Air conditioner energy-saving optimization method and system based on intelligent control

Abstract

The invention discloses an air conditioner energy-saving optimization method and system based on intelligent control, and the method comprises the following steps: 1, collecting temperature and humidity data and air conditioner setting parameters, carrying out the training through a residual error connection attention recurrent neural network model, and outputting a thermal inertia prediction value; 2, constructing a thermal behavior joint feature vector; 3, performing clustering operation by adopting a Dirichlet process mean value clustering algorithm, and generating a space-time microcell by setting a clustering penalty threshold value; 4, constructing a thermal comfort model by adopting an improved PMV model based on a Fanger heat balance theory, and obtaining a disturbance index value of each space-time microcell; 5, constructing a joint optimization objective function, and generating optimal control action parameters of each space-time microcell; and step 6, carrying out fusion calculation on the optimal control action parameters of all the space-time microcells to generate an overall control instruction. According to the method, thermal inertia prediction and dynamic clustering optimization methods are fused, and intelligent energy-saving control over the air conditioner is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy-saving optimization control of air-conditioning systems, and in particular to an air-conditioning energy-saving optimization method and system based on intelligent control. Background Art

[0002] With the growing demand for intelligent management of building energy consumption and indoor environments, optimized control technologies for air conditioning energy conservation and thermal comfort in multi-user, multi-space scenarios have attracted widespread attention. Existing air conditioning control systems often rely on rule-based control or local heat load models for parameter setting. However, these systems commonly suffer from the following problems in practical applications:

[0003] Current air-conditioning systems lack sufficient modeling of spatial thermal inertia and dynamic heat load characteristics, making it difficult to accurately predict the delayed response of air-conditioning setting changes to the evolution of temperature and humidity in each area, resulting in control lag and energy waste; traditional air-conditioning zoning methods are mostly based on physical structures or static area divisions, failing to dynamically respond to user behavior patterns and thermal environment changes, and unable to achieve sufficiently fine-grained spatiotemporal dynamic zoning; in terms of comfort control, existing methods mostly use fixed models or rule tables for adjustment, without considering the heterogeneity of users in different areas' sensitivity to temperature, humidity and wind speed factors, and lack the ability for refined intervention; in addition, most optimization algorithms use static objective functions, which cannot take into account the dynamic balance between energy saving and personalized comfort assurance, and the overall control strategy lacks a real-time perception-driven adaptive regulation mechanism.

[0004] Therefore, how to provide an air conditioning energy-saving optimization method and system based on intelligent control is a problem that those skilled in the art urgently need to solve. Summary of the Invention

[0005] One objective of the present invention is to propose an air conditioning energy-saving optimization method and system based on intelligent control. This method integrates thermal inertia prediction modeling, Dirichlet process clustering, a thermal comfort model constructed using an improved PMV model based on Fanger thermal balance theory, and an optimization algorithm to form a comprehensive air conditioning energy-saving control method. This method utilizes a residual-connected attention recurrent neural network model for training, divides spatiotemporal micro-regions, and quantifies disturbance sensitivity to achieve precise optimization of air conditioning setting parameters. Ultimately, global control instructions are generated based on the priority of the spatiotemporal micro-regions. This method offers significant energy savings, enhanced user comfort, and high adaptability, making it suitable for intelligent air conditioning control in a variety of complex spatial environments.

[0006] According to an embodiment of the present invention, an air conditioning energy-saving optimization method based on intelligent control includes the following steps:

[0007] Step 1: Collect temperature and humidity data and air conditioning setting parameters for each spatial node in the past preset time period, train them using a residual connection attention recurrent neural network model, and output the thermal inertia prediction value of each spatial node;

[0008] Step 2: Construct the joint feature vector of thermal behavior;

[0009] Step 3: Perform clustering operation based on the joint feature vector of thermal behavior using Dirichlet process mean clustering algorithm, and generate spatiotemporal micro-regions by setting cluster penalty threshold;

[0010] Step 4: A thermal comfort model is constructed using an improved PMV model based on Fanger thermal balance theory. Based on the current air conditioning setting parameters and the predicted thermal inertia, the thermal comfort model is used to calculate the sensitivity of the air conditioning setting parameter adjustment to changes in user comfort within a preset time period in the future, and the disturbance index value of each spatiotemporal micro-region is obtained;

[0011] Step 5: constructing a joint optimization objective function based on the energy consumption per unit time and the disturbance index value, solving the air conditioning setting parameters of each spatiotemporal micro-zone, and generating the optimal control action parameters of each spatiotemporal micro-zone;

[0012] Step 6: The optimal control action parameters of each spatiotemporal micro-zone are integrated and calculated, and the overall control instructions of the air-conditioning system are generated by using a priority allocation method based on the micro-zone area, heat load, and user activity.

[0013] Optionally, the step 1 is specifically as follows:

[0014] A temperature and humidity sensor and an air conditioning operating parameter collection interface are set for each spatial node to collect and record indoor temperature data, humidity data, and air conditioning setting parameters obtained within a preset time period in real time. The air conditioning setting parameters include air conditioning set temperature, wind speed level, air outlet angle, and operating mode;

[0015] The collected indoor temperature data, humidity data and air conditioning setting parameters are sorted into time series data in chronological order, wherein the time series data is composed of multiple time steps, each time step containing the temperature and humidity values ​​and air conditioning setting parameters at the current moment;

[0016] Constructing a residual connection attention recurrent neural network model, wherein the residual connection attention recurrent neural network model structure includes an input layer, an attention layer, a recurrent neural network layer, a residual connection layer and an output layer;

[0017] The input layer receives the time series data and normalizes and encodes the feature vector in each time step;

