A method and apparatus for optimizing control of building air conditioning systems based on spatiotemporal data

Abstract

This invention proposes an optimized control method and device for a building air conditioning system based on spatiotemporal data. The method includes: performing temperature simulation to obtain N simulated temperature distributions at different times throughout the year; establishing a room temperature distribution prediction model based on a graph attention neural network and a gated loop unit; performing energy consumption simulation to obtain N simulated energy consumption data at different times throughout the year; establishing an energy consumption prediction model based on a backpropagation neural network, using the air conditioning setting parameters as input and the energy consumption at the next time moment as output; optimizing the system using a multi-objective particle swarm optimization algorithm based on the room temperature distribution prediction model and the energy consumption prediction model, and determining the optimal solution using a weighted sum method, wherein the optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters; and controlling the air conditioning system according to the optimal solution, effectively improving the reliability of the building air conditioning system control.

Description

Technical Field

[0001] This invention relates to the field of air conditioning system control, and in particular to a method and apparatus for optimizing the control of building air conditioning systems based on spatiotemporal data. Background Technology

[0002] With rapid societal development and rising living standards, people have higher demands for thermal comfort in their living and working environments. Simultaneously, the global energy shortage is becoming increasingly severe, with the construction industry accounting for a large proportion of global energy consumption, and air conditioning systems being a major component of building energy consumption. Therefore, optimizing and controlling the operation of air conditioning systems is a crucial means to achieve building energy conservation and alleviate the energy crisis.

[0003] Current research often focuses on optimizing indoor thermal comfort based on average room temperature. However, in reality, room temperature distribution is often uneven, especially when the air conditioning system is first turned on, the temperature distribution is even more unequal. Moreover, improving thermal comfort often leads to increased energy consumption. These two goals are contradictory and need to be considered together, resulting in low reliability of air conditioning system operation and control.

[0004] To address this problem, the present invention provides a method and apparatus for optimizing the control of a building air conditioning system based on spatiotemporal data, thereby solving the aforementioned issues. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, this invention innovatively proposes a method and device for optimizing the control of building air conditioning systems based on spatiotemporal data, which effectively solves the problem of low control reliability of building air conditioning systems caused by the prior art and effectively improves the reliability of building air conditioning system control.

[0006] The first aspect of this invention provides a method for optimizing and controlling a building air conditioning system based on spatiotemporal data, comprising:

[0007] Create a 3D BIM model of the building rooms, divide the 3D BIM model into meshes, and create a CFD model.

[0008] In the CFD model, N simulated operating conditions with air conditioning setting parameters are defined. Temperature simulations are then performed on each of the N simulated operating conditions for weather conditions within a year to obtain N simulated temperature distributions at different times within a year. The air conditioning setting parameters include the air conditioning pre-start time, temperature set point, air supply speed, and air supply angle.

[0009] A room temperature distribution prediction model is established based on graph attention neural network and gated recurrent unit.

[0010] The 3D BIM model is input into the energy consumption calculation software, and energy consumption simulations are performed on N simulated operating conditions for air conditioning settings under different weather conditions throughout the year to obtain N energy consumption simulation data at different times throughout the year.

[0011] Using the air conditioner's settings as input and the energy consumption at the next moment as output, an energy consumption prediction model is established based on a BP neural network.

[0012] Based on the room temperature distribution prediction model and energy consumption prediction model, a multi-objective particle swarm optimization algorithm is used for optimization, and a weighted sum method is used to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters.

[0013] The air conditioning system is controlled according to the optimal solution.

[0014] A second aspect of the present invention provides an optimized control device for a building air conditioning system based on spatiotemporal data, comprising:

[0015] The first module establishes a 3D BIM model of the building rooms, divides the 3D BIM model into meshes, and establishes a CFD model.

[0016] The first simulation module defines N simulated operating conditions for air conditioning settings in the CFD model. It then performs temperature simulations on each of the N simulated operating conditions for weather conditions within a year, obtaining N simulated temperature distributions at different times within a year. The air conditioning settings include the air conditioning pre-start time, temperature set point, air supply speed, and air supply angle.

[0017] The second module establishes a room temperature distribution prediction model based on graph attention neural networks and gated recurrent units.

[0018] The second simulation module inputs the 3D BIM model into the energy consumption calculation software, performs energy consumption simulations for N simulated operating conditions of the air conditioning settings for weather conditions within a year, and obtains N energy consumption simulation data at different times within a year.

[0019] The third module takes the air conditioner's settings as input and the energy consumption at the next moment as output, and establishes an energy consumption prediction model based on a BP neural network.

[0020] The optimization module is based on the room temperature distribution prediction model and the energy consumption prediction model. It uses a multi-objective particle swarm optimization algorithm for optimization and a weighted sum method to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters.

[0021] The control module controls the air conditioning system according to the optimal solution.

