A commercial building air conditioning load intelligent scheduling control method and system

Abstract

The present application provides a method and system for intelligent scheduling and control of air-conditioning loads in commercial buildings, which relates to the field of data mining and analysis technology. The method includes: based on sensor modules deployed in each building area space in the commercial building, respectively collecting corresponding sensor load time series data; constructing a building sensor graph structure based on each of the sensor load time series data; processing the building sensor graph structure based on a time series graph attention network to update the node features of each of the graph nodes; processing the node features of each updated graph node based on an air-conditioning parameter optimization model to correspondingly determine the target air-conditioning operating parameter group for each building area space in a preset time period in the future; the air-conditioning parameter optimization model adopts a reinforcement learning model. Thus, through the combination of the time series graph attention network and the reinforcement learning model, intelligent scheduling of building air-conditioning loads is achieved, significantly improving the energy efficiency management level and air-conditioning comfort experience of commercial buildings.

Description

Technical Field

[0001] The present application relates to the technical field of data mining and analysis, and in particular to a method and system for intelligent scheduling and control of air-conditioning loads in commercial buildings. Background Art

[0002] With global warming and growing energy demand, building energy consumption is becoming increasingly prominent. According to relevant research data, air conditioning systems account for more than 40% of commercial building energy consumption, making them a key target for building energy management.

[0003] Currently, air conditioning load scheduling and control in commercial buildings mainly relies on timing control technology or manual adjustments by building managers.

[0004] In timing control technology, air conditioning systems are typically turned on and off according to a pre-set schedule. While this approach is simple and easy to implement, it lacks the ability to perceive actual environmental changes and often fails to adjust to real-time load demands. For example, during high summer temperatures, building load demands can fluctuate significantly over time. Timing control cannot respond to these changes in a timely manner, causing the air conditioning system to maintain high power during unnecessary periods, resulting in energy waste.

[0005] Furthermore, manual adjustments based on experience often rely on the subjective judgment of building managers, who manually adjust the air conditioning system based on actual conditions. However, the efficiency and accuracy of manual adjustments are limited by the manager's experience and reaction time, making it difficult to maintain optimal operation in complex and changing environments. This is especially true in commercial buildings with multiple tenants or multi-functional areas, where load demands vary significantly. A single adjustment strategy struggles to meet the comfort requirements of each area, potentially leading to overcooling or overheating in some areas, thus impacting user comfort.

[0006] To address the above issues, the industry has not yet proposed a better technical solution. Summary of the Invention

[0007] The present application provides a method, system, storage medium, computer program product and electronic device for intelligent scheduling and control of air-conditioning load in commercial buildings, which are used to at least solve the problem that timed control or manual adjustment of the air-conditioning system cannot quickly match the load requirements of different areas in the commercial building.

[0008] In the first aspect, an embodiment of the present application provides an intelligent scheduling and control method for air-conditioning loads in commercial buildings, comprising: based on sensor modules deployed in each building area space in the commercial building, respectively collecting corresponding sensor load time series data; the sensor load time series data includes multiple historical time steps and corresponding multimodal environmental sensor data and regional air-conditioning loads; based on each of the sensor load time series data, constructing a building sensor graph structure; the building sensor graph structure includes multiple graph nodes and corresponding edge connections; each of the graph nodes uniquely corresponds to a building area space, and the node features of the graph nodes are determined by the sensor load time series data of the corresponding building area space; the edge weight of each edge connection is determined according to the environmental similarity, spatial connectivity path and air-conditioning load correlation corresponding to the connected graph node; processing the building sensor graph structure based on a time-series graph attention network to update the node features of each of the graph nodes; the time-series graph attention network includes a cascaded LSTM (Long Short-Term Memory, long short-term memory network) model module and graph attention model module; based on the air-conditioning parameter optimization model, the node features of each updated graph node are processed to correspondingly determine the target air-conditioning operation parameter group of each building area space in the future preset time period; the air-conditioning parameter optimization model adopts a reinforcement learning model, and the state of the air-conditioning parameter optimization model is defined by the node features of the updated graph nodes corresponding to each building area space, the action of the air-conditioning parameter optimization model is defined according to the adjustment amount of the air-conditioning operation parameter group for each building area space, and the reward of the air-conditioning parameter optimization model is defined according to the optimization degree of the global air-conditioning load and global environmental comfort of the commercial building.

[0009] In the second aspect, an embodiment of the present application provides an intelligent scheduling and control system for air-conditioning loads in commercial buildings, comprising: a data acquisition unit for respectively collecting corresponding sensor load time series data based on sensor modules deployed in each building area space in the commercial building; the sensor load time series data comprises multiple historical time steps and corresponding multimodal environmental sensor data and regional air-conditioning loads; a graph structure construction unit for constructing a building sensor graph structure based on each of the sensor load time series data; the building sensor graph structure comprises multiple graph nodes and corresponding edge connections; each of the graph nodes uniquely corresponds to a building area space, and the node features of the graph nodes are respectively determined by the sensor load time series data of the corresponding building area space; the edge weight of each edge connection is respectively determined according to the environmental similarity, spatial connectivity path and air-conditioning load correlation corresponding to the connected graph node; the graph node is more A new unit is provided for processing the building sensor graph structure based on a time-series graph attention network to update the node features of each of the graph nodes; the time-series graph attention network includes a cascaded LSTM model module and a graph attention model module; a parameter optimization unit is provided for processing the node features of each updated graph node based on an air-conditioning parameter optimization model to correspondingly determine the target air-conditioning operation parameter group of each building area space in a preset time period in the future; the air-conditioning parameter optimization model adopts a reinforcement learning model, and the state of the air-conditioning parameter optimization model is defined by the node features of the updated graph nodes corresponding to each building area space, and the action of the air-conditioning parameter optimization model is defined according to the adjustment amount of the air-conditioning operation parameter group for each building area space, and the reward of the air-conditioning parameter optimization model is defined according to the degree of optimization of the global air-conditioning load and global environmental comfort of the commercial building.