[0018] The attention layer is used to calculate the correlation between the historical time step and the current time step. The correlation is obtained by linearly mapping the feature vectors corresponding to each historical time step and the current time step to obtain a hidden representation. The attention weight of each historical time step is obtained by calculating the vector cosine similarity between the two hidden representations and performing normalization processing;

[0019] According to the attention weights, the feature vectors of all historical time steps are weighted and summed to form an attention-weighted feature sequence;

[0020] The recurrent neural network layer is used to input the attention-weighted feature sequence into the recurrent neural network structure to extract long-term dependency features and short-term fluctuation features in the time series;

[0021] The residual connection layer is used to perform feature splicing on the feature sequence after the attention weighting and the long-term dependency features and short-term fluctuation features output by the recurrent neural network layer to generate fusion features;

[0022] The output layer is used to receive the fused feature representation and input it into a regression structure composed of a fully connected neural network, and output the thermal inertia prediction value of the current spatial node, which is used to reflect the natural change trend of the space temperature in the short future period under the condition that the current air-conditioning setting parameters remain unchanged.

[0023] Optionally, the step 2: constructing a joint feature vector of thermal behavior is specifically as follows:

[0024] Collecting user behavior data, including user presence status, dwell time, movement trajectory, and frequency of entering and leaving the room, and constructing a user behavior feature vector;

[0025] Synchronize the user behavior data with the environmental perception data corresponding to the spatial node, the environmental perception data including the temperature data, humidity data, CO2 concentration, illumination intensity and air flow rate of the spatial node, and construct an environmental perception feature vector;

[0026] The user behavior feature vector and environmental perception feature vector of each spatial node within a preset time period are encoded and processed, and a thermal behavior joint feature vector is constructed based on the encoded feature data. The thermal behavior joint feature vector is used to comprehensively reflect the user behavior characteristics and thermal load status of the spatial node within the corresponding time period.

[0027] Optionally, the step three is specifically as follows:

[0028] The joint feature vector of thermal behavior constructed by each spatial node within a preset time period is composed into a feature sample set as clustering input data;

[0029] Initialize the cluster center set to an empty set and set a cluster penalty threshold, which is used to control the probability of generating new cluster centers;

[0030] For each thermal behavior joint feature vector, calculate the distance between it and all existing cluster centers in turn, and determine whether the distance to the current nearest cluster center exceeds the cluster penalty threshold;

[0031] If the distance to the current nearest cluster center does not exceed the cluster penalty threshold, the corresponding feature sample is classified into the category of the nearest cluster center, and the cluster center position is updated based on the mean value of the feature samples in the category;

[0032] If the distance to the current nearest cluster center exceeds the cluster penalty threshold, the corresponding feature sample is created as a new cluster center according to the idea of ​​the Dirichlet process, and a new category is constructed;

[0033] All feature samples are clustered and divided, and finally several category sets with similar thermal behavior joint features are output. The spatial node corresponding to each category set is defined as a spatiotemporal micro-region.

[0034] Optionally, the step 4 is specifically as follows:

[0035] The thermal comfort model adopts an improved PMV model based on Fanger thermal balance theory. The improvement of the improved PMV model lies in performing linear regression approximation processing and fitting the PMV model coefficients:

[0036] ;

[0037] in, represents the thermal comfort score, Indicates the air temperature, represents the equivalent thermoneutral temperature of the human body, Relative humidity, Indicates the air velocity, Represents the metabolic rate of the human body, represents the thermal resistance of clothing, 、 、 、 、 and is the PMV model coefficient obtained by fitting the experimental data;

[0038] Performing a preset finite amplitude perturbation on the current air conditioning setting parameters to construct multiple perturbation parameter sets, and combining the thermal inertia prediction value as the input of the thermal comfort model to obtain thermal comfort score values ​​corresponding to the perturbation, and simultaneously obtaining an initial thermal comfort score value under the current air conditioning setting parameter conditions;

[0039] The response degree of thermal comfort changes corresponding to each set of disturbance parameters is integrated and calculated to obtain the corresponding disturbance response sensitivity value:

[0040] ;

[0041] in, Indicates the The disturbance response sensitivity value corresponding to the disturbance, represents the future prediction time interval, Indicates the The thermal comfort score corresponding to the disturbance is: Indicates the initial thermal comfort score under the current air conditioning setting parameters;

[0042] All disturbance response sensitivity values ​​are normalized and combined with the thermal inertia prediction value of the spatiotemporal micro-region to calculate the disturbance index value corresponding to the spatiotemporal micro-region, which is used to measure the comfort sensitivity of the spatiotemporal micro-region under changes in air conditioning setting parameters.

[0043] Optionally, the joint optimization objective function constructed based on the energy consumption per unit time value and the disturbance index value is specifically:

[0044] The energy consumption per unit time of each spatiotemporal micro-zone under different air-conditioning setting parameter combinations is obtained, and the energy consumption per unit time is associated with the disturbance index value of the corresponding spatiotemporal micro-zone to construct a joint optimization objective function. The joint optimization objective function is achieved by weighted integration of the energy consumption per unit time of the spatiotemporal micro-zone and the disturbance index value. The goal is to minimize the energy consumption of the air-conditioning system while ensuring the stability of the user's thermal comfort response.

[0045] Optionally, the solution of the air conditioning setting parameters of each spatiotemporal micro-zone is specifically as follows:

[0046] Based on the joint optimization objective function of each spatiotemporal micro-region, a heuristic search algorithm is used to solve the air conditioning setting parameter space;

[0047] Initialize several sets of candidate solutions in the air conditioning parameter space and calculate the joint optimization objective function value corresponding to each set of candidate solutions as the basis for evaluating the quality of the control action;

[0048] Perform search and update operations based on the joint optimization objective function values ​​of each candidate solution, and gradually approach the minimum value of the joint optimization objective function through parameter mutation and iterative replacement of air conditioning settings;

[0049] When the decrease in the value of the optimal joint optimization objective function in several consecutive generations of the search process is less than the set decrease threshold, or reaches the set maximum number of search iterations, it is determined to be converged and the search process is terminated. The solution with the minimum joint optimization objective function is selected as the optimal control action parameter of the space-time micro-region.

