A smart building maintenance method, device and medium based on edge computing

Abstract

The present invention provides a smart building maintenance method, equipment and medium based on edge computing, which relates to the field of smart building technology. The method includes: obtaining preset edge features and preset fault-related features, obtaining edge parameter data of equipment in the building corresponding to the preset edge features based on the preset edge features, inputting a target edge computing node acquisition model, obtaining initial edge computing node related information, obtaining intermediate data when the edge computing node is located at the initial edge computing node position, obtaining fault parameter data of equipment in the building corresponding to preset fault-related features, obtaining initial fault result information, using reinforcement learning to jointly obtain a target edge computing node model and a target fault diagnosis model, and training the combined model to obtain a target joint model, and using the target joint model to perform maintenance management on the smart building to improve the efficiency and accuracy of equipment maintenance.

Description

Technical Field

[0001] The present invention relates to the field of intelligent building technology, and in particular to an intelligent building maintenance method, equipment and medium based on edge computing. Background Art

[0002] With the rapid development of IoT technology, smart buildings, as a key area of ​​IoT application, have a large number of internal devices of various types, and the problem of equipment maintenance is becoming increasingly prominent. Smart buildings refer to a building form that optimizes and integrates the structure, system, service and management of buildings, and uses advanced technologies such as the Internet of Things (IoT), big data, and artificial intelligence to achieve efficient, intelligent, and sustainable operation and management of buildings. It can automatically sense, analyze, and respond to changes in the internal and external environment of the building, providing users with a comfortable, safe, and convenient environment, while improving energy efficiency and reducing operating costs.

[0003] At present, for smart buildings, traditional equipment maintenance methods mainly rely on manual inspections and regular maintenance. However, this method is inefficient, costly, and slow to respond, and cannot meet the needs of modern smart building management. Summary of the invention

[0004] In view of the above technical problems, the technical solution adopted by the present invention is: a smart building maintenance method based on edge computing, the method comprising the following steps:

[0005] Acquire a preset edge feature and a preset fault-related feature, wherein the preset edge feature is a preset feature related to the selected edge computing node, and the preset fault-related feature is a preset feature related to the fault;

[0006] Based on the preset edge features, edge parameter data of the equipment in the building corresponding to the preset edge features are obtained, and the edge parameter data is input into the target edge computing node acquisition model to obtain the initial edge computing node related information, wherein the initial edge computing node related information at least includes the edge computing node location information;

[0007] Acquire intermediate data when the edge computing node is located at the initial edge computing node position, wherein the intermediate data at least includes edge computing node attribute information and data flow information of the equipment in the initial building;

[0008] Based on the preset fault-related features, the fault parameter data of the equipment in the building corresponding to the preset fault-related features are obtained, and the intermediate data and the fault parameter data are input into the target fault diagnosis model to obtain the initial fault result information;

[0009] Use reinforcement learning to combine the target edge computing node acquisition model and the target fault diagnosis model, train the combined model, obtain the target joint model, and use the target joint model to maintain and manage the smart building;

[0010] Among them, the states in reinforcement learning include information related to edge computing nodes, data flow information of equipment in the building, system load information and fault result information. The actions include adjusting the position of edge computing nodes and increasing the number of edge computing nodes. The reward function is based on communication delay time, fault diagnosis accuracy and resource utilization.

[0011] The system load information is the information about the resources occupied by running the combined model, the communication delay time includes the delay time for transmitting the equipment data in the building to the edge computing node, the fault diagnosis accuracy is determined based on the fault result information output by the combined model during the training process and the real fault result information, and the resource utilization rate is the effective use of resources by the combined model during the training process;

[0012] The state initialization in reinforcement learning is: initial edge computing node related information, data flow information of equipment in the initial building, initial system load conditions, and initial fault result information. The initial system load conditions are the information on resources occupied when obtaining the initial edge computing node related information and the initial fault result information.

[0013] According to another aspect of the present invention, a non-transitory computer-readable storage medium is provided, in which at least one instruction or at least one program is stored, and the at least one instruction or the at least one program is loaded and executed by a processor to implement the aforementioned method.