[0010] In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the intelligent scheduling and control method for commercial building air-conditioning loads of any embodiment of the present application.

[0011] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the intelligent scheduling and control method for commercial building air-conditioning loads of any embodiment of the present application are implemented.

[0012] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the intelligent scheduling and control method for commercial building air-conditioning loads of any embodiment of the present application.

[0013] The intelligent scheduling and control method for air-conditioning load in commercial buildings provided by this application can produce at least the following technical effects:

[0014] (1) By deploying sensor modules in various areas of commercial buildings, multimodal environmental data and air conditioning load data can be collected in real time, and then a time series graph structure is constructed to form a global and dynamic building sensor network that can accurately reflect environmental changes and air conditioning load requirements in each area. Compared with traditional timing control methods, through real-time perception and analysis, real-time scheduling can be performed according to actual load requirements in different time periods, avoiding excessive or insufficient operation of air conditioning due to environmental changes, thereby reducing energy waste.

[0015] (2) The building sensor graph structure associates the environmental data and load data of each building area as graph nodes. By comprehensively considering environmental similarity, spatial connectivity paths, and load correlation through edge weights, it realizes the perception of the differentiated needs of different areas. Based on this, an independent air conditioning operating parameter group can be generated for each area, avoiding the problem of "single adjustment strategy" in traditional manual adjustment methods that is difficult to achieve balanced scheduling in multiple areas, thereby improving comfort and energy saving effects throughout the entire building.

[0016] (3) By introducing the temporal graph attention network, this solution can adapt to complex dynamic changes in the building environment, including but not limited to seasonal temperature fluctuations and changes in crowd density, and can dynamically update the node features of each area. Since the temporal graph attention network can adaptively adjust the attention paid to various types of input data at different time steps, it can ensure that the air conditioning system always operates in an optimal manner under different environmental conditions, avoiding energy efficiency losses caused by manual adjustment due to insufficient human experience or slow response.

[0017] (4) Air conditioning parameter optimization is performed based on a reinforcement learning model, where the state is defined by the updated node features of each area of ​​the building, and the action controls the load distribution by adjusting the air conditioning operating parameter group. The reinforcement learning model continuously optimizes the balance between the global air conditioning load and environmental comfort, ensuring user comfort while minimizing energy consumption. As a result, in the complex and changing commercial building environment, the air conditioning operation strategy of each area can be dynamically adjusted, so that the global air conditioning system always operates in an optimized state.

[0018] Through this technical solution, by combining the time-series graph attention network and the reinforcement learning model, intelligent scheduling of building air-conditioning loads is achieved. It can control air-conditioning operating parameters in real time, accurately, and regionally, and maximize energy-saving effects through global optimization, significantly improving the energy efficiency management level and air-conditioning comfort experience of commercial buildings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 A flowchart illustrating an example of a method for intelligent scheduling and control of air-conditioning loads in commercial buildings according to an embodiment of the present application is shown;

[0021] Figure 2 A schematic diagram showing an example of a state transition action in a reinforcement learning model;

[0022] Figure 3 A schematic diagram showing a structural connection of an example of an air conditioning parameter optimization model according to an embodiment of the present application is shown;

[0023] Figure 4 A schematic diagram showing the structural connection of an example of a graph attention model module according to an embodiment of the present application is shown;

[0024] Figure 5 A structural block diagram of an example of an intelligent dispatching control system for air-conditioning loads in commercial buildings according to an embodiment of the present application is shown;

[0025] Figure 6 This is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0026] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0027] In the technical solutions of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.

[0028] Figure 1 A flowchart of an example of a method for intelligent scheduling and control of air-conditioning loads in commercial buildings according to an embodiment of the present application is shown.

[0029] Regarding the execution entity of the method of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities. By introducing real-time acquisition of multimodal data, dynamic construction of time series graph structure, deep learning of graph attention network and global optimization of reinforcement learning, the intelligence level and energy utilization efficiency of commercial building air-conditioning systems are significantly improved.

[0030] In some examples, it can be a commercial building air conditioning load scheduling platform, and can be integrated and configured in an electronic device or terminal through software, hardware, or a combination of software and hardware, and the type of terminal or electronic device can be diverse, such as a mobile phone, tablet computer, or desktop computer, etc.

[0031] like Figure 1 As shown, in step S110, based on the sensor modules deployed in the space of each building area in the commercial building, the corresponding sensor load time series data are collected respectively. The sensor load time series data includes multiple historical time steps and corresponding multimodal environmental sensor data and regional air-conditioning loads.

[0032] In some embodiments, multimodal environmental sensor modules are deployed in various areas of a commercial building. These sensors may include temperature sensors, humidity sensors, CO2 concentration sensors, light intensity sensors, personnel flow detectors, etc. Each sensor module continuously collects environmental data within the area it covers. Accordingly, the parameter types of the multimodal environmental sensor data include temperature, humidity, CO2 concentration, light intensity, and personnel activity density. At the same time, the air conditioning energy consumption of each area is monitored in real time through the air conditioning load sensor, and the operating parameters of the air conditioning system (such as power, wind speed, outlet temperature, etc.) are obtained. The collected sensor load time series data is arranged in chronological order, including data for multiple historical time steps or historical sampling moments, to reflect the environmental change trend and air conditioning load history of each area. In some examples, the air conditioning parameter type for the platform to perform load scheduling calculations, that is, the parameter type of the air conditioning operation parameter group, can be selected, which may specifically include supply air temperature, supply air volume, wind speed, and air conditioning switching time.