[0050] Optionally, the step six is ​​specifically as follows:

[0051] Obtain the optimal control action parameters for each spatiotemporal micro-region and extract the corresponding micro-region area, heat load and user activity indicators;

[0052] Calculate the fusion weight for each spatiotemporal micro-region, where the fusion weight is determined by weighted calculation of the micro-region area, heat load, and user activity;

[0053] Using the priority allocation method, the spatiotemporal micro-region with a larger fusion weight has a higher corresponding priority;

[0054] Through weighted fusion, the optimal control action parameters of all spatiotemporal micro-regions are integrated according to the corresponding fusion weights to generate the overall control instructions of the air-conditioning system.

[0055] An air conditioning energy-saving optimization system based on intelligent control according to an embodiment of the present invention includes the following modules:

[0056] The data collection module is used to collect temperature and humidity data, air conditioning setting parameters, user behavior data and environmental perception data of each spatial node within the past preset time period;

[0057] The thermal inertia prediction module is used to build a residual connection attention recurrent neural network model, model and train the historical temperature and humidity data of spatial nodes and air conditioning setting parameters, and output the thermal inertia prediction value of the spatial node under the current air conditioning setting parameters;

[0058] Thermal behavior feature construction module, used to synchronize and encode user behavior features with the environmental perception data of spatial nodes to construct a thermal behavior joint feature vector;

[0059] A micro-area partitioning module is used to perform cluster analysis based on the thermal behavior joint feature vector using a Dirichlet process mean clustering algorithm, and dynamically partition and generate a number of spatiotemporal micro-areas according to a cluster penalty threshold;

[0060] The disturbance calculation module is used to construct an improved PMV model based on Fanger thermal balance theory. It combines the thermal inertia prediction value with the current air conditioning setting parameters to calculate the thermal comfort response changes of each spatiotemporal micro-area under disturbance and output the disturbance index value.

[0061] The optimization control solution module is used to construct a joint optimization objective function based on the unit time energy consumption value and the disturbance index value of each micro-zone, and use a heuristic search algorithm to solve the air conditioning setting parameter space to generate the optimal control action parameters for each spatiotemporal micro-zone;

[0062] The fusion allocation module is used to calculate the fusion weight based on the area, heat load and user activity of each micro-zone. It adopts a priority allocation strategy to weightedly fuse the optimal control action parameters of each spatiotemporal micro-zone to generate the overall control instructions of the air conditioning system;

[0063] The execution instruction output module is used to output the overall control instruction and send it to the air conditioning system for execution.

[0064] The beneficial effects of the present invention are:

[0065] The present invention uses the joint modeling of residual-connected attention recurrent neural network and Dirichlet process mean clustering algorithm to address the problems of spatial thermal inertia response delay, rough static area division and insufficient differentiated control of user comfort in air-conditioning energy-saving control. The thermal inertia prediction model is constructed by using time-series temperature and humidity data for spatial nodes and air-conditioning setting parameters. The attention mechanism is introduced to calculate time-dependent weights, and the residual structure is combined to alleviate the gradient degradation problem, thereby achieving accurate prediction of indoor thermal inertia change trends. In the micro-area division stage, based on the thermal behavior joint feature vector that integrates user behavior and environmental perception, the Dirichlet process mean clustering algorithm is used to adaptively determine the micro-area based on the feature distribution without relying on the preset number of clusters. Zone structure, dynamically generate high-resolution spatiotemporal micro-zone division; in the comfort evaluation link, construct an improved PMV model based on Fanger thermal balance theory, and quantify the disturbance index value through disturbance response sensitivity integration and normalization processing, which comprehensively reflects the comfort sensitivity of the micro-zone to the change of air-conditioning settings; finally, construct a joint optimization objective function based on the disturbance index value and the energy consumption per unit time, adopt a heuristic search algorithm to efficiently solve the air-conditioning setting parameter space, integrate the micro-zone area, heat load and user activity to perform weighted integration of the optimal control action parameters of each spatiotemporal micro-zone, and output the overall control instructions of the air-conditioning system. The present invention realizes personalized intelligent regulation of dynamic micro-zones, which significantly improves the energy saving level of air-conditioning while ensuring comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0067] Figure 1 This is an overall flow chart of an air conditioning energy-saving optimization method based on intelligent control proposed by the present invention;

[0068] Figure 2 This is a structural diagram of an air conditioning energy-saving optimization system based on intelligent control proposed by the present invention. DETAILED DESCRIPTION

[0069] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0070] refer to Figure 1 , an air conditioning energy-saving optimization method based on intelligent control, comprising the following steps:

[0071] Step 1: Collect temperature and humidity data and air conditioning setting parameters for each spatial node in the past preset time period, train them using a residual connection attention recurrent neural network model, and output the thermal inertia prediction value of each spatial node;

[0072] Step 2: Construct the joint feature vector of thermal behavior;

[0073] Step 3: Perform clustering operation based on the joint feature vector of thermal behavior using Dirichlet process mean clustering algorithm, and generate spatiotemporal micro-regions by setting cluster penalty threshold;

[0074] Step 4: A thermal comfort model is constructed using an improved PMV model based on Fanger thermal balance theory. Based on the current air conditioning setting parameters and the predicted thermal inertia, the thermal comfort model is used to calculate the sensitivity of the air conditioning setting parameter adjustment to changes in user comfort within a preset time period in the future, and the disturbance index value of each spatiotemporal micro-region is obtained;

[0075] Step 5: constructing a joint optimization objective function based on the energy consumption per unit time and the disturbance index value, solving the air conditioning setting parameters of each spatiotemporal micro-zone, and generating the optimal control action parameters of each spatiotemporal micro-zone;

[0076] Step 6: The optimal control action parameters of each spatiotemporal micro-zone are integrated and calculated, and the overall control instructions of the air-conditioning system are generated by using a priority allocation method based on the micro-zone area, heat load, and user activity.

