Building intelligent inspection method and system based on AI algorithm

Abstract

The invention provides an AI algorithm-based intelligent building inspection method and system. The method comprises the steps of obtaining first information, second information, third information and fourth information; performing sensing data feature extraction according to the first information and the second information to generate a sensing data feature matrix; according to the first information and the third information, identifying and generating an equipment state feature vector according to the equipment state; performing anomaly detection according to the sensing data feature matrix and the equipment state feature vector to obtain an anomaly detection result; constructing a fault diagnosis model according to the fourth information to generate a fault diagnosis report; and performing path planning processing according to the first information and the fault diagnosis report to generate an inspection scheme. According to the invention, by adopting wavelet transformation, multi-scale feature extraction, the graph neural network and a dynamic clustering algorithm, the abnormal states of equipment and environment in a building can be accurately identified, and the comprehensiveness and accuracy of anomaly detection are improved by combining cross-layer information propagation and correlation modeling.

Description

Technical Field

[0001] The present invention relates to the field of intelligent inspection technology, and in particular to an AI algorithm-based intelligent building inspection method and system. Background Art

[0002] In the current field of building management and inspection, with the continuous development of intelligence and digitalization, the demand for automated management of multi-story complex buildings is growing. A large number of sensors, cameras and other equipment are deployed in modern buildings, which can collect environmental data, equipment status and energy consumption information in real time. These data provide the basis for intelligent inspection of buildings. However, existing building inspection technologies usually rely on rule-driven manual inspections and simple static data analysis methods. These methods have the following problems: First, rule-driven analysis cannot dynamically adapt to the complex spatial structure and environmental changes of buildings, and it is difficult to detect potential anomalies and faults in time; second, the existing methods lack deep mining of the correlation between multimodal data (such as sensor data and image data), resulting in low anomaly detection accuracy; finally, for the inspection path planning of multi-story buildings, most methods fail to consider the priority and energy consumption constraints of cross-layer equipment, resulting in low inspection efficiency and high maintenance costs. Existing technologies attempt to solve the above problems through fixed rules or simple statistical methods, but their effects are limited in complex multi-story building scenarios and cannot meet the requirements of efficiency and accuracy of intelligent inspection.

[0003] Based on the shortcomings of the existing technology, there is an urgent need for an intelligent building inspection method and system based on AI algorithm. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent building inspection method and system based on AI algorithm to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides an intelligent building inspection method based on an AI algorithm, comprising:

[0006] Acquire first information, second information, third information and fourth information, wherein the first information is a building topology map, the building topology map includes the spatial structure of the building, the hierarchical interconnection relationship and the equipment deployment location, the second information includes the sensor deployment location and sensor data, the third information includes the camera deployment location and image information, and the fourth information includes the historical log data of the equipment;

[0007] Extract sensor data features based on the first information and the second information, perform multi-scale analysis on the sensor data using wavelet transform, and perform feature aggregation on sensor data of different regions in combination with the spatial position relationship of the building to generate a sensor data feature matrix;

[0008] Identify the device state based on the first information and the third information, use a preset deep learning model to identify the operating state and appearance changes of the device, and generate a device state feature vector;

[0009] Anomaly detection is performed based on the sensor data feature matrix and the device state feature vector, and the spatial association between devices is modeled using a graph neural network to identify the anomaly propagation path and impact range, thereby obtaining anomaly detection results;

[0010] Building a fault diagnosis model according to the fourth information, and using the abnormality detection result as an input value of the fault diagnosis model, and generating a fault diagnosis report by analyzing the time series and spatial association between abnormal points;

[0011] Path planning is performed based on the first information and the fault diagnosis report, and an inspection plan is generated in combination with the priority of abnormal equipment, the connectivity between floors and energy consumption constraints. The inspection plan includes an optimal inspection path map and corresponding maintenance recommendations.

[0012] In the second aspect, the present application also provides an AI algorithm-based intelligent building inspection system, including:

[0013] an acquisition module, configured to acquire first information, second information, third information, and fourth information, wherein the first information is a building topology map, the building topology map including the spatial structure, hierarchical interconnection relationship, and device deployment location of the building, the second information including the sensor deployment location and sensor data, the third information including the camera deployment location and image information, and the fourth information including historical log data of the device;

[0014] an extraction module, configured to extract sensor data features according to the first information and the second information, perform multi-scale analysis on the sensor data using wavelet transform, and perform feature aggregation on sensor data of different regions in combination with the spatial position relationship of the building to generate a sensor data feature matrix;