[0014] According to yet another aspect of the present invention, an electronic device is provided, comprising a processor and the aforementioned non-transitory computer-readable storage medium.

[0015] The present invention has at least the following beneficial effects: In summary, preset edge features and preset fault-related features are obtained, and based on the preset edge features, edge parameter data of the equipment in the building corresponding to the preset edge features are obtained, and the edge parameter data is input into the target edge computing node acquisition model to obtain the initial edge computing node related information, and the intermediate data when the edge computing node is located at the initial edge computing node position is obtained, and based on the preset fault-related features, the fault parameter data of the equipment in the building corresponding to the preset fault-related features is obtained, and the intermediate data and the fault parameter data are input into the target fault diagnosis model to obtain the initial fault result information, and the reinforcement learning is used to jointly obtain the target edge computing node model and the target fault diagnosis model, and the combined model is trained to obtain the target joint model, and the target joint model is used to perform maintenance and management on the smart building. The present invention introduces edge computing and models to realize real-time monitoring and intelligent analysis of equipment status to improve the efficiency and accuracy of equipment maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A flowchart of a smart building maintenance method based on edge computing provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0019] The embodiment of the present invention provides a smart building maintenance method based on edge computing, such as Figure 1 As shown, the method comprises the following steps:

[0020] S100, obtaining preset edge features and preset fault-related features, wherein the preset edge features are preset features related to the selected edge computing node, and the preset fault-related features are preset features related to the fault.

[0021] S200, based on the preset edge features, obtain edge parameter data of the equipment in the building corresponding to the preset edge features, and input the edge parameter data into the target edge computing node acquisition model to obtain the initial edge computing node related information, wherein the initial edge computing node related information at least includes the edge computing node location information.

[0022] Specifically, the edge computing node is a component for implementing edge computing and is a physical or virtual device that performs edge computing tasks. The edge computing nodes are distributed at the edge of the network and collect data from various data sources.

[0023] In one embodiment of the present invention, the edge parameter data includes: temperature parameters, pressure parameters, current parameters, humidity parameters, and vibration parameters of the air-conditioning system; power parameters, load condition parameters, voltage parameters, current parameters, and power factor parameters of the power management system; and light intensity parameters and energy consumption parameters of the lighting system.

[0024] S300, obtaining intermediate data when the edge computing node is located at the initial edge computing node position, wherein the intermediate data at least includes edge computing node attribute information and data flow information of equipment in the initial building.

[0025] S400, based on preset fault-related features, obtaining fault parameter data of equipment in the building corresponding to preset fault-related features, and inputting intermediate data and fault parameter data into a target fault diagnosis model to obtain initial fault result information.

[0026] S500, use reinforcement learning to jointly acquire a target edge computing node model and a target fault diagnosis model, train the combined model, acquire a target joint model, and use the target joint model to perform maintenance and management on the smart building.

[0027] Among them, the states in reinforcement learning include information related to edge computing nodes, data flow information of equipment in the building, system load information and fault result information. Actions include adjusting the position of edge computing nodes and increasing the number of edge computing nodes. The reward function is based on communication delay time, fault diagnosis accuracy and resource utilization.

[0028] The system load information is the information about resources occupied by running the combined model. The communication delay time includes the delay time for transmitting equipment data in the building to the edge computing node. The fault diagnosis accuracy is determined based on the fault result information output by the combined model during the training process and the actual fault result information. The resource utilization rate is the degree of effective use of resources by the combined model during the training process.

[0029] The state initialization in reinforcement learning is: initial edge computing node related information, data flow information of equipment in the initial building, initial system load conditions, and initial fault result information. The initial system load conditions are the information on resources occupied when obtaining the initial edge computing node related information and the initial fault result information.

[0030] Specifically, the reward function is a weighted sum of communication delay time, fault diagnosis accuracy and resource utilization.