[0033] In step S120 , a building sensor graph structure is constructed based on the time series data of each sensor load.

[0034] The building sensor graph structure consists of multiple graph nodes and corresponding edge connections. Each graph node uniquely corresponds to a building area, and its node characteristics are determined by the time series data of the sensor load in the corresponding building area. The edge weight of each edge connection is determined based on the environmental similarity, spatial connectivity path, and air conditioning load correlation corresponding to the connected graph node.

[0035] Specifically, each building area is considered a node in the graph, such as each floor or each room. The node characteristics are determined by the sensor load time series data of the corresponding area. Each node feature contains the environmental data and air conditioning load data of the area at different time steps, which can reflect the historical load and environmental dynamics of the area. In addition, the edge weights between nodes are determined based on the environmental similarities between areas (such as similarities in temperature and humidity), spatial connectivity paths (such as adjacent or interconnected areas, ventilation duct connections, etc.), and air conditioning load correlation (for example, air conditioning load fluctuations in certain areas may affect adjacent areas). In this way, the building sensor graph structure can intuitively reflect the relationship and influence between areas. By establishing the connection relationship between nodes and edges, it is possible to identify potential collaborative control opportunities between areas, thereby optimizing the overall operating efficiency of the air conditioning system.

[0036] In step S130 , the building sensor graph structure is processed based on a temporal graph attention network to update the node features of each graph node. The temporal graph attention network includes a cascaded LSTM model module and a graph attention model module.

[0037] Here, a Spatio-Temporal Graph Attention Network (ST-GAT) is used to process the constructed building sensor graph structure. First, the LSTM model module is used to process the sensor load time series data of each graph node to capture the dependencies and dynamic change characteristics in the time series. Next, the graph attention model module is used to process the graph structure, calculate the attention weights between each node and its neighboring nodes, and thus update the features of each node. This allows dynamic attention to be paid to neighboring nodes that have a significant impact on the current node's features, so that the updated node features not only contain the historical information of the original time series data, but also integrate the associated information of the neighboring areas, improving the ability to express spatiotemporal features.

[0038] In step S140 , the node features of each updated graph node are processed based on the air-conditioning parameter optimization model to correspondingly determine a target air-conditioning operating parameter group for each building area space in a future preset time period.

[0039] Here, the air conditioning parameter optimization model can adopt various types of machine learning models, such as a multi-objective optimization model, a genetic algorithm, etc. In some examples of the embodiments of the present application, the air conditioning parameter optimization model adopts a reinforcement learning model.

[0040] Figure 2 A schematic diagram showing an example of state transition actions in a reinforcement learning model.

[0041] like Figure 2As shown, it involves actions corresponding to state transitions in the state space of multiple states S1 to Sn. For example, a1 represents the action of transitioning from state S1 to S2, a2 ​​represents the action of transitioning from S2 to S1, a3 represents the state transition from S1 to S3, and so on. Here, corresponding state transitions can occur based on policies, and each state transition policy can be used to generate different transitions. For example, based on the transition policy for state S1, action a2 or a3 can occur.

[0042] It should be noted that the range of states that a state can transfer to (also called transferable states) may be limited or conditional. For example, none of S1 to S3 will transfer to S4 to Sn, while the states that state S1 can transfer to are S2 and S3, and so on.

[0043] In some embodiments, each action has a corresponding action reward, and each action reward can be determined based on a preset reward function. Generally, the larger the transfer reward, the more valuable the transferred action is, and the system will prioritize executing this action. For example, if the reward corresponding to action a1 is greater than the reward corresponding to a3, then the transferred action a1 is more valuable.

[0044] In some examples of the embodiments of the present application, the state of the air-conditioning parameter optimization model is defined by the node features of the updated graph nodes corresponding to each building area space, so that the state variables of the reinforcement learning model can contain the historical environmental data of the area, the current air-conditioning load, and the association information with other areas. The actions of the air-conditioning parameter optimization model are defined according to the adjustment amount of the air-conditioning operating parameter group for each building area space, so that through learning and optimization, it is determined how each area should adjust the air-conditioning operating parameters (such as temperature setting, wind speed adjustment, etc.) in the future time period. The reward of the air-conditioning parameter optimization model is defined according to the degree of optimization of the global air-conditioning load and global environmental comfort of the commercial building. The model continuously adjusts and optimizes the air-conditioning operating parameters of each area through state transition actions, so that the entire building achieves the optimal balance between energy saving and comfort.

[0045] Through the embodiments of the present application, based on the state migration principle of the reinforcement learning model, the air-conditioning parameter optimization model can adaptively adjust the air-conditioning operating parameters to cope with environmental changes and load fluctuations. By comprehensively considering the global load and comfort, the air-conditioning loads in different areas of the building can be differentiated and controlled. Under the premise of ensuring user comfort, the energy-saving effect of the air-conditioning system can be maximized, thereby improving the operating efficiency and reliability of the overall air-conditioning system.

[0046] In some examples of the embodiments of the present application, the air-conditioning parameter optimization model adopts a deep Q-network (DQN).

[0047] Figure 3 A structural connection diagram of an example of an air-conditioning parameter optimization model according to an embodiment of the present application is shown.

[0048] like Figure 3 As shown, the air conditioning parameter optimization model 300 includes an input layer 310 , a hidden layer 320 and an output layer 330 .

[0049] Input layer 310 is used to receive the state vector corresponding to each updated graph node's node features. The updated graph node features have integrated environmental data, air conditioning load, and information related to other regions, ensuring that the air conditioning parameter optimization model can fully receive state information from each region. Furthermore, to ensure input validity and efficiency, the state vectors are normalized to reduce magnitude differences between features, thereby improving the model's convergence speed and accuracy.

[0050] The hidden layer 320 is used to process the state vector through a multi-layer neural network to extract at least one state-action mapping relationship, thereby obtaining a corresponding plurality of potential actions.