[0077] In this embodiment, the step 1 is specifically as follows:

[0078] A temperature and humidity sensor and an air conditioning operating parameter collection interface are set for each spatial node to collect and record indoor temperature data, humidity data, and air conditioning setting parameters obtained at fixed time intervals within a preset time period in real time. The air conditioning setting parameters include air conditioning set temperature, wind speed level, air outlet angle, and operating mode;

[0079] Spatial nodes are basic units that divide a building's interior space into actual use areas or grids. Each spatial node corresponds to a specific physical location, such as a room or a small area. By deploying temperature and humidity sensors and air conditioning operating parameter collection interfaces at each spatial node, precise perception and independent control of local environmental conditions are achieved, providing foundational support for thermal inertia modeling and intelligent control.

[0080] The collected indoor temperature data, humidity data and air conditioning setting parameters are sorted into time series data in chronological order, wherein the time series data is composed of multiple time steps, each time step containing the temperature and humidity values ​​and air conditioning setting parameters at the current moment;

[0081] Constructing a residual connection attention recurrent neural network model, wherein the residual connection attention recurrent neural network model structure includes an input layer, an attention layer, a recurrent neural network layer, a residual connection layer and an output layer;

[0082] The input layer receives the time series data and normalizes and encodes the feature vector in each time step;

[0083] The attention layer is used to calculate the correlation between the historical time step and the current time step. The correlation is obtained by linearly mapping the feature vectors corresponding to each historical time step and the current time step to obtain a hidden representation. The attention weight of each historical time step is obtained by calculating the vector cosine similarity between the two hidden representations and performing normalization processing;

[0084] According to the attention weights, the feature vectors of all historical time steps are weighted and summed to form an attention-weighted feature sequence;

[0085] The recurrent neural network layer is used to input the attention-weighted feature sequence into the recurrent neural network structure to extract long-term dependency features and short-term fluctuation features in the time series;

[0086] The residual connection layer is used to perform feature splicing on the feature sequence after the attention weighting and the long-term dependency features and short-term fluctuation features output by the recurrent neural network layer to generate fusion features;

[0087] The output layer is used to receive the fused feature representation and input it into a regression structure composed of a fully connected neural network, and output the thermal inertia prediction value of the current spatial node, which is used to reflect the natural change trend of the space temperature in the short future period under the condition that the current air-conditioning setting parameters remain unchanged.

[0088] In this embodiment, the step 2: constructing a joint feature vector of thermal behavior is specifically as follows:

[0089] Collecting user behavior data, including user presence status, dwell time, movement trajectory, and frequency of entering and leaving the room, and constructing a user behavior feature vector;

[0090] Synchronize the user behavior data with the environmental perception data corresponding to the spatial node, the environmental perception data including the temperature data, humidity data, CO2 concentration, illumination intensity and air flow rate of the spatial node, and construct an environmental perception feature vector;

[0091] The user behavior feature vector and environmental perception feature vector of each spatial node within a preset time period are encoded and processed, and a thermal behavior joint feature vector is constructed based on the encoded feature data. The thermal behavior joint feature vector is used to comprehensively reflect the user behavior characteristics and thermal load status of the spatial node within the corresponding time period.

[0092] In this embodiment, the step three is specifically as follows:

[0093] The joint feature vector of thermal behavior constructed by each spatial node within a preset time period is composed into a feature sample set as clustering input data;

[0094] Initialize the cluster center set to an empty set and set a cluster penalty threshold, which is used to control the probability of generating new cluster centers;

[0095] For each thermal behavior joint feature vector, calculate the distance between it and all existing cluster centers in turn, and determine whether the distance to the current nearest cluster center exceeds the cluster penalty threshold;

[0096] If the distance to the current nearest cluster center does not exceed the cluster penalty threshold, the corresponding feature sample is classified into the category of the nearest cluster center, and the cluster center position is updated based on the mean value of the feature samples in the category;

[0097] If the distance to the current nearest cluster center exceeds the cluster penalty threshold, then according to the idea of ​​the Dirichlet process, instead of relying on the preset number of categories, the corresponding feature sample is directly created as a new cluster center and a new category is constructed, thus achieving dynamic expansion of the potential category space;

[0098] All feature samples are clustered and finally several category sets with similar thermal behavior joint features are output. The spatial node corresponding to each category set is defined as a spatiotemporal micro-region. The number of spatiotemporal micro-regions is not set manually, but is determined by the distribution of thermal behavior joint features and the clustering penalty threshold. The smaller the clustering penalty threshold, the more micro-regions are generated and the finer the clustering granularity.

[0099] The use of a Dirichlet process-based mean clustering algorithm for spatiotemporal micro-zoning offers significant advantages, including eliminating the need for a pre-set number of clusters, adapting to data distribution characteristics, and improving segmentation flexibility. Traditional clustering methods (such as K-means) typically require manual pre-determination of the number of clusters, making them difficult to adapt to the dynamic changes in user behavior and thermal environment across different areas and time periods within a building. This implementation, by incorporating the Dirichlet process concept and combining it with a dynamic control mechanism using a cluster penalty threshold, automatically determines whether to create new clusters based on the true distance relationship between features when processing joint feature vectors of thermal behavior, thereby dynamically expanding the latent class space. This approach avoids both excessive merging caused by insufficient clusters and computational complexity and control interference caused by an excessive number of clusters. Cluster granularity is controlled by a penalty threshold, allowing for fine-tuning of micro-zoning density based on scenario requirements. This enables adaptive thermal-aware spatial decomposition, improves subsequent air conditioning control accuracy and personalized adjustment capabilities, and ultimately achieves the intelligent optimization goal of balancing energy conservation and comfort.

