparative analysis of the controllable parameter variation curves: Comparison benchmark: The original manual experience control method was used as the control group, and the NMPC-IPSO optimized control method in this embodiment was used as the experimental group. The changes in the system controllable parameters within 288 time steps (corresponding to 288 minutes of continuous operation) were compared to verify the differences in control methods.

[0112] See Figure 5 The control curves of the main control parameters show the changes in supply air temperature, supply air humidity, supply air volume, and DC cooling capacity. Under the original manual control, the changes in each main control parameter were relatively small, relying mainly on experience-based settings and local adjustments, making it difficult to promptly match the dynamic changes in the heat and humidity load of a large space. After adopting the NMPC-IPSO control of this invention, the supply air temperature, supply air humidity, supply air volume, and DC cooling capacity exhibit segmented, step-like changes. New setpoints are provided based on predicted load demand in different control cycles, and these values ​​remain relatively stable within the cycle. This change pattern indicates that this invention can adjust the main control parameters in advance based on future temperature and humidity deviations, avoiding temperature and humidity fluctuations caused by the lag in manual control.

[0113] Therefore, through Figure 5 It can be seen that by predictively adjusting the main control parameters in segments, the system's ability to respond to changes in total load is enhanced, enabling the air conditioning system to actively compensate before or in the early stages of disturbance, thereby improving the stability of global temperature and humidity control.

[0114] See Figure 6 The control curves for the DC water valve openings (1-4) show that, under the original control, some valves exhibited continuous fluctuations or significant periodic changes, indicating a lack of unified and coordinated valve actions. After adopting the control method of this invention, the valve openings are adjusted in a stepwise manner mainly at control cycle nodes, remaining relatively stable within the cycle. This demonstrates that the controller can adjust the terminal water valves as needed according to the heat and humidity requirements of the corresponding area.

[0115] Therefore, through Figure 6 It can be seen that the frequent operation of the water valve has reduced the execution disturbance and improved the stability of temperature and humidity regulation in the local area.

[0116] See Figure 7 The DC water valve opening control curves (5-8) show that, compared to the original control, the openings of different water valves no longer change synchronously under the control of this invention, but rather form differentiated opening distributions according to the load demands of different areas. Some valves undertake the main regulating role, while others maintain a lower opening or participate in regulation in stages, reflecting the characteristics of zoned load distribution.

[0117] Therefore, through Figure 7 It can be seen that by differentiating the opening degree of the local cold plate water valve, the system oscillation caused by the simultaneous large-scale operation of all terminals can be avoided, so that different sub-regions can obtain the adjustment amount that matches their load demand.

[0118] See Figure 8 The DC water valve opening control curve (9-12) shows that, under the original control, some valves operated at high or low opening for extended periods, easily leading to over-adjustment or under-adjustment in localized areas. Under the control of this invention, the valve opening is adjusted in stages based on the predicted deviation, compensating for localized thermal and humidity disturbances at different time periods.

[0119] Therefore, through Figure 8 It can be seen that the ability to regulate edge areas or local disturbance areas has been improved, and the effect of coordinated temperature and humidity control in large spaces and multiple areas has been enhanced.

[0120] for Figures 9-12 The temperature and humidity control effect curves are compared and analyzed as follows: See Figure 9 The master temperature and humidity change curves are obtained by PCA principal component analysis weighted calculation, which represent the overall temperature and humidity level of the entire space and are a direct representation of the total load demand.

[0121] Temperature dimension: The original control's master temperature curve fluctuated wildly, covering an overall range of 22.8℃ to 23.4℃, frequently touching the upper and lower boundaries of the process's allowable range, posing a risk of exceeding limits at certain times; moreover, the response to load changes was lag-dependent, with the temperature only starting to fall back after a significant peak, requiring a long time to return to stability. The master temperature controlled by this invention quickly converges to near the process center value of 23℃ in the initial stage of operation, with minimal fluctuations throughout, a flat and stable curve, and always remaining within the process's allowable range without boundary touches.