[0022] The technical solution adopted in this invention has the following technical effects:

[0023] 1. The technical solution of this invention obtains N simulated temperature distributions and energy consumption simulation data at different times within a year; based on the room temperature distribution prediction model and energy consumption prediction model, a multi-objective particle swarm optimization algorithm is used for optimization, and a weighted sum method is used to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters. This effectively solves the problem of low control reliability of building air conditioning systems caused by existing technologies and effectively improves the control reliability of building air conditioning systems.

[0024] 2. The technical solution of this invention is based on BIM, CFD and EnergyPlus modeling, which effectively reduces the data collection cycle and improves the efficiency of model training; moreover, based on spatiotemporal data, a room temperature distribution prediction model is established through graph attention neural network and gated recurrent unit, which improves the accuracy of room temperature distribution prediction.

[0025] 3. In the room temperature field topology diagram of the present invention, the nodes can be either sitting type nodes or standing type nodes for people whose stay time is longer than the preset time. Alternatively, the nodes can be determined by calculating the temperature gradient of the temperature distribution in the room area, which improves the flexibility of node selection.

[0026] 4. In the technical solution of this invention, the feature vector h' of the target node i output by the aggregation graph attention neural network at time t is... i Preprocessing is performed, specifically multi-head attention processing, to improve the learning ability of the room temperature distribution prediction model.

[0027] 5. The air conditioning system control based on the optimal solution in the technical solution of the present invention includes: optimized control of air conditioning pre-start and optimized control of air conditioning during room usage periods; different optimized controls can be performed according to the usage periods of the building air conditioning, further improving the reliability and flexibility of the optimized control of the building air conditioning system.

[0028] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1This is a schematic flowchart of the method in Embodiment 1 of the present invention;

[0031] Figure 2 This is another flowchart illustrating the method in Embodiment 1 of the present invention.

[0032] Figure 3 This is a flowchart illustrating step S3 in the method of Embodiment 1 of the present invention;

[0033] Figure 4 This is a schematic diagram illustrating the determination of node positions within a building room based on the duration of human stay and the location of the human in the method of Embodiment 1 of the present invention.

[0034] Figure 5 This is a schematic diagram of the gated loop unit in the method of Embodiment 1 of the present invention;

[0035] Figure 6 This is a flowchart illustrating the multi-objective particle swarm optimization algorithm in Embodiment 1 of the present invention.

[0036] Figure 7 This is a flowchart illustrating the weighted sum method in Embodiment 1 of the present invention.

[0037] Figure 8 This is a schematic diagram of the device in Embodiment 2 of the present invention. Detailed Implementation

[0038] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0039] Example 1

[0040] like Figures 1-2 As shown, this invention provides an optimized control method for a building air conditioning system based on spatiotemporal data, comprising:

[0041] Step S1: Create a 3D BIM model of the building rooms, divide the 3D BIM model into meshes, and create a CFD model.

[0042] Step S2: Define N simulated operating conditions for air conditioning settings in the CFD model, and perform temperature simulations for the N simulated operating conditions of the CFD model for weather conditions within a year to obtain N simulated temperature distributions at different times within a year; wherein, the air conditioning settings include air conditioning pre-start time, temperature set point, air supply speed and air supply angle.

[0043] Step S3: Based on Graph Attention Networks (GAT) and Gated Recurrent Units (GRU), establish a room temperature distribution prediction model.

[0044] Step S4: Input the 3D BIM model into the energy consumption calculation software, and perform energy consumption simulations for N simulated operating conditions of the air conditioning settings for weather conditions within a year, to obtain N energy consumption simulation data at different times within a year.

[0045] Step S5: Using the air conditioner's settings as input and the energy consumption at the next moment as output, establish an energy consumption prediction model based on a BP neural network.

[0046] Step S6: Based on the room temperature distribution prediction model and energy consumption prediction model, the multi-objective particle swarm optimization algorithm is used for optimization, and the weighted sum method is used to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters.

[0047] Step S7: Control the air conditioning system according to the optimal solution.

[0048] In step S1, a three-dimensional solid model of the building rooms is first created using BIM (Building Information Modeling) software based on drawings and on-site measurement information. This model includes interior and exterior walls, doors and windows, floors, roofs, structural columns, etc.

[0049] The BIM model is imported into mesh generation software to create a mesh model. The mesh model is then imported into CFD (Computational Fluid Dynamics) software, where model selection and settings, material settings, boundary settings, initial condition settings, and solution settings are performed to finally create the CFD model.

[0050] In step S2, N simulated operating conditions with air conditioning setting parameters are defined in the CFD model (air conditioning setting parameters include air conditioning pre-start time, temperature set point, air supply speed, and air supply angle). One of the operating conditions is selected to perform the simulation. Multiple temperature measuring instruments are evenly placed in the room to measure the room temperature. The measured room temperature is compared with the simulated temperature to verify the accuracy of the CFD model.