[0100] In this embodiment, the step 4 is specifically as follows:

[0101] The thermal comfort model adopts an improved PMV model based on Fanger thermal balance theory. The improvement of the improved PMV model lies in performing linear regression approximation processing and fitting the PMV model coefficients:

[0102] ;

[0103] in, represents the thermal comfort score, Indicates the air temperature, represents the equivalent thermoneutral temperature of the human body, Relative humidity, Indicates the air velocity, Represents the metabolic rate of the human body, represents the thermal resistance of clothing, 、 、 、 、 and is the PMV model coefficient obtained by fitting the experimental data;

[0104] The PMV model, proposed by PO Fanger, is used to predict the average thermal sensation rating of a large number of people in a specific thermal environment. It estimates the difference between heat generation and heat loss by the human body through a complex heat balance equation, thereby inferring the subjective thermal sensation. To adapt to real-time computing and deep learning modeling, this paper simplifies the original complex PMV expression into a differentiable, linear combination form, and experimentally optimizes the PMV model coefficients, thereby simplifying the computational complexity and improving the efficiency of micro-area disturbance prediction.

[0105] Performing a preset finite amplitude perturbation on the current air conditioning setting parameters to construct multiple perturbation parameter sets, and combining the thermal inertia prediction value as the input of the thermal comfort model to obtain thermal comfort score values ​​corresponding to the perturbation, and simultaneously obtaining an initial thermal comfort score value under the current air conditioning setting parameter conditions;

[0106] The response degree of thermal comfort changes corresponding to each set of disturbance parameters is integrated and calculated to obtain the corresponding disturbance response sensitivity value:

[0107] ;

[0108] in, Indicates the The disturbance response sensitivity value corresponding to the disturbance, represents the future prediction time interval, Indicates the The thermal comfort score corresponding to the disturbance is: Indicates the initial thermal comfort score under the current air conditioning setting parameters;

[0109] All disturbance response sensitivity values ​​are normalized and combined with the thermal inertia prediction value of the spatiotemporal micro-region to calculate the disturbance index value corresponding to the spatiotemporal micro-region, which is used to measure the comfort sensitivity of the spatiotemporal micro-region under changes in air conditioning setting parameters.

[0110] In this embodiment, an improved PMV model based on Fanger's thermal balance theory is used to model thermal comfort, and the calculation of disturbance response sensitivity and disturbance index values ​​is introduced, which has significant practical value and engineering advantages. Although the traditional PMV model has a certain theoretical basis for thermal comfort prediction, its complex nonlinear thermal balance equation makes it difficult to directly integrate it into real-time optimization and intelligent control systems. To this end, the present invention improves the PMV model through linear regression approximation, simplifying it into a differentiable linear combination expression, and experimentally fitting various model parameters. This allows the thermal comfort model to retain the perceptual accuracy of the original model while significantly improving computational efficiency and embeddability. In addition, by applying disturbances to the air-conditioning setting parameters, constructing a disturbance parameter set, and combining the thermal inertia prediction results, the changes in the impact of air-conditioning adjustment on comfort in future time periods are dynamically simulated, and the disturbance response sensitivity is further calculated. After normalizing the sensitivity index, the disturbance index value corresponding to the spatiotemporal micro-region is calculated, which can more accurately reflect the differences in comfort response of each spatiotemporal micro-region to parameter changes, thereby providing a refined and quantifiable reference basis for energy-saving optimization and enhancing the system's regional adaptability and intelligent adjustment capabilities.

[0111] In this embodiment, the joint optimization objective function constructed based on the energy consumption per unit time value and the disturbance index value is specifically:

[0112] The energy consumption per unit time of each spatiotemporal micro-zone under different air-conditioning setting parameter combinations is obtained, and the energy consumption per unit time is associated with the disturbance index value of the corresponding spatiotemporal micro-zone to construct a joint optimization objective function. The joint optimization objective function is achieved by weighted integration of the energy consumption per unit time of the spatiotemporal micro-zone and the disturbance index value. The goal is to minimize the energy consumption of the air-conditioning system while ensuring the stability of the user's thermal comfort response.

[0113] In this embodiment, the air conditioning setting parameters of each spatiotemporal micro-zone are solved as follows:

[0114] Based on the joint optimization objective function of each spatiotemporal micro-region, a heuristic search algorithm is used to solve the air conditioning setting parameter space;

[0115] Initialize several sets of candidate solutions in the air conditioning parameter space and calculate the joint optimization objective function value corresponding to each set of candidate solutions as the basis for evaluating the quality of the control action;

[0116] Perform search and update operations based on the joint optimization objective function values ​​of each candidate solution, and gradually approach the minimum value of the joint optimization objective function through parameter mutation and iterative replacement of air conditioning settings;

[0117] When the decrease in the value of the optimal joint optimization objective function in several consecutive generations of the search process is less than the set decrease threshold, or reaches the set maximum number of search iterations, it is determined to be converged and the search process is terminated. The solution with the minimum joint optimization objective function is selected as the optimal control action parameter of the space-time micro-region.

[0118] In this embodiment, the step six is ​​specifically as follows:

[0119] Obtain the optimal control action parameters for each spatiotemporal micro-region and extract the corresponding micro-region area, heat load and user activity indicators;

[0120] A fusion weight is calculated for each spatiotemporal micro-zone. The fusion weight is determined by weighted calculation of the micro-zone area, heat load, and user activity. User activity is comprehensively evaluated based on the user's presence, duration of stay, and movement frequency within the spatiotemporal micro-zone. The heat load is calculated based on the temperature and humidity levels within the micro-zone and the external heat exchange conditions.