[0051] In some embodiments, the hidden layer is composed of a multi-layer neural network, such as a fully connected layer structure. Each layer of neurons processes the input data using a nonlinear activation function (e.g., ReLU, Sigmoid, etc.) to extract complex feature relationships. Through the hidden layer, useful features are extracted from the input state vector, capturing complex state-action relationships, thereby providing a precise strategic basis for the optimal control of the air conditioning system.

[0052] The output layer 330 is used to determine the Q value corresponding to each potential action, and determine the target air conditioning operating parameter group according to the air conditioning operating parameter group adjustment amount corresponding to the potential action with the maximum Q value.

[0053] Here, the output layer 330 is used to evaluate the potential actions extracted by the hidden layer and calculate the Q value corresponding to each action. The Q value is defined based on the expected cumulative reward that can be obtained by taking the corresponding potential action under the state vector.

[0054] Specifically, in the output layer, by calculating the Q value of each potential action, and then sorting them according to the Q value of each potential action, the action with the largest Q value is selected, and the adjustment amount of the air-conditioning operation parameter group corresponding to the action is used as the final target air-conditioning operation parameter group.

[0055] Through the embodiments of this application, the calculation and maximization of Q-values ​​ensures that the adjustments taken by the air conditioning system under the current environmental conditions can achieve the optimal overall benefits. The target air conditioning operating parameter set ultimately determined can achieve the dual goals of minimizing energy consumption and maximizing user comfort. Furthermore, because Q-values ​​incorporate the expected cumulative reward, the output layer can comprehensively consider both current and future impacts, ensuring not only short-term energy consumption optimization but also long-term system efficiency.

[0056] By combining the collaborative work of the input layer, hidden layer, and output layer, the entire air-conditioning parameter optimization model achieves efficient feedback from state input to optimal action output. It can not only respond to changes in the current environment, but also has certain predictive and foresight capabilities, ensuring the long-term stability and energy-saving effects of the system, and significantly improving the intelligent load scheduling capabilities of the air-conditioning system in commercial buildings.

[0057] During the training process of the deep Q network, an experience replay mechanism is used to randomly extract small batches of data from past experience for training, which helps to break data correlation and stabilize the network training process.

[0058] In addition, the target network Q(s,a;θ - ) to calculate the target Q value to reduce the estimated deviation, and the parameters θ of the target network - Online learning is achieved by periodically updating the main network (i.e., the Deep Q-Network). For example, parameters are copied and updated from the Deep Q-Network every certain number of training steps. Furthermore, an epsilon-greedy strategy can be used to balance exploration and exploitation. As training progresses, the epsilon value is gradually reduced to reduce random exploration. Later in training, the algorithm can more consistently exploit the best policy actions learned, ensuring higher rewards.

[0059] It should be noted that in a deep Q-network, the reward function is used during reinforcement learning to evaluate the immediate benefit or effect of an action in a specific state. It provides immediate information about environmental feedback and is used to guide network updates. For more details, please refer to the relevant descriptions above. The loss function is used in DQN to measure the difference between the current Q-value estimate and the target Q-value. It is used to guide the optimization of network parameters. The loss function of DQN is typically based on the Bellman Equation and is calculated as the mean squared error (MSE) between the current Q-value and the target Q-value.

[0060] In some examples of the embodiments of the present application, the loss function of the air-conditioning parameter optimization model can be combined with a multi-scale and time dynamic enhancement mechanism. By combining multi-scale time characteristics and time dynamic weights, it aims to simultaneously consider short-term and long-term load change characteristics, and dynamically adjust the weights of different time periods to enhance the adaptability and robustness of the model in complex environments.

[0061] More specifically, the loss function of the air conditioning parameter optimization model can be expressed as follows:

[0062]

[0063] Where L(θ) represents the overall loss function of the deep Q network, θ is the parameter of the current Q network; k represents the index of the time scale, K represents the total number of time scales, and λ k represents the weight coefficient of the kth time scale; T(t) is the time dynamic weight function, which represents the weight at time t and is used to dynamically adjust the sensitivity of the loss function to different time periods; is the expected value operator, which represents the expected value of all state s, action a, immediate reward r and next state s′ experience samples; r k is the immediate reward at time scale k, which represents the immediate benefit obtained by taking action a in the current state s; k is the discount factor at time scale k, which represents the discount rate of future rewards; Q k (s,a;θ) is the Q value estimate of the current network parameter θ for state s and action a at time scale k; max a′ Q k (s′,a′;θ - ) is the parameter θ of the target network at time scale k - The maximum Q-value estimate for the next state s′ represents the expected cumulative reward that can be obtained by selecting the optimal action a′ in the next state s′.

[0064] Regarding the explanation of formula (1), different time scales k are used to capture the characteristics of air conditioning load in different time ranges, such as short-term (such as hourly level) and long-term (such as daily or weekly level) load changes. Each time scale corresponds to an independent Q value estimation. In addition, the discount factor γ for each time scale is k May be different to weigh the importance of current rewards versus future rewards. k may be small, while the longer time scale γ kThe time dynamic weight function T(t) is used to dynamically adjust the sensitivity of the loss function to the Q-value estimation at different time periods of the day. For example, a higher weight T(t) can be assigned during peak hours (such as daytime working hours) to ensure more accurate air conditioning load scheduling at critical moments. k , which regulates the contribution of each time scale in the overall loss function.