[0051] Temperature simulations are performed on N operating conditions with specified air conditioning settings in the CFD model for weather conditions within a year (taking 365 days as an example). The temperature distribution output step size is set to R minutes. Therefore, N temperature simulation distributions at different times within a year are obtained, i.e., 365×(24×60) / R×N temperature distributions.

[0052] like Figure 3 Specifically, step S3 includes:

[0053] S31, Establish a room temperature field topology diagram. The room temperature field topology diagram is composed of nodes and edges. The nodes are used to represent different selected locations in the building room, and the edges are used to represent the correlation between different nodes.

[0054] In step S31, different points are selected from the building rooms to establish a topological structure diagram of the room temperature field. The topological structure diagram of the room temperature field consists of nodes and edges, represented as graph G = (Q, O), where Q is a set of nodes in the topological structure diagram of the room temperature field, representing different selected locations in the room, and O is a set of edges in the topological structure diagram of the room temperature field, representing the correlation between different nodes.

[0055] One method for selecting nodes in the room temperature field topology diagram is subjective selection, primarily based on areas where people spend a significant amount of time within the room. For example, in office buildings... Figure 4 As shown, if a person sits at their workstation for an extended period, three nodes can be selected from the room space corresponding to the head, chest, and legs of the person. Nodes can also be selected from the room space corresponding to the head when the person is standing, and from the room space corresponding to the head, chest, and legs of a standing person in the aisle. Specifically, these can be nodes representing the sitting posture type and the standing posture type. The sitting posture type nodes include the room space corresponding to the head when the person stands up, the room space corresponding to the head while sitting, the room space corresponding to the chest while sitting, and the room space corresponding to the legs while sitting. The standing posture type nodes include the room space corresponding to the head, the room space corresponding to the chest while standing, and the room space corresponding to the legs while standing.

[0056] Another method for determining nodes in the room temperature field topology diagram is to objectively select nodes. This involves measuring the room's temperature distribution using temperature acquisition instruments, calculating the temperature gradient within the room area, and identifying a node at a location where the temperature gradient exceeds a preset threshold and there are no other nodes within a preset distance. The specific process is as follows: First, temperature acquisition instruments such as thermal imagers and thermocouples are used. If necessary, ultrasonic travel time tomography (which involves two steps: first, measuring the signal propagation time using an ultrasonic sensor to calculate the sound velocity along the propagation path; second, reconstructing the spatial temperature distribution of the entire test area using tomographic imaging. Common reconstruction methods in tomographic imaging include iterative and non-iterative methods. Iterative methods include Landweber iteration, synchronous iterative reconstruction, and synchronous algebraic reconstruction; non-iterative methods include truncated singular value decomposition and Tikhonov regularization) are used to measure the room's temperature distribution. Then, the temperature gradient within the area is calculated. If no other nodes exist within a distance *r*, then a node is selected, where ΔT is the temperature difference, Δh is the spatial distance change, and ε is the node selection threshold. Under specific spatial distance conditions, a smaller threshold indicates a higher requirement for room temperature. The selection of the threshold ε is generally first determined by expert scoring to establish different levels of thresholds, and then the corresponding level of threshold is selected based on the room temperature requirements.

[0057] Common empirical values, for example, the vertical temperature difference at 0.1m and 1.1m above the floor (considering sitting posture) is divided into three levels, with temperature differences of less than 2℃, 3℃ and 4℃ respectively; the vertical temperature difference between 1.8m and 0.1m (considering standing posture) is no greater than 3℃.

[0058] The edges of the room temperature field topology graph are the lines connecting the selected nodes.

[0059] S32, construct the time series feature matrix for each moment within a preset time period, and normalize the time series feature matrix; wherein, the time series feature matrix includes outdoor meteorological parameters, node temperatures output after temperature simulation, and relevant air conditioning setting parameters; the time series feature matrix at a certain moment is e rows and b columns, where e represents the number of nodes and b represents the number of feature data.

[0060] The input features of the room temperature distribution prediction model include *b* features: outdoor meteorological parameters, node temperatures (obtained from the temperature distribution output by the CFD model; that is, temperature simulations are performed on N operating conditions with specified air conditioning settings in the CFD model for weather conditions within a year, with the temperature distribution output step size set to R minutes, resulting in N simulated temperature distributions at different times within a year), and relevant air conditioning settings (i.e., air conditioning pre-start time, temperature setpoint, air supply speed, and air supply angle). The outdoor meteorological parameters and relevant air conditioning settings correspond to each node and, together with the node temperature, serve as the node's features. Therefore, a feature matrix is ​​established for each time point within the first *n* hours at time intervals of R minutes. The feature matrix for a given time point is *e* rows and *b* columns, where *e* represents the number of nodes and *b* represents the feature data.