[0121] Using the priority allocation method, the spatiotemporal micro-region with a larger fusion weight has a higher corresponding priority;

[0122] Through weighted fusion, the optimal control action parameters of all spatiotemporal micro-regions are integrated according to the corresponding fusion weights to generate the overall control instructions of the air-conditioning system.

[0123] For example, the temperature in space-time micro-zone A is high because there are many people, so it is recommended to set it to 23℃. The temperature in space-time micro-zone B is low because there are few people, so it is recommended to set it to 26℃. After integration, the set temperature of the entire area is 24.5℃, and the wind speed and air supply direction are fine-tuned to make space-time micro-zone A feel a stronger air supply, while space-time micro-zone B feels a softer air supply, thus achieving a balance between energy saving and comfort.

[0124] refer to Figure 2 , an air conditioning energy-saving optimization system based on intelligent control, including the following modules:

[0125] The data collection module is used to collect temperature and humidity data, air conditioning setting parameters, user behavior data and environmental perception data of each spatial node within the past preset time period;

[0126] The thermal inertia prediction module is used to build a residual connection attention recurrent neural network model, model and train the historical temperature and humidity data of spatial nodes and air conditioning setting parameters, and output the thermal inertia prediction value of the spatial node under the current air conditioning setting parameters;

[0127] Thermal behavior feature construction module, used to synchronize and encode user behavior features with the environmental perception data of spatial nodes to construct a thermal behavior joint feature vector;

[0128] A micro-area partitioning module is used to perform cluster analysis based on the thermal behavior joint feature vector using a Dirichlet process mean clustering algorithm, and dynamically partition and generate a number of spatiotemporal micro-areas according to a cluster penalty threshold;

[0129] The disturbance calculation module is used to construct an improved PMV model based on Fanger thermal balance theory. It combines the thermal inertia prediction value with the current air conditioning setting parameters to calculate the thermal comfort response changes of each spatiotemporal micro-area under disturbance and output the disturbance index value.

[0130] The optimization control solution module is used to construct a joint optimization objective function based on the unit time energy consumption value and the disturbance index value of each micro-zone, and use a heuristic search algorithm to solve the air conditioning setting parameter space to generate the optimal control action parameters for each spatiotemporal micro-zone;

[0131] The fusion allocation module is used to calculate the fusion weight based on the area, heat load and user activity of each micro-zone. It adopts a priority allocation strategy to weightedly fuse the optimal control action parameters of each spatiotemporal micro-zone to generate the overall control instructions of the air conditioning system;

[0132] The execution instruction output module is used to output the overall control instruction and send it to the air conditioning system for execution.

[0133] Example 1:

[0134] To verify the feasibility of this invention, we applied it to the optimization of the central air conditioning system of a large commercial complex. The building, with a total area of ​​over 80,000 square meters, houses multiple functional areas including offices, shops, a cinema, and restaurants. The complex spatial structure, diverse user behaviors, and significant fluctuations in cooling and heating loads make it a typical application scenario for intelligent energy-saving control technology.

[0135] During the deployment phase, the entire building space was divided into several spatial nodes based on function and structure. Temperature and humidity sensors, CO2 concentration detectors, and user behavior recognition equipment were deployed at each node to collect data including temperature, humidity, air conditioning settings, and occupant presence, movement trajectories, and dwell time. This data was updated every five minutes, and a rolling 48-hour time series was constructed. This was used to train a residual-connected attention recurrent neural network model to predict the thermal inertia response of spatial nodes.

[0136] User behavior characteristics and environmental perception characteristics are synchronously encoded to construct a joint thermal behavior feature vector. This feature vector is then dynamically partitioned using the Dirichlet process mean clustering algorithm, ultimately forming multiple spatiotemporal micro-regions with similar thermal behavior characteristics. Each micro-region is then fed into the improved PMV thermal comfort model, and the sensitivity of the response to comfort changes under different parameter perturbations is calculated to generate a disturbance index value.

[0137] A joint optimization objective function is then constructed, with the optimization goals of minimizing energy consumption and comfort disturbances. A heuristic search algorithm is then used to find the optimal control action within the air conditioning parameter space. Finally, the optimal control parameters for each micro-zone are integrated based on micro-zone area, heat load, and user activity to generate unified air conditioning system control instructions.

[0138] Table 1 Comparison of energy consumption per unit time and thermal comfort disturbance before and after system deployment

[0139]

[0140] It can be seen from the data in Table 1 above that after deploying the method of the present invention, the energy consumption and comfort performance of the air-conditioning system at different ambient temperatures are significantly improved.

[0141] From the perspective of average daily energy consumption, as the ambient temperature gradually increased (from 28.5°C to 36.6°C), the system's energy consumption before deployment continued to rise, from 7300.2 kWh to 8678.1 kWh, an increase of 18.8%. After deployment, although energy consumption also increased with the ambient temperature (from 6401.8 kWh to 7542.5 kWh), the overall level remained significantly lower than before deployment, demonstrating the system's strong energy-saving and control capabilities. The energy saving rate remained stable between 12.3% and 13.2%, demonstrating that the present invention can maintain stable energy savings even in high-temperature load scenarios.

[0142] In terms of the thermal comfort disturbance index, the average disturbance index gradually increased from 0.145 to 0.205 before system deployment, but significantly decreased after deployment, ranging from 0.090 to 0.118, with smaller fluctuations. This demonstrates that the present invention, through disturbance index modeling and optimized control, significantly improves the dynamic response capability of the air conditioning system, effectively suppressing indoor temperature and humidity fluctuations and enhancing user thermal comfort.