[0065] Through the embodiments of the present application, multiple time scales are introduced into the loss function, and the enhanced loss function can simultaneously capture the short-term, medium-term and long-term air-conditioning load variation characteristics, so that the DQN model can flexibly adjust its scheduling strategy when facing load fluctuations at different time scales, which can not only meet short-term energy efficiency requirements, but also optimize long-term environmental comfort and system stability. Through the time dynamic weight function, the enhanced loss function can dynamically adjust the training focus of the model in different time periods of the day. Therefore, by introducing multi-scale characteristics and time dynamic weights, it helps to smooth the Q value estimation at different time scales, reduce the risk of overestimation due to environmental changes, and prevent the model from making overly optimistic predictions in complex environments, thereby maintaining the robustness of the scheduling strategy.

[0066] It should be noted that the air conditioning load in different areas of a building may be affected by a variety of factors, such as the external environment (temperature, humidity, etc.) and the internal environment (human activity, equipment operation, etc.). These factors have different dynamic characteristics over time. In some cases, the load changes rapidly, and the model needs to focus on the most recent time series information; in other cases, when the load changes slowly or is stable, a longer historical time step may be more valuable.

[0067] In view of this, for the LSTM model module in the time series graph attention network, in some implementations, it can adopt an adaptive time series window mechanism to dynamically adjust the time range of LSTM processing, so that the model can dynamically adjust the time step according to the current environment, thereby capturing more relevant time series information.

[0068] More specifically, the LSTM model module is used to determine the adaptive timing window corresponding to each graph node, and determine the calculation time range corresponding to each graph node based on the adaptive timing window, and process the sensor load timing data of the corresponding graph node according to each calculation time range.

[0069] The input feature vector matrix to be processed by LSTM within the calculation time range corresponding to the graph node i is expressed as follows:

[0070]

[0071] Where w iis the adaptive timing window size of graph node i, Represents the input feature vector of the sensor load time series data corresponding to the graph node i at time t;

[0072] Then, the adaptive timing window corresponding to each graph node is determined, including:

[0073] For the sensor load time series data of each graph node, the load change rate and environmental change rate corresponding to the sensor load time series data are calculated according to the reference window. The corresponding time series window of the graph node is determined based on the load change rate and environmental change rate. Specifically, the following steps are performed:

[0074]

[0075] Where, LCR i and ECR i They represent the load change rate and environment change rate corresponding to the graph node i, w base Indicates the preset reference window; is the air conditioning load value corresponding to the graph node i at time t′; is the pth environmental parameter corresponding to the graph node i at time t′, P represents the total number of parameter types of environmental parameters; β1 and β2 represent the factor adjustment factors respectively; w min and w max They represent the minimum and maximum values ​​of the adaptive timing window respectively; the clip(x,a,b) function means limiting the variable x to the numerical interval [a,b].

[0076] Here, the load change rate (LCR) is used to measure the speed of change of the regional air conditioning load in the past several time steps. The environmental change rate (ECR) is used to measure the speed of change of the external environment (such as temperature, humidity, etc.). Furthermore, based on the load change rate and the environmental change rate, the size of the time series window is dynamically adjusted so that the LSTM can focus on the historical time step with the most reference value. For example, when the LCR i or ECR i When it is larger, it means that the load or environment changes drastically, and the adaptive timing window size w of the graph node i i Will shrink, making the model pay more attention to the most recent time series data. i and ECR i When it is smaller, it means that the load and environment are relatively stable, and the timing window w i will expand, allowing the model to refer to longer historical time steps.

[0077] Through the embodiments of the present application, the time series window size used for LSTM input is adaptively adjusted based on real-time load changes and environmental dynamics to ensure that the selected window covers the most valuable historical data. By properly adjusting the time series window size, the adaptive mechanism can also avoid unnecessary computational overhead and optimize the model's resource utilization efficiency. In addition, by dynamically analyzing the load change rate and environmental change rate of different building areas, the adaptive time series window can be adaptively adjusted, ensuring that the LSTM maintains efficient time series feature extraction capabilities in diverse and complex environments.

[0078] For the graph attention model module in the temporal graph attention network, in some embodiments, a multi-scale convolution kernel is introduced into the basic graph attention model to process node neighborhood information at different distances. Through multi-scale information fusion, the network can simultaneously consider the influence of nearby and distant neighboring nodes on the current node, thereby more accurately capturing the complex dependencies between different areas of the building.

[0079] Figure 4 A structural connection diagram of an example of a graph attention model module according to an embodiment of the present application is shown.

[0080] like Figure 4 As shown, the graph attention model module 400 includes a near neighbor scale extraction submodule 410 , a far neighbor scale extraction submodule 420 and a scale fusion submodule 430 .

[0081] The neighbor scale extraction submodule 410 is used to extract and fuse information of neighbor nodes of the neighbor scale to obtain neighbor scale features:

[0082]

[0083] Where, Represents the neighbor scale feature of graph node i; Represents the set of neighboring nodes of graph node i, where the edge weights between graph node i and each graph node in the neighboring node set do not exceed the preset neighbor weight threshold; Represents the hidden state vector output by the LSTM model module for the neighboring graph node j of i at time t; is the neighbor attention weight between i and j, W near is the neighbor weight matrix; σ represents the activation function.

[0084] The distant neighbor scale extraction submodule 420 is used to extract and fuse information of neighbor nodes of the near neighbor scale to obtain distant neighbor scale features:

[0085]

[0086] Where, Represents the distant neighbor scale feature of graph node i; Represents the set of distant neighbor nodes of node i, and the edge weight between graph node i and each graph node in the distant neighbor node set is greater than the nearest neighbor weight threshold and less than the distant neighbor weight threshold; Represents the hidden state vector output by the LSTM model module for the distant neighbor graph node m of i at time t; is the distant neighbor attention weight between i and m, W far is the distant neighbor weight matrix.

[0087] The scale fusion submodule 430 is used to fuse the near-neighbor scale features and the far-neighbor scale features to obtain an updated node feature representation:

[0088]

[0089] Where, represents the updated node feature of graph node i, and || represents a vector concatenation operation.