[0061] S33, establish the adjacency matrix of the room temperature field topology diagram, and determine whether any two nodes in the room temperature field topology diagram are adjacent based on the adjacency matrix;

[0062] Adjacency matrices are used to represent the relationships between nodes. The matrix elements U of an adjacency matrix are... IJ U represents the relationship between node I and node J. IJ The calculation method is as follows:

[0063]

[0064] Wherein, P(vx) I vx J ), P(vx I vx J ) and P(vx I vx J ) are the Pearson correlation coefficients of the velocity components of nodes I and J along the x, y, and z axes, respectively. The velocity components represent the magnitude of the velocity in different directions along the three coordinate axes, describing the airflow and thus influencing the temperature field distribution; ε is the threshold for the correlation coefficient; U IJ A value of 1 indicates that nodes I and J share an edge, meaning they are strongly correlated and neighbors; U IJ If the value is 0, it means that the two nodes I and J are not adjacent and are unrelated.

[0065] S34, extract the spatial features of each node based on the graph attention neural network, wherein the spatial features of the node include the attention coefficients between the target node and its neighboring nodes;

[0066] S35, calculate the attention coefficient of all neighboring nodes to the target node. The attention coefficient of a neighboring node to the target node represents the importance of the neighboring node to the target node. The specific calculation method of the attention coefficient is as follows:

[0067]

[0068] Where, α ij Here, σi represents the normalized attention coefficients of target node i and neighbor node j; i is the target node; j is neighbor node j; k is neighbor node k; σ1 is the LeakyReLU activation function, which is a non-linear activation function; N i Let represent all neighboring nodes of target node i; || represents the concatenation operation between feature vectors; a represents the learnable parameter vector; W1 is the linear transformation matrix used for node feature transformation, with the same size as the feature matrix, used for feature transformation; h j h is the feature vector of target node i, that is, the i-th row of the feature matrix; j h is the feature vector of neighbor node j; k The feature vector of neighbor node k;

[0069] S36, based on the attention coefficients of all neighboring nodes towards the target node, aggregate the spatial features of the neighboring nodes adjacent to the target node to obtain the feature vector of the target node output by the aggregated graph attention neural network; wherein, the expression of the feature vector of the target node output by the aggregated graph attention neural network is:

[0070]

[0071] Among them, h' i The aggregated graph shows the feature vector of the target node output by the neural network, where σ2 is the Sigmoid activation function.

[0072] S37, based on the feature vector of each target node i output by the aggregated graph attention neural network from the gated recurrent unit, the corresponding time features are extracted to obtain the room temperature distribution prediction model; where, as Figure 5 As shown, the gated loop unit includes a reset gate and an update gate. The expression of the gated loop unit specifically includes:

[0073] r t =σ2(W r x t +U r h t-1 +b r ),

[0074] z t =σ2(W z x t +U z h t-1 +b z ),

[0075]

[0076] Where, r t To reset the gate output, z t To update the gate output; x t For the aggregated graph, note the feature vector h' of each target node i output by the neural network at time t. i W r W is the weight matrix input to the reset gate; z To update the weight matrix of the gate; W h U is the weight matrix of the candidate hidden states; r The hidden state h at time t-1 t-1 Weight matrix to reset gate; U z U is the weight matrix of the update gate at time t-1; h b is the weight matrix of the candidate hidden states at time t-1; r b z and b h These are the biases for resetting the gate, updating the gate, and the candidate hidden state, respectively; h t-1 This represents the hidden state at time t-1; tanh is the matrix element-wise product; tanh is the hyperbolic tangent function. Let h be the candidate hidden state at time t; t Let h be the hidden state at time t, and the last h in the time series. t The data is input into the fully connected layer to predict the temperature distribution at the next time step.

[0077] S38. Set the basic parameters of the room temperature distribution prediction model, divide the dataset into training set and test set according to the preset ratio, and use the training set to train the room temperature distribution prediction model to obtain the trained room temperature distribution prediction model.

[0078] The preset ratio can be to divide the temperature distribution simulation dataset into a training set and a test set in a 7:3 ratio, and use the training set to train the room temperature distribution prediction model based on S31 to S37. After the room temperature distribution prediction model is trained, use the test set to test the performance of the room temperature distribution prediction model. If the performance of the room temperature distribution prediction model meets the requirements, output the temperature distribution prediction result for the next moment.

[0079] Model evaluation metrics can include root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and correlation coefficient (R²). 2 One or more of them are not limited to the present invention.

[0080]

[0081]

[0082] In the formula, For predicted values; y f This is the actual value; is the average of the actual values; n is the number of samples.

[0083] Preferably, the feature vector h' of the target node i output by the aggregated graph attention neural network at time t is... i Preprocessing is performed, specifically multi-head attention processing. The feature vector of the target node i output by the preprocessed graph attention neural network at time t is as follows:

[0084]

[0085] Among them, h' i ' is the feature vector of target node i output by the preprocessed graph attention neural network at time t; M represents the number of heads in the multi-head attention; Let be the attention coefficient of the m-th head; Let be the weight matrix of the m-th head.