[0143] This embodiment achieves refined intelligent control of the air conditioning system by introducing thermal inertia prediction, user behavior modeling, and spatiotemporal microzoning, combined with disturbance index-based thermal comfort evaluation and energy consumption trade-off optimization. Compared to traditional extensive control methods that rely on global temperature settings, this invention can dynamically adjust setting parameters based on local thermal load changes and user behavior characteristics, significantly improving the system's response sensitivity and environmental adaptability. Furthermore, by jointly optimizing the objective function to guide the evolution of the air conditioning control strategy, it not only effectively reduces overall energy consumption, but also ensures the consistency and stability of the user's thermal comfort experience, demonstrating excellent energy efficiency and practicality.

[0144] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An air conditioning energy-saving optimization method based on intelligent control, characterized in that: The steps include: Step 1: Collect temperature and humidity data and air conditioning setting parameters for each spatial node in the past preset time period, train them using a residual connection attention recurrent neural network model, and output the thermal inertia prediction value of each spatial node; Step 2: Construct the joint feature vector of thermal behavior; Step 3: Perform clustering operation based on the joint feature vector of thermal behavior using Dirichlet process mean clustering algorithm, and generate spatiotemporal micro-regions by setting cluster penalty threshold; Step 4: A thermal comfort model is constructed using an improved PMV model based on Fanger thermal balance theory. Based on the current air conditioning setting parameters and the predicted thermal inertia, the thermal comfort model is used to calculate the sensitivity of the air conditioning setting parameter adjustment to changes in user comfort within a preset time period in the future, and the disturbance index value of each spatiotemporal micro-region is obtained; Step 5: constructing a joint optimization objective function based on the energy consumption per unit time and the disturbance index value, solving the air conditioning setting parameters of each spatiotemporal micro-zone, and generating the optimal control action parameters of each spatiotemporal micro-zone; Step 6: The optimal control action parameters of each spatiotemporal micro-zone are integrated and calculated, and the overall control instructions of the air-conditioning system are generated by using a priority allocation method based on the micro-zone area, heat load, and user activity.

2. The air conditioning energy saving optimization method based on intelligent control according to claim 1 is characterized in that: The step 1 is specifically as follows: A temperature and humidity sensor and an air conditioning operating parameter collection interface are set for each spatial node to collect and record indoor temperature data, humidity data, and air conditioning setting parameters obtained within a preset time period in real time. The air conditioning setting parameters include air conditioning set temperature, wind speed level, air outlet angle, and operating mode; The collected indoor temperature data, humidity data and air conditioning setting parameters are sorted into time series data in chronological order, wherein the time series data is composed of multiple time steps, each time step containing the temperature and humidity values ​​and air conditioning setting parameters at the current moment; Constructing a residual connection attention recurrent neural network model, wherein the residual connection attention recurrent neural network model structure includes an input layer, an attention layer, a recurrent neural network layer, a residual connection layer and an output layer; The input layer receives the time series data and normalizes and encodes the feature vector in each time step; The attention layer is used to calculate the correlation between the historical time step and the current time step. The correlation is obtained by linearly mapping the feature vectors corresponding to each historical time step and the current time step to obtain a hidden representation. The attention weight of each historical time step is obtained by calculating the vector cosine similarity between the two hidden representations and performing normalization processing; According to the attention weights, the feature vectors of all historical time steps are weighted and summed to form an attention-weighted feature sequence; The recurrent neural network layer is used to input the attention-weighted feature sequence into the recurrent neural network structure to extract long-term dependency features and short-term fluctuation features in the time series; The residual connection layer is used to perform feature splicing on the feature sequence after the attention weighting and the long-term dependency features and short-term fluctuation features output by the recurrent neural network layer to generate fusion features; The output layer is used to receive the fused feature representation and input it into a regression structure composed of a fully connected neural network, and output the thermal inertia prediction value of the current spatial node, which is used to reflect the natural change trend of the space temperature in the short future period under the condition that the current air-conditioning setting parameters remain unchanged.

3. The air conditioning energy saving optimization method based on intelligent control according to claim 1, characterized in that: The second step is to construct a joint feature vector of thermal behavior, specifically: Collecting user behavior data, including user presence status, dwell time, movement trajectory, and frequency of entering and leaving the room, and constructing a user behavior feature vector; Synchronize the user behavior data with the environmental perception data corresponding to the spatial node, the environmental perception data including the temperature data, humidity data, CO2 concentration, illumination intensity and air flow rate of the spatial node, and construct an environmental perception feature vector; The user behavior feature vector and environmental perception feature vector of each spatial node within a preset time period are encoded and processed, and a thermal behavior joint feature vector is constructed based on the encoded feature data. The thermal behavior joint feature vector is used to comprehensively reflect the user behavior characteristics and thermal load status of the spatial node within the corresponding time period.

4. The air conditioning energy saving optimization method based on intelligent control according to claim 1, characterized in that: The step three is specifically as follows: The joint feature vector of thermal behavior constructed by each spatial node within a preset time period is composed into a feature sample set as clustering input data; Initialize the cluster center set to an empty set and set a cluster penalty threshold, which is used to control the probability of generating new cluster centers; For each thermal behavior joint feature vector, calculate the distance between it and all existing cluster centers in turn, and determine whether the distance to the current nearest cluster center exceeds the cluster penalty threshold; If the distance to the current nearest cluster center does not exceed the cluster penalty threshold, the corresponding feature sample is classified into the category of the nearest cluster center, and the cluster center position is updated based on the mean value of the feature samples in the category; If the distance to the current nearest cluster center exceeds the cluster penalty threshold, the corresponding feature sample is created as a new cluster center according to the idea of ​​the Dirichlet process, and a new category is constructed; All feature samples are clustered and divided, and finally several category sets with similar thermal behavior joint features are output. The spatial node corresponding to each category set is defined as a spatiotemporal micro-region.