[0090] It should be noted that, according to the description of the above related embodiments, the edge weight W ij The correlation between different regions in terms of physical space and functionality is comprehensively considered to reflect the comprehensive similarity between node i and node j. Here, the edge weight W ij The size of the edge can be used to classify node j as a close neighbor or a distant neighbor of node i. For example, if the edge weight W ij Greater than a certain threshold τ near , then node j is considered as a neighbor node of node i, if the edge weight W ij Higher than τ near and below a specific threshold τ far , node j is considered a distant neighbor of node i. Thus, edge weights directly reflect the strength of the association between nodes and distinguish which nodes have a direct influence on the target node from which nodes have a more indirect influence, thereby improving the model's ability to capture spatiotemporal information. Furthermore, using edge weights for partitioning reduces the computational effort for redundant nodes, enabling the model to more efficiently utilize computing resources when processing complex graph structures.

[0091] Through the embodiments of the present application, multi-scale information fusion is adopted, so that the graph attention model module can simultaneously consider the influence of the neighboring and distant nodes inside the building. By introducing convolution kernels and attention mechanisms of multiple scales, the model can capture the complex spatial dependencies between nodes at different distances. The characteristics of the neighboring nodes reflect the direct connection between the local area and can provide accurate local load information; the characteristics of the distant neighbor nodes capture a wider range of correlations, such as the dependencies between areas with similar functions but large physical distances. By processing the node features of the neighbors and distant neighbors at the same time, the model obtains richer spatial information, which enables it to consider more factors when performing load forecasting and scheduling, and realizes comprehensive spatial dependency capture capabilities, so that the model can make more accurate scheduling decisions when facing complex building structures, and better meet the actual system load scheduling needs.

[0092] Here, edge weights comprehensively consider multiple factors, such as environmental similarity, spatial connectivity paths, and air conditioning load correlation, to accurately reflect the strength of association between different areas of a building. In some examples of the present application, edge weights are calculated by comprehensively considering environmental similarity, spatial connectivity coefficients, and air conditioning load correlation, thereby accurately reflecting the strength of association between different areas of a building.

[0093] Specifically, edge weights are calculated as follows:

[0094]

[0095] Where W uv is the edge weight between graph node u and graph node v; Indicates the environmental similarity between u and v; represents the spatial connectivity coefficient between u and v, represents the air conditioning load correlation between u and v, γ1, γ2 and γ3 are the adjustment coefficients of the corresponding factors in the edge weight; E u,p and E v,p are the average values ​​of u and v for the pth environmental parameter at all times, σ p is the normalization factor of the pth environmental parameter; d uv is the spatial connectivity distance between the building area spaces corresponding to u and v, and λ is the attenuation parameter used to control the influence of the spatial connectivity distance; and denote the air-conditioning loads of u and v at time t, respectively, and B is the total number of moments considered when calculating the correlation of air-conditioning loads.

[0096] It should be noted that by adjusting the three coefficients γ1, γ2, and γ3, the impact of various factors on edge weights can be flexibly adjusted, allowing the model to maintain efficient adaptability in different buildings or environments. In addition, for environmental similarity, by measuring the similarity of environmental parameters such as temperature and humidity between regions, because regions with similar environments typically have similar air conditioning requirements, it reflects the environmental consistency between the two regions. By using an exponential function to handle environmental differences, it can more sensitively capture subtle environmental changes and appropriately amplify these differences in the calculation to avoid them being ignored.

[0097] The spatial connectivity coefficient uses the shortest path length to measure the connectivity between two areas. Compared to simple physical distance, the shortest path length better reflects the impact of actual spatial structure on air conditioning scheduling. For example, even if adjacent rooms are physically close, the actual connectivity will be significantly reduced if multiple doors or staircases are required. In addition, by using an exponential decay function to process the shortest path length, the model can appropriately reduce the impact of similarity between areas on edge weights when connectivity is poor.

[0098] Regarding air conditioning load correlation, the correlation between historical air conditioning loads in two regions reflects similarities in their air conditioning usage patterns. Regions with high load correlation are likely to have synchronized air conditioning demands in future scheduling. Through standardized calculation methods, the model can stably capture the load correlation between different regions, thereby making scheduling strategies more precise.

[0099] Therefore, through the above-mentioned calculation process of edge connection weights, the model can carefully capture the multi-dimensional correlation information between different regions and reasonably calculate the edge weights between regions, which can improve the model's prediction accuracy for future air-conditioning loads and improve the overall efficiency of scheduling.

[0100] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of combined actions, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0101] Figure 5 A structural block diagram of an example of an intelligent scheduling control system for air-conditioning loads in a commercial building according to an embodiment of the present application is shown.

[0102] like Figure 5As shown, the intelligent dispatching and control system 500 for air-conditioning loads in commercial buildings includes a data acquisition unit 510 , a graph structure construction unit 520 , a graph node updating unit 530 and a parameter optimization unit 540 .

[0103] The data acquisition unit 510 is used to collect corresponding sensor load time series data based on sensor modules deployed in various building area spaces in commercial buildings; the sensor load time series data includes multiple historical time steps and corresponding multimodal environmental sensor data and regional air-conditioning loads.

[0104] The graph structure construction unit 520 is used to construct a building sensor graph structure based on each of the sensor load time series data; the building sensor graph structure includes multiple graph nodes and corresponding edge connections; each of the graph nodes uniquely corresponds to a building area space, and the node characteristics of the graph nodes are determined by the sensor load time series data of the corresponding building area space; the edge weight of each edge connection is determined according to the environmental similarity, spatial connectivity path and air-conditioning load correlation corresponding to the connected graph node.

[0105] The graph node updating unit 530 is used to process the building sensor graph structure based on a temporal graph attention network to update the node features of each of the graph nodes; the temporal graph attention network includes a cascaded LSTM model module and a graph attention model module.