[0086] That is, in step S37, based on the feature vector of each target node i output by the aggregated graph attention neural network or the feature vector of the target node i output by the preprocessed graph attention neural network at time t, the corresponding time features are extracted to obtain the room temperature distribution prediction model; wherein, in the expression of the gated recurrent unit, x t For the aggregated graph, note the feature vector h' of each target node i output by the neural network at time t. i Or, note the feature vector h' of the target node i output by the neural network at time t after preprocessing. i '.

[0087] In step S4, the BIM model is input into the EnergyPlus energy consumption calculation software (energy consumption model), and the measured energy consumption data of a selected period is compared with the simulated energy consumption data to verify the accuracy of the energy consumption model; then, energy consumption simulation is performed for N operating conditions of air conditioning settings for 365×N weather conditions within a year, with the energy consumption output step size being R minutes, thus obtaining 365×(24×60) / R×N energy consumption data.

[0088] In step S5, the air conditioner's setting parameters are used as input, and the energy consumption at the next moment is used as output (using 365×(24×60) / R×N energy consumption data obtained in step S4). A deep training based on a BP neural network is performed to establish an energy consumption prediction model. That is, the energy consumption dataset is divided into a training set and a test set in a 7:3 ratio. The training set is used to train the energy consumption prediction model. After the energy consumption prediction model is trained, the test set is used to test the performance of the energy consumption prediction model. If the performance of the energy consumption prediction model meets the requirements, the energy consumption prediction result at the next moment is output.

[0089] In step S6, such as Figure 6 As shown, the optimization using the multi-objective particle swarm optimization algorithm specifically includes:

[0090] Initialize the particle swarm and external archive, and set the termination conditions for the multi-objective particle swarm algorithm;

[0091] Calculate the fitness of each particle, which includes a thermal comfort objective function F1 and an energy consumption objective function F2;

[0092] Update the historical best position pBest and the global best position gBest for each particle;

[0093] Update the velocity and position of each particle, where the particle velocity and position are updated in the following way:

[0094]

[0095] in, γ1 and γ2 are the particle velocities of particle h in the (e+1)th and eth iterations, respectively; c1 and c2 are the acceleration coefficients; γ1 and γ2 are random numbers between [0,1]. ω represents the historical optimal position and the global optimal position of particle h in the e-th iteration, respectively; ω is the inertia weight. These are the positions of particle h in the (e+1)th and eth iterations, respectively;

[0096] Update the external archive based on the updated velocity and position of each particle;

[0097] The Pareto set is output until the current iteration number or the fitness function value meets the termination condition. The termination condition includes the iteration number meeting a preset maximum iteration number or the fitness function value being less than a preset fitness threshold.

[0098] The thermal comfort objective function is represented by the sum of the differences between the predicted temperatures of all target nodes and the set optimal comfort temperature, specifically:

[0099]

[0100] The energy consumption objective function is as follows:

[0101] min F2=E,

[0102] Where: F1 is the thermal comfort objective function, represented by the absolute value of the difference between the predicted room temperature at the next time step and the set optimal comfort temperature; F2 is the energy consumption objective function; Q is the total number of objective nodes; T q Let be the predicted temperature of the q-th target node; Set the most comfortable temperature for the q-th target node; E is the energy consumption of the air conditioning system between the current moment and the next moment (one time step).

[0103] like Figure 7 As shown, determining the optimal solution using the weighted sum method specifically includes:

[0104] Weights are assigned to each objective function to optimize the target.

[0105] Based on the weights of each optimization objective function, the weighted sum of each solution is calculated, whereby the weighted sum of each solution is calculated as follows:

[0106] S z =ζ1F z1 +ζ2F z2 ,

[0107] Among them, S z Let be the weighted sum of the z-th scheme; ζ1 and ζ2 are the weights of the thermal comfort objective function and the energy consumption objective function, respectively; F z1 F z2 These are the thermal comfort objective function value and the energy consumption objective function value for the z-th scheme, respectively;

[0108] Sort all schemes according to the weighted sum (total value) of each scheme;

[0109] The optimal solution is the one with the smallest weighted sum.

[0110] In step S7, controlling the air conditioning system according to the optimal solution includes: optimized control of air conditioning pre-start and optimized control of air conditioning during room usage periods;

[0111] Among them, the optimization variables in the air conditioning pre-start optimization control include air conditioning pre-start time, temperature set point, air supply speed and air supply angle. The optimization process is performed during the non-use period (early morning) of the building rooms to improve the thermal comfort and energy consumption at the initial use time of the rooms.

[0112] The optimization variables in the air conditioning system's optimized control during room usage periods include the temperature setpoint, air supply speed, and air supply angle. The optimization process is performed during the building's room usage periods.