[0031] In summary, preset edge features and preset fault-related features are obtained, and based on the preset edge features, edge parameter data of the equipment in the building corresponding to the preset edge features are obtained, and the edge parameter data is input into the target edge computing node acquisition model to obtain the initial edge computing node related information, and the intermediate data when the edge computing node is located at the initial edge computing node position is obtained. Based on the preset fault-related features, the fault parameter data of the equipment in the building corresponding to the preset fault-related features are obtained, and the intermediate data and the fault parameter data are input into the target fault diagnosis model to obtain the initial fault result information, and the reinforcement learning is used to jointly obtain the target edge computing node model and the target fault diagnosis model, and the combined model is trained to obtain the target joint model, and the target joint model is used to perform maintenance and management on the smart building. The present invention introduces edge computing and models to realize real-time monitoring and intelligent analysis of equipment status to improve the efficiency and accuracy of equipment maintenance.

[0032] Specifically, the preset edge features are obtained through the following steps:

[0033] An initial edge feature list is obtained, where the initial edge feature list includes a plurality of initial edge features.

[0034] The preset edge features at least include: equipment density in the building, total data flow of equipment in the building. The preset edge features also include: communication delay data and bandwidth utilization, the equipment density in the building is the ratio of the number of equipment to the building area; the total data flow of equipment in the building is the product of the amount of data transmitted per unit time by the equipment and the number of equipment in the building.

[0035] A historical data list set is obtained, where the historical data list set includes a plurality of historical data lists, and the historical data lists include edge feature values ​​corresponding to a plurality of initial edge features.

[0036] Each edge feature value in the historical data list set is standardized to obtain a first intermediate data list set, wherein the first intermediate data list set includes a plurality of first intermediate data lists, and the first intermediate data list includes a plurality of first intermediate data.

[0037] Based on the first intermediate data list set, the PCA principal component analysis method is used to obtain the principal component edge features of the initial edge features and the edge distribution vectors corresponding to the principal component edge features. Specifically, it can be understood that based on the PCA principal component analysis method, the covariance matrix of the initial edge features is obtained, the eigenvalues ​​and eigenvectors of the covariance matrix are obtained, the eigenvalues ​​are sorted from large to small, the first k eigenvalues ​​are obtained, and the initial edge features corresponding to each eigenvalue in the first k eigenvalues ​​are obtained, and the initial edge features are used as the principal component edge features. Among them, k can be determined according to actual needs.

[0038] The principal component edge feature is used as the preset edge feature.

[0039] Furthermore, after obtaining the edge distribution vector corresponding to the edge feature of the principal component, it also includes:

[0040] The edge distribution vector corresponding to the principal component edge feature is input into the random forest model to obtain the weight of each preset edge feature.

[0041] In summary, an initial edge feature list is obtained, a historical data list set is obtained, and based on the first intermediate data list set, the PCA principal component analysis method is used to obtain the principal component edge features of the initial edge features and the edge distribution vector principal component edge features corresponding to the principal component edge features as preset edge features. The present invention uses the PAC principal component analysis method to obtain more important features as preset edge features, thereby more accurately calculating the position of the initial edge calculation node.

[0042] Furthermore, the target edge computing node acquisition model is obtained through the following steps:

[0043] Obtain a historical edge data list set, the historical edge data list set includes several historical edge data lists, the historical edge data list includes the location label of the historical edge computing node and the historical edge values ​​corresponding to several preset edge features. The historical edge value is standardized data with the same scale; the location label of the historical edge computing node is represented by Boolean, specifically, 1 indicates that the location is selected, and 0 indicates that it is not selected.

[0044] Use the historical edge data list set to train the initial edge computing node and obtain the target edge computing node acquisition model.

[0045] Specifically, the historical edge data list set is divided into a training set and a test set to ensure that the training and evaluation of the model can objectively reflect its performance. The model parameters are adjusted to minimize the difference between the predicted value and the true value. The hyperparameters of the model are adjusted using methods such as grid search or random search to further improve the model performance.

[0046] Specifically, the target fault diagnosis model is obtained through the following steps:

[0047] Obtain an initial fault diagnosis model.

[0048] Obtain a fault training list set, wherein the fault training list set includes fault training lists corresponding to several preset events, and the fault training list includes real fault types corresponding to the preset events and feature values ​​of preset fault-related features corresponding to the preset events; wherein the real fault types include at least a normal type, a first preset fault type, and a second preset fault type.