[0106] The parameter optimization unit 540 is used to process the node features of each updated graph node based on the air-conditioning parameter optimization model to correspondingly determine the target air-conditioning operation parameter group of each building area space in a preset time period in the future; the air-conditioning parameter optimization model adopts a reinforcement learning model, and the state of the air-conditioning parameter optimization model is defined by the node features of the updated graph nodes corresponding to each building area space, and the action of the air-conditioning parameter optimization model is defined according to the adjustment amount of the air-conditioning operation parameter group for each building area space, and the reward of the air-conditioning parameter optimization model is defined according to the optimization degree of the global air-conditioning load and global environmental comfort of the commercial building.

[0107] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices, etc.) to execute the steps of any of the above-mentioned commercial building air-conditioning load intelligent scheduling and control methods of the present application.

[0108] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the steps of any one of the above-mentioned commercial building air-conditioning load intelligent scheduling and control methods.

[0109] In some embodiments, an embodiment of the present application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the intelligent scheduling and control method for commercial building air-conditioning loads.

[0110] Figure 6 FIG. 1 is a schematic diagram of the hardware structure of an electronic device for executing an intelligent dispatching control method for air-conditioning loads in commercial buildings provided by another embodiment of the present application. Figure 6 As shown, the device includes:

[0111] One or more processors 610 and memory 620, Figure 6 A processor 610 is taken as an example.

[0112] The device for executing the intelligent dispatching control method for commercial building air-conditioning load may further include: an input device 630 and an output device 640 .

[0113] The processor 610, the memory 620, the input device 630 and the output device 640 may be connected via a bus or other means. Figure 6 The bus connection is taken as an example.

[0114] Memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the intelligent scheduling and control method for commercial building air conditioning loads in the embodiments of this application. Processor 610 executes the non-volatile software programs, instructions, and modules stored in memory 620 to execute various server functional applications and data processing, thereby implementing the intelligent scheduling and control method for commercial building air conditioning loads in the aforementioned method embodiment.

[0115] The memory 620 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device, etc. In addition, the memory 620 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include a memory remotely located relative to the processor 610, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0116] The input device 630 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 640 may include a display device such as a display screen.

[0117] The one or more modules are stored in the memory 620 and, when executed by the one or more processors 610 , perform the intelligent dispatching and control method for air-conditioning load of commercial buildings in any of the above method embodiments.

[0118] The above-mentioned product can execute the method provided in the embodiment of this application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of this application.

[0119] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0120] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and their primary purpose is to provide voice and data communications. These terminals include smartphones, multimedia phones, feature phones, and low-end phones.

[0121] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers and have computing and processing capabilities, and generally also have mobile Internet access. These terminals include PDAs, MIDs, and UMPCs.

[0122] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0123] (4) Other onboard electronic devices with data interaction functions, such as onboard computer devices installed in vehicles.

[0124] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0125] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, or of course, by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiment.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for intelligent dispatching and controlling air conditioning loads in commercial buildings, comprising: Based on the sensor modules deployed in each building area space in the commercial building, the corresponding sensor load time series data are collected respectively; The sensor load time series data includes multiple historical time steps and corresponding multi-modal environmental sensor data and regional air conditioning load; Based on each of the sensor load time series data, a building sensor graph structure is constructed; the building sensor graph structure includes a plurality of graph nodes and corresponding edge connections; each of the graph nodes uniquely corresponds to a building area space, and the node characteristics of the graph nodes are determined by the sensor load time series data of the corresponding building area space; the edge weight of each edge connection is determined according to the environmental similarity, spatial connectivity path, and air conditioning load correlation corresponding to the connected graph node; Processing the building sensor graph structure based on a temporal graph attention network to update node features of each of the graph nodes; The temporal graph attention network includes a cascaded LSTM model module and a graph attention model module; Based on the air-conditioning parameter optimization model, the node features of each updated graph node are processed to determine the target air-conditioning operation parameter group of each building area space in a preset time period in the future; the air-conditioning parameter optimization model adopts a reinforcement learning model, and the state of the air-conditioning parameter optimization model is defined by the node features of the updated graph nodes corresponding to each building area space, the action of the air-conditioning parameter optimization model is defined according to the adjustment amount of the air-conditioning operation parameter group for each building area space, and the reward of the air-conditioning parameter optimization model is defined according to the optimization degree of the global air-conditioning load and global environmental comfort of the commercial building.

2. The method according to claim 1, wherein The air conditioning parameter optimization model adopts a deep Q network, which includes an input layer, a hidden layer and an output layer; The input layer is used to receive the state vector corresponding to the node feature of each updated graph node; The hidden layer is used to process the state vector through a multi-layer neural network to extract at least one state-action mapping relationship, thereby obtaining a corresponding plurality of potential actions; The output layer is used to determine the Q value corresponding to each potential action and determine the target air conditioning operating parameter group based on the air conditioning operating parameter group adjustment corresponding to the potential action with the largest Q value; wherein the Q value is defined based on the expected cumulative reward that can be obtained by taking the corresponding potential action under the state vector.

3. The method according to claim 2, wherein: The loss function of the air conditioning parameter optimization model is: Where L(θ) represents the overall loss function of the deep Q network, θ is the parameter of the current Q network; k represents the index of the time scale, K represents the total number of time scales, and λ k represents the weight coefficient of the kth time scale; T(t) is the time dynamic weight function, which represents the weight at time t and is used to dynamically adjust the sensitivity of the loss function to different time periods; is the expected value operator, which represents the expected value of all state s, action a, immediate reward r and next state s′ experience samples; r k is the immediate reward at time scale k, which represents the immediate benefit obtained by taking action a in the current state s; k is the discount factor at time scale k, which represents the discount rate of future rewards; Q k (s,a;θ) is the Q value estimate of the current network parameter θ for state s and action a at time scale k; max a′ Q k (s′,a′;θ - ) is the parameter θ of the target network at time scale k - The maximum Q-value estimate for the next state s′ represents the expected cumulative reward that can be obtained by selecting the optimal action a′ in the next state s′.