[0113] The technical solution of this invention obtains N simulated temperature distributions and energy consumption simulation data at different times within a year; based on the room temperature distribution prediction model and energy consumption prediction model, a multi-objective particle swarm optimization algorithm is used for optimization, and a weighted sum method is used to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters. This effectively solves the problem of low control reliability of building air conditioning systems caused by existing technologies and effectively improves the control reliability of building air conditioning systems.

[0114] The technical solution of this invention is based on BIM, CFD and EnergyPlus modeling, which effectively reduces the data collection cycle and improves the efficiency of model training; moreover, based on spatiotemporal data, a room temperature distribution prediction model is established through graph attention neural network and gated recurrent unit, which improves the accuracy of room temperature distribution prediction.

[0115] In the technical solution of this invention, the nodes in the room temperature field topology diagram can be either sitting type nodes or standing type nodes for people whose stay duration is longer than the preset duration. Alternatively, the nodes can be determined by calculating the temperature gradient of the temperature distribution within the room area, which improves the flexibility of node selection.

[0116] In the technical solution of this invention, the feature vector h' of the target node i output by the aggregated graph attention neural network at time t is... i Preprocessing is performed, specifically multi-head attention processing, to improve the learning ability of the room temperature distribution prediction model.

[0117] The technical solution of this invention controls the air conditioning system according to the optimal solution, including: optimized control of air conditioning pre-start and optimized control of air conditioning during room usage periods; different optimized controls can be performed according to the usage periods of the building air conditioning, further improving the reliability and flexibility of the optimized control of the building air conditioning system.

[0118] Example 2

[0119] like Figure 8 As shown, the present invention also provides a building air conditioning system optimization control device based on spatiotemporal data, comprising:

[0120] The first module 101 is used to create a 3D BIM model of the building rooms, divide the 3D BIM model into meshes, and create a CFD model.

[0121] The first simulation module 102 defines N simulated operating conditions for air conditioning settings in the CFD model, and performs temperature simulations for the N simulated operating conditions of the CFD model for weather conditions within a year, to obtain N simulated temperature distributions at different times within a year; wherein, the air conditioning settings include air conditioning pre-start time, temperature set point, air supply speed and air supply angle.

[0122] The second module 103 establishes a room temperature distribution prediction model based on graph attention neural network and gated loop unit.

[0123] The second simulation module 104 inputs the three-dimensional BIM model into the energy consumption calculation software, performs energy consumption simulations for N simulated operating conditions of air conditioning settings for weather conditions within a year, and obtains N energy consumption simulation data at different times within a year.

[0124] The third module 105 takes the air conditioner's setting parameters as input and the energy consumption at the next moment as output, and establishes an energy consumption prediction model based on a BP neural network.

[0125] The optimization module 106 is based on the room temperature distribution prediction model and the energy consumption prediction model. It uses a multi-objective particle swarm optimization algorithm for optimization and a weighted sum method to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters.

[0126] Control module 107 controls the air conditioning system according to the optimal solution.

[0127] It should be noted that the first establishment module 101, the first simulation module 102, the second establishment module 103, the second simulation module 104, the third establishment module 105, the optimization module 106, and the control module 107 in this embodiment correspond to the method steps in Embodiment 1, and will not be described in detail here.

[0128] The technical solution of this invention obtains N simulated temperature distributions and energy consumption simulation data at different times within a year; based on the room temperature distribution prediction model and energy consumption prediction model, a multi-objective particle swarm optimization algorithm is used for optimization, and a weighted sum method is used to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters. This effectively solves the problem of low control reliability of building air conditioning systems caused by existing technologies and effectively improves the control reliability of building air conditioning systems.

[0129] The technical solution of this invention is based on BIM, CFD and EnergyPlus modeling, which effectively reduces the data collection cycle and improves the efficiency of model training; moreover, based on spatiotemporal data, a room temperature distribution prediction model is established through graph attention neural network and gated recurrent unit, which improves the accuracy of room temperature distribution prediction.

[0130] In the technical solution of this invention, the nodes in the room temperature field topology diagram can be either sitting type nodes or standing type nodes for people whose stay duration is longer than the preset duration. Alternatively, the nodes can be determined by calculating the temperature gradient of the temperature distribution within the room area, which improves the flexibility of node selection.

[0131] In the technical solution of this invention, the feature vector h' of the target node i output by the aggregated graph attention neural network at time t is... i Preprocessing is performed, specifically multi-head attention processing, to improve the learning ability of the room temperature distribution prediction model.

[0132] The technical solution of this invention controls the air conditioning system according to the optimal solution, including: optimized control of air conditioning pre-start and optimized control of air conditioning during room usage periods; different optimized controls can be performed according to the usage periods of the building air conditioning, further improving the reliability and flexibility of the optimized control of the building air conditioning system.