[0049] In one embodiment of the present invention, the first preset fault type is a mechanical fault, and the second preset fault type is a software fault.

[0050] The initial fault diagnosis model is trained using the fault training list set to obtain the target fault diagnosis model.

[0051] Specifically, select an appropriate machine learning algorithm as the initial fault diagnosis model. If the data size is small, it is recommended to use traditional machine learning models such as random forest and SVM; if the data size is large, use methods such as DNN for deeper feature learning;

[0052] In an exemplary description of the present invention, the architecture of the initial fault diagnosis model is as follows:

[0053] First hidden layer: 128 neurons, ReLU activation function;

[0054] Second hidden layer: 64 neurons, ReLU activation function;

[0055] The third hidden layer: 32 neurons, ReLU activation function;

[0056] The output layer uses the Softmax activation function for multi-classification tasks and outputs the probability of each class. The number of output nodes is equal to the number of fault classification categories (such as normal, mechanical failure, and software failure).

[0057] Specifically, the cross entropy loss function and Adam optimizer are used to train the model, evaluate the model, and tune it.

[0058] Furthermore, the preset fault related features are obtained through the following steps:

[0059] An initial fault-related feature list is obtained, where the initial fault-related feature list includes a plurality of initial fault-related features.

[0060] The initial fault-related features include at least: the mean temperature of the equipment in the building, the frequency information of the vibration of the equipment in the building. The initial fault-related features also include: the mean, standard deviation, maximum, minimum, spectrum center and spectrum bandwidth, and time-frequency domain characteristics of each parameter. The parameters can be set according to actual needs, such as the temperature, pressure, current, humidity, and vibration of the air-conditioning system.

[0061] A training fault data list set is obtained, wherein the training fault data list set includes a plurality of training fault data lists, and the training fault data lists include a plurality of training fault feature values ​​corresponding to initial fault related features.

[0062] Each training fault feature value in the training fault data list set is standardized to obtain a second intermediate data list set, wherein the second intermediate data list set includes a plurality of second intermediate data lists, and the second intermediate data lists include a plurality of second intermediate data.

[0063] Based on the second intermediate data list set, the PCA principal component analysis method is used to obtain the principal component fault-related features of the initial fault-related features and the fault distribution vector corresponding to the principal component fault-related features. Specifically, a dimensionality reduction technique such as the principal component analysis (PCA) method is used to reduce the feature dimension and retain the main information. It also includes: using the chi-square test to evaluate the independence of the feature and the target variable, and using the analysis of variance (ANOVA) to evaluate the impact of different features on the target variable. Alternatively, the recursive feature elimination (RFE) method is used to recursively train the model and remove unimportant features.

[0064] The Pearson correlation coefficient between each principal component fault-related feature and the actual fault type is obtained, and the principal component fault-related feature corresponding to the Pearson correlation coefficient greater than a preset correlation threshold is used as the preset fault-related feature.

[0065] In summary, an initial fault-related feature list is obtained, a training fault data list set is obtained, and each training fault feature value in the training fault data list set is standardized to obtain a second intermediate data list set. In the second intermediate data list set, the PCA principal component analysis method is used to obtain the principal component fault-related features of the initial fault-related features and the fault distribution vector corresponding to the principal component fault-related features, and the Pearson correlation coefficient between each principal component fault-related feature and the actual fault type is obtained. The principal component fault-related features corresponding to the Pearson correlation coefficient greater than the preset correlation threshold are used as preset fault-related features. The present invention obtains preset fault-related features through the PCA principal component analysis method and the Pearson coefficient, thereby improving the accuracy of fault prediction.

[0066] Furthermore, the edge computing nodes will schedule and allocate computing resources according to the decisions of the target joint model to avoid overloading of certain nodes or waste of idle resources. Help edge computing nodes to better utilize network bandwidth and reduce communication delays. By dynamically adjusting the location or communication strategy of the nodes, ensure that the communication between sensors and edge computing nodes is as fast and reliable as possible in different scenarios. The optimization results of DQN can ensure that edge computing nodes can quickly and accurately collect device data and run fault diagnosis models in real time to detect the health status of the equipment. The edge computing node determines the current status of the device through the diagnostic model and feeds back the diagnostic results to the central control system or equipment maintenance personnel.