[0122] Humidity dimension: The original controlled humidity fluctuated repeatedly within the range of 45% to 50%, with large fluctuations and poor stability. The controlled humidity of this invention rapidly rises to about 50% in the middle of the process range at the beginning of the control, and then only fluctuates slightly, maintaining a stable state in the long term.

[0123] Therefore, through Figure 9 It can be seen that the control accuracy and stability of the overall temperature and humidity of the entire space have been significantly improved. The main control temperature and humidity are stably maintained near the central value of the process requirements, reducing the risk of large fluctuations and boundary exceedances under manual control, and greatly improving the control accuracy of the overall load.

[0124] See Figure 10 The temperature variation curves for points 1-3 are shown. Points 1-3 correspond to the core process area, which is the area with the highest environmental accuracy requirements for the production process. Under the original control method, the temperatures at all three measuring points fluctuated significantly, and the temperature differences between the measuring points were large, exhibiting the characteristics of "single-point fluctuation and multi-point unevenness." The temperature recovered slowly after being disturbed, requiring a long time to return to near the target value.

[0125] Under the control method of this invention, the temperature of the three measuring points converges rapidly to around 23.0℃, the curves of each measuring point are highly overlapping and flat, the temperature difference between regions is minimal, and the spatial temperature uniformity is significantly improved; the temperature of all measuring points remains stable within the process allowable range of 23±0.3℃ throughout the process, the response speed to local disturbances is faster, and it can quickly return to the center value after a slight deviation.

[0126] Therefore, through Figure 10 It can be seen that the temperature control accuracy and uniformity in the core process area have reached a higher level, basically meeting the environmental requirements for high-precision process production, and the temperature fluctuation range and regional temperature difference have been greatly reduced.

[0127] See Figure 11 The temperature variation curves for points 4-6 are shown. Points 4-6 correspond to the edge of the space and the area near the heat source, which are weak points in traditional control. Under the original control method, the temperature fluctuation range of this group of measuring points is much larger than that of the core area. The peak values ​​of some measuring points exceed 24℃, which significantly exceeds the process requirement of 23±0.3℃. Moreover, the temperature difference with the core area is large, and the temperature distribution throughout the space is extremely uneven, making it difficult for the edge area to consistently meet the standards.

[0128] Under the control method of this invention, the temperature of the edge area is effectively controlled through zoned fine-tuning and priority load allocation: the temperature of all measuring points is stable within the allowable range of the process, and the fluctuation range is significantly narrowed compared with the original control; at the same time, the temperature difference between the edge area and the core area is significantly reduced, and the uniformity of the temperature distribution in the whole space is significantly improved.

[0129] Therefore, through Figure 11 As can be seen, this embodiment solves the problem of uneven spatial distribution in traditional industrial air conditioning control, where "the core area is easy to meet the standard, while the edge area often exceeds the standard." It achieves stable temperature compliance in all areas of a large space and significantly improves the consistency of temperature and humidity control in a large space.

[0130] See Figure 12 The relative humidity variation curves from point 1 to point 3 are shown. Points 1 to 3 cover different areas of the entire space, reflecting the overall humidity distribution characteristics. Under the original control method, the humidity at all three points was below the process center value of 50%, with significant fluctuations and noticeable differences between points, indicating insufficient uniformity of humidity in the space.

[0131] Under the control method of this invention, the humidity at the three measuring points quickly stabilizes in the middle of the process range of 48% to 50%, with a significant reduction in fluctuation amplitude and a generally flat and stable curve; the humidity difference between each measuring point is significantly reduced, the humidity distribution in the entire space is uniform and consistent, and there are no cases of exceeding the allowable range of the process throughout the entire process.

[0132] Therefore, through Figure 12It can be seen that the stability and uniformity of humidity control in the whole space are improved simultaneously, and the humidity parameters are always maintained within the optimal range of the process range, avoiding the adverse effects of humidity fluctuations and boundary exceedances on the production process.

[0133] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment, characterized in that, Includes the following steps: Collect historical operational data; Predict the total load required to maintain constant temperature and humidity in a large space; Predict the load demand required by different sub-regions under non-physical boundaries of a large space; Find the optimal operating mode that meets the temperature and humidity process requirements of each region within each optimization cycle.