[0015] an identification module, configured to identify the device state according to the first information and the third information, identify the operating state and appearance change of the device using a preset deep learning model, and generate a device state feature vector;

[0016] A detection module, used to perform anomaly detection based on the sensor data feature matrix and the device state feature vector, use a graph neural network to model the spatial association between devices, identify the anomaly propagation path and impact range, and obtain anomaly detection results;

[0017] A diagnosis module, configured to construct a fault diagnosis model according to the fourth information, and use the abnormality detection result as an input value of the fault diagnosis model, and generate a fault diagnosis report by analyzing the time series and spatial association between abnormal points;

[0018] A planning module is used to perform path planning processing according to the first information and the fault diagnosis report, and generate an inspection plan in combination with the priority of abnormal equipment, the interconnection relationship between floors and energy consumption constraints. The inspection plan includes an optimal inspection path map and corresponding maintenance suggestions.

[0019] The beneficial effects of the present invention are:

[0020] The present invention can accurately identify abnormal conditions of equipment and environment in buildings by adopting wavelet transform, multi-scale feature extraction, graph neural network and dynamic clustering algorithm, and improve the comprehensiveness and accuracy of abnormality detection by combining cross-layer information propagation and association modeling. In terms of inspection path planning, the present invention combines building topology, equipment priority and energy consumption constraints, and uses reinforcement learning and multi-objective optimization algorithms to generate inspection paths with optimal energy consumption, significantly improving inspection efficiency and reducing maintenance costs. In addition, the present invention designs hierarchical feature aggregation and dynamic clustering algorithms to address the uneven distribution of equipment and the complexity of floor connections in multi-story complex buildings. It has good scene adaptability and can flexibly meet the needs of buildings of different sizes and layouts. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 It is a schematic diagram of the flow of the intelligent building inspection method described in an embodiment of the present invention;

[0023] Figure 2 Schematic diagram of the structure of the intelligent building inspection system described in the embodiment of the present invention

[0024] Figure 3 It is a schematic diagram of the structure of the intelligent building inspection equipment described in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0027] Embodiment 1:

[0028] This embodiment provides a building intelligent inspection method based on AI algorithm, see Figure 1 , the figure shows that the method includes steps S100 to S600.

[0029] Step S100, obtaining first information, second information, third information and fourth information, wherein the first information is a building topology map, the building topology map includes the spatial structure of the building, the hierarchical interconnection relationship and the equipment deployment location, the second information includes the sensor deployment location and sensor data, the third information includes the camera deployment location and image information, and the fourth information includes the historical log data of the device;

[0030] It is understandable that the building topology map is generated through spatial modeling technology and topology analysis algorithm, which not only provides the physical layout of the building, but also clarifies the specific distribution of equipment in three-dimensional space and the connection between floors, such as stairs, elevators and ventilation ducts. Through the topology map, the association relationship between devices can be efficiently constructed, providing an accurate spatial reference for data integration. Sensor data includes the deployment location of the sensor and the real-time information collected by the sensor, covering various physical parameters such as temperature and humidity, gas concentration, and energy consumption. Preferably, in order to ensure the validity and availability of the data, the steps combine heterogeneous data integration technology, time alignment algorithm and signal correction mechanism to process different types of sensor data and solve possible timestamp errors, signal noise and data loss problems. The camera information provides intuitive visual monitoring capabilities for the environment and equipment operation status through the combination of deployment location and image data. The field of view modeling and image coordinate mapping methods are used to ensure that the camera's shooting area can accurately correspond to the physical space of the building, so that the video stream and the spatial layout are associated one by one, which is convenient for subsequent analysis. The equipment history log provides the historical status and change trend data of equipment operation. This data is processed through log parsing and cleaning technology to ensure the integrity and consistency of the information, and extract key features such as equipment operation cycle and fault history to support comprehensive understanding of building status.

[0031] Step S200, extracting sensor data features according to the first information and the second information, performing multi-scale analysis on the sensor data using wavelet transform, and performing feature aggregation on sensor data of different regions in combination with the spatial position relationship of the building to generate a sensor data feature matrix;

[0032] This step first uses the first information to clarify the distribution and regional division relationship of sensors in the building space, ensuring that the data of each sensor can be correctly attributed in space. Then, based on the second information, the data is analyzed at multiple scales through wavelet transform, and the time series signal is decomposed into components of different frequency bands using wavelet decomposition technology, so as to capture various change patterns in the sensor data, including low-frequency trends, periodic fluctuations, and high-frequency anomalies. This multi-scale feature extraction can not only identify long-term trends, but also keenly capture short-term sudden changes, improving the ability to understand the complex dynamics of sensor data. On this basis, combined with the spatial position relationship of the building, the sensor data of different regions are feature aggregated. Feature aggregation introduces a spatial weighting mechanism to construct a weighted feature matrix based on the physical adjacency and functional correlation of sensors in space. Specifically, the spatial correlation weights between sensors are calculated through the building topology map, and the feature values ​​of sensors in the same area are integrated using the weighted average method. The aggregated results generate spatially weighted comprehensive features on each floor or specific area to form a sensor data feature matrix.