5. The air conditioning energy saving optimization method based on intelligent control according to claim 1, characterized in that: The step 4 is specifically as follows: The thermal comfort model adopts an improved PMV model based on Fanger thermal balance theory. The improvement of the improved PMV model lies in performing linear regression approximation processing and fitting the PMV model coefficients: ； in, represents the thermal comfort score, Indicates the air temperature, represents the equivalent thermoneutral temperature of the human body, Relative humidity, Indicates the air flow rate, Represents the metabolic rate of the human body, represents the thermal resistance of clothing, 、 、 、 、 and is the PMV model coefficient obtained by fitting the experimental data; Performing a preset finite amplitude perturbation on the current air conditioning setting parameters to construct multiple perturbation parameter sets, and combining the thermal inertia prediction value as the input of the thermal comfort model to obtain thermal comfort score values ​​corresponding to the perturbation, and simultaneously obtaining an initial thermal comfort score value under the current air conditioning setting parameter conditions; The response degree of thermal comfort changes corresponding to each set of disturbance parameters is integrated and calculated to obtain the corresponding disturbance response sensitivity value: ； in, Indicates the The disturbance response sensitivity value corresponding to the disturbance, represents the future prediction time interval, Indicates the The thermal comfort score corresponding to the disturbance is: Indicates the initial thermal comfort score under the current air conditioning setting parameters; All disturbance response sensitivity values ​​are normalized and combined with the thermal inertia prediction value of the spatiotemporal micro-region to calculate the disturbance index value corresponding to the spatiotemporal micro-region, which is used to measure the comfort sensitivity of the spatiotemporal micro-region under changes in air conditioning setting parameters.

6. The air conditioning energy saving optimization method based on intelligent control according to claim 1, characterized in that: The joint optimization objective function constructed based on the energy consumption per unit time value and the disturbance index value is specifically: The energy consumption per unit time of each spatiotemporal micro-zone under different air-conditioning setting parameter combinations is obtained, and the energy consumption per unit time is associated with the disturbance index value of the corresponding spatiotemporal micro-zone to construct a joint optimization objective function. The joint optimization objective function is achieved by weighted integration of the energy consumption per unit time of the spatiotemporal micro-zone and the disturbance index value. The goal is to minimize the energy consumption of the air-conditioning system while ensuring the stability of the user's thermal comfort response.

7. The air conditioning energy saving optimization method based on intelligent control according to claim 1, characterized in that: The specific solution for the air conditioning setting parameters of each time-space micro-zone is: Based on the joint optimization objective function of each spatiotemporal micro-region, a heuristic search algorithm is used to solve the air conditioning setting parameter space; Initialize several sets of candidate solutions in the air conditioning parameter space and calculate the joint optimization objective function value corresponding to each set of candidate solutions as the basis for evaluating the quality of the control action; Perform search and update operations based on the joint optimization objective function values ​​of each candidate solution, and gradually approach the minimum value of the joint optimization objective function through parameter mutation and iterative replacement of air conditioning settings; When the decrease in the value of the optimal joint optimization objective function in several consecutive generations of the search process is less than the set decrease threshold, or reaches the set maximum number of search iterations, it is determined to be converged and the search process is terminated. The solution with the minimum joint optimization objective function is selected as the optimal control action parameter of the space-time micro-region.

8. The air conditioning energy saving optimization method based on intelligent control according to claim 1 is characterized in that: The step six is ​​specifically as follows: Obtain the optimal control action parameters for each spatiotemporal micro-region and extract the corresponding micro-region area, heat load and user activity indicators; Calculate the fusion weight for each spatiotemporal micro-region, where the fusion weight is determined by weighted calculation of the micro-region area, heat load, and user activity; Using the priority allocation method, the spatiotemporal micro-region with a larger fusion weight has a higher corresponding priority; Through weighted fusion, the optimal control action parameters of all spatiotemporal micro-regions are integrated according to the corresponding fusion weights to generate the overall control instructions of the air-conditioning system.

9. An air conditioning energy-saving optimization system based on intelligent control, executing an air conditioning energy-saving optimization method based on intelligent control according to any one of claims 1 to 8, characterized in that: Includes the following modules: The data collection module is used to collect temperature and humidity data, air conditioning setting parameters, user behavior data and environmental perception data of each spatial node within the past preset time period; The thermal inertia prediction module is used to build a residual connection attention recurrent neural network model, model and train the historical temperature and humidity data of spatial nodes and air conditioning setting parameters, and output the thermal inertia prediction value of the spatial node under the current air conditioning setting parameters; Thermal behavior feature construction module, used to synchronize and encode user behavior features with the environmental perception data of spatial nodes to construct a thermal behavior joint feature vector; A micro-area partitioning module is used to perform cluster analysis based on the thermal behavior joint feature vector using a Dirichlet process mean clustering algorithm, and dynamically partition and generate a number of spatiotemporal micro-areas according to a cluster penalty threshold; The disturbance calculation module is used to construct an improved PMV model based on Fanger thermal balance theory. It combines the thermal inertia prediction value with the current air conditioning setting parameters to calculate the thermal comfort response changes of each spatiotemporal micro-area under disturbance and output the disturbance index value. The optimization control solution module is used to construct a joint optimization objective function based on the unit time energy consumption value and the disturbance index value of each micro-zone, and use a heuristic search algorithm to solve the air conditioning setting parameter space to generate the optimal control action parameters for each spatiotemporal micro-zone; The fusion allocation module is used to calculate the fusion weight based on the area, heat load and user activity of each micro-zone. It adopts a priority allocation strategy to weightedly fuse the optimal control action parameters of each spatiotemporal micro-zone to generate the overall control instructions of the air conditioning system; The execution instruction output module is used to output the overall control instruction and send it to the air conditioning system for execution.

Images