[0133] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for optimizing the control of a building air conditioning system based on spatiotemporal data, characterized in that it includes: Create a 3D BIM model of the building rooms, divide the 3D BIM model into meshes, and create a CFD model. In the CFD model, N simulated operating conditions with air conditioning setting parameters are defined. Temperature simulations are then performed on each of the N simulated operating conditions for weather conditions within a year to obtain N simulated temperature distributions at different times within a year. The air conditioning setting parameters include the air conditioning pre-start time, temperature set point, air supply speed, and air supply angle. A room temperature distribution prediction model is established based on a graph attention neural network and a gated recurrent unit; specifically, the establishment of the room temperature distribution prediction model based on the graph attention neural network and the gated recurrent unit includes: Establish a room temperature field topology diagram, which is composed of nodes and edges. The nodes are used to represent different selected locations in the building room, and the edges are used to represent the correlation between different nodes. Construct a time series feature matrix for each moment within a preset time period, and normalize the time series feature matrix. The time series feature matrix includes outdoor meteorological parameters, node temperatures output after temperature simulation, and relevant air conditioning settings. The time series feature matrix at a certain moment is e rows and b columns, where e represents the number of nodes and b represents the number of feature data. Establish an adjacency matrix for the room temperature field topology diagram, and determine whether any two nodes in the room temperature field topology diagram are adjacent based on the adjacency matrix. The spatial features of each node are extracted based on a graph attention neural network, and the spatial features of the node include the attention coefficients between the target node and its neighboring nodes. Calculate the attention coefficients of all neighboring nodes to the target node. The specific calculation method for the attention coefficients is as follows: in, Let be the normalized attention coefficients of target node i and neighbor node j; i is the target node; j is neighbor node j; k is neighbor node k. Use the LeakyReLU activation function; For all neighboring nodes of target node i; This is a cascade operation; A learnable parameter vector; This is the linear transformation matrix used for node feature transformation; Let i be the feature vector of the target node i; Let be the feature vector of neighbor node j; The feature vector of neighbor node k; Based on the attention coefficients of all neighboring nodes towards the target node, the spatial features of the neighboring nodes adjacent to the target node are aggregated to obtain the feature vector of the target node output by the aggregated graph attention neural network; where the expression of the feature vector of the target node output by the aggregated graph attention neural network is: in, The aggregated graph shows the feature vector of the target node output by the neural network. Use the Sigmoid activation function; Pay attention to the feature vector of the target node i output by the neural network at time t after aggregation. Preprocessing is performed, specifically multi-head attention processing. The feature vector of the target node i output by the preprocessed graph attention neural network at time t is as follows: in, Let be the feature vector of the target node i output by the preprocessed graph attention neural network at time t; M represents the number of heads in the multi-head attention; Let be the attention coefficient of the m-th head; Let be the weight matrix for the m-th head; Based on the gated loop unit, the corresponding time features are extracted from the feature vector of each target node i in the output of the aggregated graph attention neural network to obtain the room temperature distribution prediction model; wherein, the gated loop unit includes a reset gate and an update gate, and the expression of the gated loop unit specifically includes: ， ， ， ， in, To reset the gate output, To update the gate output; For the aggregated graph, note the feature vector of each target node i output by the neural network at time t. ; The weight matrix is ​​input to the reset gate; To update the weight matrix of the gate; Let be the weight matrix of the candidate hidden states; The hidden state at time t-1 The weight matrix to the reset gate; Let be the weight matrix of the update gate at time t-1; Let be the weight matrix of the candidate hidden states at time t-1; , and These are the biases for resetting the door, updating the door, and the candidate hidden state, respectively. This represents the hidden state at time t-1; tanh is the matrix element-wise product; tanh is the hyperbolic tangent function. Let t be the candidate hidden state; Let t be the hidden state at time t, the last one in the time series. The data is input into the fully connected layer to predict the temperature distribution at the next time step. Set the basic parameters of the room temperature distribution prediction model, divide the temperature distribution simulation dataset into training set and test set according to a preset ratio, and use the training set to train the room temperature distribution prediction model to obtain the trained room temperature distribution prediction model. The 3D BIM model is input into the energy consumption calculation software, and energy consumption simulations are performed on N simulated operating conditions for air conditioning settings under different weather conditions throughout the year to obtain N energy consumption simulation data at different times throughout the year. Using the air conditioner's settings as input and the energy consumption at the next moment as output, an energy consumption prediction model is established based on a BP neural network. Based on room temperature distribution prediction models and energy consumption prediction models, a multi-objective particle swarm optimization algorithm is used for optimization, and a weighted sum method is employed to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters. The multi-objective particle swarm optimization specifically includes: Initialize the particle swarm and external archive, and set the termination conditions for the multi-objective particle swarm algorithm; Calculate the fitness of each particle, which includes a thermal comfort objective function. and energy consumption objective function ; Update the historical best position pBest and the global best position gBest for each particle; Update the velocity and position of each particle, where the particle velocity and position are updated in the following way: ， ， in, , , respectively, represent the particle velocities of particle h in the (e+1)th and eth iterations; , These are the acceleration coefficients; , These are random numbers between [0, 1]. , These are the historical optimal position and the global optimal position of particle h in the e-th iteration, respectively; Inertial weight; , These are the positions of particle h in the (e+1)th and eth iterations, respectively; Update the external archive based on the updated velocity and position of each particle; The Pareto set is output until the current iteration number or the fitness function value meets the termination condition. The termination condition includes the iteration number meeting a preset maximum iteration number or the fitness function value being less than a preset fitness threshold. The air conditioning system is controlled according to the optimal solution.