[0067] Specifically, reinforcement learning is used to jointly acquire a target edge computing node model and a target fault diagnosis model, and the combined model is trained to obtain a target joint model, including:

[0068] Initialize the state in reinforcement learning and define a reward function, which is a weighted sum of communication delay time, fault diagnosis accuracy and resource utilization.

[0069] Status t = (Node Position, Data Flow, Sensor Data Rate, Resource Usage), Node Position is the edge computing node related information, Data Flow is the equipment data flow information in the building, Sensor Data Rate is the system load information, and Resource Usage is the fault result information;

[0070] Reward function r t =-α1(Delay)-α2(Fault Error Rate)+α3(Resource Efficiency),

[0071] α1 is the first coefficient used to adjust the reward weight, α2 is the second coefficient used to adjust the reward weight, α3 is the third coefficient used to adjust the reward weight, Delay is the communication delay time, Fault Error Rate is the fault diagnosis accuracy, and Resource Efficiency is the resource utilization.

[0072] Use the ε-greedy strategy to select actions and get the selected action a t .

[0073] Perform the selected action a t , get the reward r of environmental feedback t and the new state s t+1 .

[0074] Estimated Q value Q (s t , a t )=E[r t +γmaxQ(s t+1 ,a0)], E[] is the expected function, γ is the discount factor, maxQ(s t+1 ,a0) is in the next state The estimated Q value of the best action a0 is selected.

[0075] Will (s t ,a t ,r t ,s t+1 ) samples are stored in the experience pool.

[0076] The target network is used to obtain the target Q value, and the estimated Q value is optimized by gradient descent to make it close to the target Q value. Wherein, the parameters of the target network are regularly updated, and the target network can be manually preset.

[0077] Action selection, state update, and network training are repeated until the system converges.

[0078] In summary, the present invention will avoid the local optimality problem caused by single optimization in traditional models by co-optimizing factors such as communication delay, computing resource allocation, node deployment, etc. with the accuracy of fault diagnosis. This enables edge computing nodes to not only improve system performance, but also better support efficient equipment maintenance and fault prediction. Traditional fault diagnosis systems run independently on equipment data, while this model deeply integrates node location with fault prediction models, and adjusts node location and data acquisition frequency through Q value functions, thereby improving the accuracy and real-time performance of fault prediction. This design ensures that the distance between nodes and devices is minimized, reduces communication delays, and ensures the real-time performance of fault diagnosis.

[0079] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to implementing a method in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided in the above embodiment.

[0080] An embodiment of the present invention further provides an electronic device, comprising a processor and the aforementioned non-transitory computer-readable storage medium.

[0081] Although some specific embodiments of the present invention have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are only for illustration, not for limiting the scope of the present invention. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention.

Claims

1. A smart building maintenance method based on edge computing, characterized in that: The method comprises the following steps: Acquire a preset edge feature and a preset fault-related feature, wherein the preset edge feature is a preset feature related to the selected edge computing node, and the preset fault-related feature is a preset feature related to the fault; Based on the preset edge features, edge parameter data of the equipment in the building corresponding to the preset edge features are obtained, and the edge parameter data is input into the target edge computing node acquisition model to obtain the initial edge computing node related information, wherein the initial edge computing node related information at least includes the edge computing node location information; Acquire intermediate data when the edge computing node is located at the initial edge computing node position, wherein the intermediate data at least includes edge computing node attribute information and data flow information of the equipment in the initial building; Based on the preset fault-related features, the fault parameter data of the equipment in the building corresponding to the preset fault-related features are obtained, and the intermediate data and the fault parameter data are input into the target fault diagnosis model to obtain the initial fault result information; Use reinforcement learning to combine the target edge computing node acquisition model and the target fault diagnosis model, train the combined model, obtain the target joint model, and use the target joint model to maintain and manage the smart building; Among them, the states in reinforcement learning include information related to edge computing nodes, data flow information of equipment in the building, system load information and fault result information. The actions include adjusting the position of edge computing nodes and increasing the number of edge computing nodes. The reward function is obtained based on communication delay time, fault diagnosis accuracy and resource utilization. The system load information is the information about the resources occupied by running the combined model, the communication delay time includes the delay time for transmitting the equipment data in the building to the edge computing node, the fault diagnosis accuracy is determined based on the fault result information output by the combined model during the training process and the real fault result information, and the resource utilization rate is the effective use of resources by the combined model during the training process; The state initialization in reinforcement learning is: initial edge computing node related information, data flow information of equipment in the initial building, initial system load conditions, and initial fault result information. The initial system load conditions are the information on resources occupied when obtaining the initial edge computing node related information and the initial fault result information.