[0033] Step S300: Identify the device state according to the first information and the third information, use a preset deep learning model to identify the operating state and appearance change of the device, and generate a device state feature vector;

[0034] The core of equipment status recognition is to analyze image data using a preset deep learning model. Therefore, this step extracts features of the equipment display (such as numerical readings), indicator lights (such as color changes, flashing states), and overall equipment appearance (such as surface damage, degree of contamination) based on a model based on a convolutional neural network (CNN) for the equipment's operating status. For dynamic changes in the operating status, the operating status switching frequency and stability are analyzed by combining the continuous frame information of the time series using models such as the long short-term memory network (LSTM). On the other hand, image difference and edge detection techniques are used to capture subtle changes in the equipment's appearance, such as corrosion, cracks, or surface wear.

[0035] Step S400: perform anomaly detection based on the sensor data feature matrix and the device state feature vector, use the graph neural network to model the spatial association between devices, identify the anomaly propagation path and impact range, and obtain the anomaly detection result;

[0036] It should be noted that the spatial association between devices is defined by the first information, and the physical distance, functional association, and data similarity between devices are quantified by constructing an adjacency matrix. The construction of a spatial graph can capture potential abnormal associations between devices. For example, sensor anomalies between adjacent devices are propagative, and the connectivity of devices between floors may amplify the impact of specific faults. The graph neural network dynamically updates the anomaly score of each device by aggregating node features layer by layer. Specifically, the sensor data feature matrix and the device state feature vector are used as the initial node features, combined with the spatial graph adjacency relationship, to propagate and aggregate abnormal feature signals layer by layer to identify the association between devices. In this way, the propagation path of the anomaly in the spatial dimension can be detected and its scope of influence can be quantified. The advantage of the graph neural network is that it can capture the nonlinear relationship between high-dimensional features, which is particularly suitable for the association analysis of equipment and environmental data in multi-story complex buildings.

[0037] Step S500: construct a fault diagnosis model according to the fourth information, and use the abnormality detection result as the input value of the fault diagnosis model, and generate a fault diagnosis report by analyzing the time series and spatial association between abnormal points;

[0038] In order to deeply analyze the causal relationship and propagation logic between these anomalies, the diagnostic model constructs the temporal and spatial correlation structure of the anomalies through a dynamic Bayesian network. In the temporal dimension, the model uses causal inference technology to analyze the time series relationship between anomalies and explore the temporal pattern and triggering sequence of fault propagation. In the spatial dimension, combined with the spatial association weights of the equipment in the building topology map, the functional linkage and physical proximity between the anomalies are identified, thereby capturing the spatial characteristics of abnormal propagation. The fault diagnosis model uses deep learning classification models (such as Transformer or LSTM) to further enhance the ability to predict fault types. Based on the comprehensive feature vectors of the anomalies, the model classifies potential fault modes and predicts possible fault types and their probability distributions. Combining time series and spatial associations, the model can also generate a key cause chain of faults, clarifying the origin of the fault, the main propagation path, and possible secondary effects.

[0039] Step S600: perform path planning processing according to the first information and the fault diagnosis report, and generate an inspection plan in combination with the priority of abnormal equipment, the interconnection relationship between floors and energy consumption constraints. The inspection plan includes an optimal inspection path map and corresponding maintenance suggestions.

[0040] It should be noted that the first step of path planning is to sort the inspection targets based on fault priority. Using the multi-attribute decision-making algorithm, the target devices are prioritized and the optimized inspection target sequence is generated by comprehensively considering the abnormality of the equipment, the scope of fault propagation, and the criticality of the equipment to the building operation. Next, the shortest path algorithm is used to initialize the path based on the spatial position and connectivity relationship in the building topology to ensure that the inspection path has the shortest travel while covering all target points. In order to further optimize the path to meet the energy consumption constraint, the reinforcement learning algorithm is used for dynamic path adjustment. The energy consumption constraint model calculates the total energy consumption of the path based on the connectivity between floors (such as elevator consumption, walking time, etc.), which is used as the optimization target of reinforcement learning. In the reinforcement learning environment, each path selection adjusts the strategy according to the reward function, and finally converges to the optimal inspection path that covers all target devices and has the lowest energy consumption. Finally, maintenance recommendations are generated based on the inspection path and equipment status. The maintenance recommendations generate specific operation instructions (such as replacing parts, cleaning equipment, or adjusting parameters) based on the fault type, equipment characteristics, and priority allocation in the diagnostic report. At the same time, the optimal path map presents the inspection route and key task nodes in a visual way, making it easier for inspection personnel to perform their tasks efficiently.