2. The method for optimizing and controlling a building air conditioning system based on spatiotemporal data according to claim 1, characterized in that, In the room temperature field topology diagram, the nodes represent the sitting posture type nodes and the standing posture type nodes for people whose dwell time exceeds the preset time. The sitting posture type nodes include the room position node corresponding to the head when the person stands up from a sitting position, the room position node corresponding to the head when the person is sitting, the room position node corresponding to the chest when the person is sitting, and the room position node corresponding to the legs when the person is sitting. The standing posture type nodes include the room position node corresponding to the head when the person is standing, the room position node corresponding to the chest when the person is standing, and the room position node corresponding to the legs when the person is standing.

3. The method for optimizing and controlling a building air conditioning system based on spatiotemporal data according to claim 1, characterized in that, The nodes in the room temperature field topology diagram are determined as follows: The room temperature distribution is measured by a temperature acquisition instrument, and the temperature gradient is calculated for the temperature distribution within the room area. When the temperature gradient value corresponding to a certain location in the room is greater than the preset gradient threshold and there are no other nodes within the preset distance, then that location is a node.

4. The method for optimizing and controlling a building air conditioning system based on spatiotemporal data according to claim 1, characterized in that, heat... The comfort objective function is represented by the sum of the differences between the predicted temperatures of all target nodes and the set optimal comfort temperature, specifically: ， The energy consumption objective function is as follows: ， in: The objective function is thermal comfort. Let Q be the energy consumption objective function; Q is the total number of target nodes; Let be the predicted temperature of the q-th target node; Set the most comfortable temperature for the q-th target node; This represents the energy consumption of the air conditioning system between the current moment and the next moment.

5. The method for optimizing and controlling a building air conditioning system based on spatiotemporal data according to claim 4, characterized in that, The weighted sum method for determining the optimal solution specifically includes: Weights are assigned to each objective function to optimize the target. Based on the weights of each optimization objective function, the weighted sum of each solution is calculated, wherein the weighted sum of each solution is calculated as follows: ， in, Let z be the weighted sum of the z-th scheme; , These are the weights for the thermal comfort objective function and the energy consumption objective function, respectively. , These are the thermal comfort objective function value and the energy consumption objective function value for the z-th scheme, respectively; Sort all schemes according to the weighted sum of each scheme; The optimal solution is the one with the smallest weighted sum.

6. The method for optimizing and controlling a building air conditioning system based on spatiotemporal data according to claim 1, characterized in that, Controlling the air conditioning system according to the optimal solution includes: optimized control of air conditioning pre-start and optimized control of air conditioning during room usage periods; Among them, the optimization variables in the pre-start optimization control of air conditioning include air conditioning pre-start time, temperature set point, air supply speed and air supply angle. The optimization process is performed during the non-use period of the building rooms to improve the thermal comfort and energy consumption at the initial use time of the rooms. The optimization variables in the air conditioning system's optimized control during room usage periods include the temperature setpoint, air supply speed, and air supply angle. The optimization process is performed during the building's room usage periods.

7. A building air conditioning system optimization control device based on spatiotemporal data, characterized in that, This method is implemented based on the spatiotemporal data-based building air conditioning system optimization control method described in claim 1, and includes: The first module establishes a 3D BIM model of the building rooms, divides the 3D BIM model into meshes, and establishes a CFD model. The first simulation module defines N simulated operating conditions for air conditioning settings in the CFD model. It then performs temperature simulations on each of the N simulated operating conditions for weather conditions within a year, obtaining N simulated temperature distributions at different times within a year. The air conditioning settings include the air conditioning pre-start time, temperature set point, air supply speed, and air supply angle. The second module establishes a room temperature distribution prediction model based on graph attention neural networks and gated recurrent units. The second simulation module inputs the 3D BIM model into the energy consumption calculation software, performs energy consumption simulations for N simulated operating conditions of the air conditioning settings for weather conditions within a year, and obtains N energy consumption simulation data at different times within a year. The third module takes the air conditioner's settings as input and the energy consumption at the next moment as output, and establishes an energy consumption prediction model based on a BP neural network. The optimization module is based on the room temperature distribution prediction model and the energy consumption prediction model. It uses a multi-objective particle swarm optimization algorithm for optimization and a weighted sum method to determine the optimal solution. The optimization objectives are room thermal comfort and air conditioning energy consumption, and the decision variables are the air conditioning setting parameters. The control module controls the air conditioning system according to the optimal solution.

Images