2. The edge computing-based smart building maintenance method according to claim 1, characterized in that: Get the preset edge features by following the steps below: Acquire an initial edge feature list, wherein the initial edge feature list includes a plurality of initial edge features; Acquire a historical data list set, wherein the historical data list set includes a plurality of historical data lists, and the historical data lists include edge feature values ​​corresponding to a plurality of initial edge features; Performing standardization processing on each edge feature value in the historical data list set, thereby obtaining a first intermediate data list set, wherein the first intermediate data list set includes a plurality of first intermediate data lists, and the first intermediate data list includes a plurality of first intermediate data; Based on the first intermediate data list set, a principal component analysis method is used to obtain principal component edge features of the initial edge features and edge distribution vectors corresponding to the principal component edge features; The principal component edge feature is used as the preset edge feature.

3. The edge computing-based smart building maintenance method according to claim 2, characterized in that: After obtaining the edge distribution vector corresponding to the principal component edge feature, it also includes: The edge distribution vector corresponding to the principal component edge feature is input into the random forest model to obtain the weight of each preset edge feature.

4. The edge computing-based smart building maintenance method according to claim 2, characterized in that: The preset edge features include at least: equipment density in the building and total data flow of equipment in the building.

5. The edge computing-based smart building maintenance method according to claim 1, characterized in that: Obtain the target fault diagnosis model through the following steps: Obtaining an initial fault diagnosis model; Obtain a fault training list set, the fault training list set including a number of fault training lists corresponding to preset events, the fault training list including a real fault type corresponding to the preset event and a feature value of a preset fault-related feature corresponding to the preset event; wherein the real fault type includes at least a normal type, a first preset fault type, and a second preset fault type; The initial fault diagnosis model is trained using the fault training list set to obtain the target fault diagnosis model.

6. The edge computing-based smart building maintenance method according to claim 5, characterized in that: Obtain the preset fault-related features through the following steps: Acquire an initial fault related feature list, wherein the initial fault related feature list includes a plurality of initial fault related features; Acquire a training fault data list set, wherein the training fault data list set includes a plurality of training fault data lists, and the training fault data lists include a plurality of training fault feature values ​​corresponding to initial fault related features; Performing standardization processing on each training fault feature value in the training fault data list set, thereby obtaining a second intermediate data list set, wherein the second intermediate data list set includes a plurality of second intermediate data lists, and the second intermediate data lists include a plurality of second intermediate data; Based on the second intermediate data list set, a principal component analysis method is used to obtain the principal component fault-related features of the initial fault-related features and the fault distribution vector corresponding to the principal component fault-related features; The Pearson correlation coefficient between each principal component fault-related feature and the actual fault type is obtained, and the principal component fault-related feature corresponding to the Pearson correlation coefficient greater than a preset correlation threshold is used as the preset fault-related feature.

7. The edge computing-based smart building maintenance method according to claim 6, characterized in that: The preset fault-related characteristics include at least: the average temperature of the equipment in the building and the frequency information of the vibration of the equipment in the building.

8. The edge computing-based smart building maintenance method according to claim 1, characterized in that: The reward function is a weighted sum of communication delay time, fault diagnosis accuracy and resource utilization.

9. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the edge computing-based smart building maintenance method as described in any one of claims 1-8.

10. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 9.

Images