4. The method according to claim 1, wherein The LSTM model module is used to determine the adaptive timing window corresponding to each graph node, and determine the calculation time range corresponding to each graph node according to the adaptive timing window, and process the sensor load time series data of the corresponding graph node according to each of the calculation time ranges; The input feature vector matrix to be processed by LSTM within the calculation time range corresponding to the graph node i is expressed as follows: Where w i is the adaptive timing window size of graph node i, Represents the input feature vector of the sensor load time series data corresponding to the graph node i at time t; Determining the adaptive timing window corresponding to each graph node includes: For the sensor load time series data of each graph node, the load change rate and the environment change rate corresponding to the sensor load time series data are calculated according to the reference window, and the time series window corresponding to the graph node is determined based on the load change rate and the environment change rate, specifically including: Where, LCR i and ECR i They represent the load change rate and environment change rate corresponding to the graph node i, w base Indicates the preset reference window; is the air conditioning load value corresponding to the graph node i at time t′; is the pth environmental parameter corresponding to the graph node i at time t′, P represents the total number of parameter types of environmental parameters; β1 and β2 represent the factor adjustment factors respectively; w min and w max They represent the minimum and maximum values ​​of the adaptive timing window respectively; the clip(x,a,b) function means limiting the variable x to the numerical interval [a,b].

5. The method according to claim 4, wherein The graph attention model module includes a near-neighbor scale extraction submodule, a far-neighbor scale extraction submodule, and a scale fusion submodule; The neighbor scale extraction submodule is used to extract and fuse information of neighbor nodes of the neighbor scale to obtain neighbor scale features: Where, Represents the neighbor scale feature of graph node i; Represents the set of neighboring nodes of graph node i, where the edge weights between graph node i and each graph node in the neighboring node set do not exceed the preset neighbor weight threshold; represents the hidden state vector output by the LSTM model module for the neighboring graph node j of i at time t; is the neighbor attention weight between i and j, W near is the neighbor weight matrix; σ represents the activation function; The distant neighbor scale extraction submodule is used to extract and fuse information of neighbor nodes at the near neighbor scale to obtain distant neighbor scale features: Where, Represents the distant neighbor scale feature of graph node i; Represents a set of distant neighbor nodes of node i, where the edge weight between graph node i and each graph node in the distant neighbor node set is greater than the neighbor weight threshold and less than the distant neighbor weight threshold; Represents the hidden state vector output by the LSTM model module for the distant neighbor graph node m of i at time t; is the distant neighbor attention weight between i and m, W far is the distant neighbor weight matrix; The scale fusion submodule is used to fuse the near-neighbor scale features and the far-neighbor scale features to obtain an updated node feature representation: Where, represents the updated node feature of graph node i, and || represents a vector concatenation operation.

6. The method according to claim 5, wherein: The edge weights are calculated as follows: Where W uv is the edge weight between graph node u and graph node v; Indicates the environmental similarity between u and v; represents the spatial connectivity coefficient between u and v, represents the air conditioning load correlation between u and v, γ1, γ2 and γ3 are the adjustment coefficients of the corresponding factors in the edge weight; E u,p and E v,p are the average values ​​of u and v for the pth environmental parameter at all times, σ p is the normalization factor of the pth environmental parameter; d uv is the spatial connectivity distance between the building area spaces corresponding to u and v, and λ is the attenuation parameter used to control the influence of the spatial connectivity distance; and denote the air-conditioning loads of u and v at time t, respectively, and B is the total number of moments considered when calculating the correlation of air-conditioning loads.

7. The method according to any one of claims 1 to 6, wherein The parameter types of the multimodal environmental sensing data include temperature, humidity, CO2 concentration, light intensity and human activity density.

8. The method according to any one of claims 1 to 6, wherein The parameter types of the air conditioning operation parameter group include air supply temperature, air supply volume, wind speed and air conditioning on / off time.

9. An intelligent dispatching and control system for air conditioning loads in commercial buildings, comprising: A data acquisition unit is configured to collect corresponding sensor load time series data based on sensor modules deployed in various building area spaces in a commercial building; the sensor load time series data includes multiple historical time steps and corresponding multimodal environmental sensor data and regional air conditioning loads; A graph structure construction unit is configured to construct a building sensor graph structure based on each of the sensor load time series data; the building sensor graph structure comprises a plurality of graph nodes and corresponding edge connections; each of the graph nodes uniquely corresponds to a building area space, and the node characteristics of the graph nodes are determined by the sensor load time series data of the corresponding building area space; and the edge weight of each edge connection is determined based on the environmental similarity, spatial connectivity path, and air conditioning load correlation corresponding to the connected graph node; A graph node updating unit, configured to process the building sensor graph structure based on a temporal graph attention network to update node features of each graph node; The temporal graph attention network includes a cascaded LSTM model module and a graph attention model module; A parameter optimization unit is used to process the node features of each updated graph node based on the air-conditioning parameter optimization model to correspondingly determine the target air-conditioning operation parameter group of each building area space in a preset time period in the future; the air-conditioning parameter optimization model adopts a reinforcement learning model, and the state of the air-conditioning parameter optimization model is defined by the node features of the updated graph nodes corresponding to each building area space, the action of the air-conditioning parameter optimization model is defined according to the adjustment amount of the air-conditioning operation parameter group for each building area space, and the reward of the air-conditioning parameter optimization model is defined according to the optimization degree of the global air-conditioning load and global environmental comfort of the commercial building.

Images