2. The method for multi-zone temperature and humidity control in a large-space air-conditioned environment according to claim 1, characterized in that, The prediction of the total load requirement for maintaining constant temperature and humidity in a large space includes the following steps: Determine the main control temperature and humidity values ​​to characterize the total load demand; Forecast total load demand.

3. The method for multi-zone temperature and humidity control in a large-space air-conditioned environment according to claim 2, characterized in that, The process of determining the main control temperature and humidity values ​​to characterize the total load demand includes the following steps: A master temperature and humidity value is reconstructed by measuring the temperature and humidity values ​​at all global measurement points to reflect the actual total load demand of the space. PCA method was used to analyze temperature and humidity data from multiple measurement points in a large space to identify the influence weight of each temperature and humidity measurement point on the overall load change of the space. Based on the PCA principal component analysis results, the weighted average of each set of real-time monitoring data is used to calculate the main control temperature and humidity values ​​required for this invention.

4. The method for multi-zone temperature and humidity control in a large-space air-conditioned environment according to claim 2, characterized in that, The predicted total load demand refers to: Based on the master control parameters, a master control temperature and humidity prediction model is constructed using TCN and BiLSTM time-series prediction modules to predict the total load demand of a large space in the future time domain.

5. The method for multi-zone temperature and humidity control in a large-space air-conditioned environment according to claim 1, characterized in that, The method for predicting the load requirements of different sub-regions under non-physical boundaries of a large space includes the following steps: Determine the temperature and humidity influence characteristics of each sub-region and the mapping of measurement points; Predict temperature and humidity changes in each sub-region.

6. The method for multi-zone temperature and humidity control in a large-space air-conditioned environment according to claim 5, characterized in that, The process of determining the temperature and humidity influence characteristics of each sub-region and mapping them to the measuring points includes the following steps: Clearly define the thermal and humidity requirements of each sub-region; Based on airflow organization and measurement point layout, non-physical boundary sub-regions are divided, a one-to-one mapping relationship between sub-regions and measurement points is established, and core characteristic parameters affecting the temperature and humidity of each sub-region are screened. Analyze the response patterns of each sub-region to air conditioning control.

7. A method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment according to claim 5, characterized in that, The prediction of temperature and humidity changes in each sub-region includes the following steps: Based on system operating parameters and meteorological data, a multi-point temperature and humidity prediction model is constructed using a CNN-Transformer hybrid neural network to predict the temperature and humidity changes of each sub-region in the future time domain. Calculate the temperature and humidity deviations of each sub-region, quantify their cooling, heating, and humidity load requirements, and classify the control priorities according to the degree of deviation.

8. A method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment according to claim 1, characterized in that, The process of finding the optimal operating mode that meets the temperature and humidity requirements of each region within each optimization cycle includes the following steps: Read the current system operation and environment data; Predict the changes in main control temperature and humidity values ​​during the optimization cycle; Predict the changes in temperature and humidity values ​​at each measuring point during the optimization period; Determine the specific circumstances of the total load demand in the large space; Determine the temperature and humidity values ​​of each sub-area in a large space.

9. A method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment according to claim 8, characterized in that, The specific circumstances under which the total load demand of a large space is determined refer to: If it is the first case, the range of adjustment of the operating parameters is too small to ensure that the main control temperature and humidity values ​​are close to the center value of the process requirements; if it is the second case, the range of adjustment of the operating parameters is too large to ensure that the main control temperature and humidity values ​​recover to the range of the process requirements more quickly.

10. A method for controlling temperature and humidity in multiple zones of a large-space air-conditioned environment according to claim 8, characterized in that, The determination of the temperature and humidity values ​​of each sub-region of the large space refers to: For sub-regions that are within the process requirements, the first case is adopted, where the allocated load only needs to offset local thermal and moisture disturbances and ensure that it remains at the center value of the process requirements. For sub-regions that are not within the process requirements, the second case is adopted, where the allocated load is given priority, including the load that offsets heat and moisture generation and ensures that it returns to the process requirements zone within 1 to 2 optimization cycles.

Images