[0041] Further, step S200 includes step S210 to step S240.

[0042] Step S210: performing time series smoothing processing according to the second information, and obtaining preliminary multi-scale features by multi-scale decomposition and denoising of the data;

[0043] Since time series data are usually affected by many factors, such as instantaneous failures of equipment, environmental fluctuations or external interference, these noises may mask the real trend or periodic changes, so smoothing is required. Therefore, in this step, the time series is decomposed into multiple frequency bands through wavelet transform to identify the changes in different frequency components. In this way, high-frequency noise and low-frequency trends can be separated to ensure that the most important information in the signal is retained. Specifically, wavelet decomposition can not only remove those fluctuations that are irregular in the short term (such as occasional reading errors of sensors), but also highlight long-term trends and periodic characteristics (such as seasonal changes in temperature and humidity in buildings or the normal operation status of equipment).

[0044] Step S220: performing hierarchical feature decomposition according to the preliminary multi-scale features and the first information, enhancing the features of the sensor data in different regions by spatial position association, and obtaining a spatially weighted hierarchical feature matrix, wherein each row in the hierarchical feature matrix represents a comprehensive feature of a level;

[0045] Therefore, in this step, hierarchical feature decomposition ensures that the sensor data of adjacent areas or floors can be reasonably integrated in the feature matrix by considering the functional association and physical proximity between floors, areas or equipment. The feature matrix of each row is enhanced in space and level, so that the data of each area shows clearer spatial and functional characteristics. The hierarchical feature matrix is ​​expressed as:

[0046]

[0047] Among them, S l,m represents the comprehensive feature value of floor l in the mth feature dimension; T :,m represents the multi-scale eigenvalues ​​of all sensors in the mth feature dimension; W i,: represents the spatial association weight between sensor i and other sensors; P l,: represents the cross-layer propagation weight between floor l and other floors; represents the sensor set on floor l; i represents the sensor number; l represents the floor number; m represents the feature dimension number; represents the eigenvalue of floor k in the mth feature dimension in the previous iteration; P l,k represents the information propagation weight between floor l and floor k; L represents the total number of floors; t-1 represents the previous iteration; α represents the cross-layer propagation adjustment coefficient; σ is the nonlinear activation function.

[0048] Step S230, performing feature aggregation according to the hierarchical feature matrix, capturing feature distribution patterns through a dynamic clustering algorithm and identifying feature distributions of similar regions, and obtaining regional feature aggregation results;

[0049] This step uses a clustering algorithm to achieve intelligent grouping of data from different regions, allowing the system to automatically identify and summarize regions with similar characteristics in the building. Compared with traditional manual region division or static analysis methods, dynamic clustering methods are more flexible and efficient, and can adaptively handle complex and changing environments and data patterns in building spaces. The dynamic clustering algorithm formula is as follows:

[0050]

[0051] Among them, R c represents the comprehensive feature vector of the cth cluster; C c represents the floor index set belonging to the cth cluster; c represents the cluster number; S l,: represents the feature vector of floor l; S k,: represents the feature vector of floor k; λ is the feature balance factor; K represents the total number of clusters.

[0052] Step S240: globally model the cross-layer features according to the regional feature aggregation results, perform dimensionality reduction processing on the cross-layer features through principal component analysis and extract key features to generate the final sensor data feature matrix.

[0053] Since the regional feature aggregation results have completed the integration of sensor data in spatial and hierarchical dimensions, these features still have high-dimensional redundancy problems, especially in multiple floors or areas, some features are highly correlated. PCA maps data from high-dimensional space to low-dimensional space through linear transformation, while retaining the variability of the data (i.e., the amount of information) to the greatest extent. In this step, PCA compresses the feature dimensions from the original high-dimensional data space to a lower dimension by identifying the direction of maximum variance between cross-layer features, eliminating redundant information, and extracting the principal components that best represent data changes. Through this dimensionality reduction process, the sensor data feature matrix finally generated not only reduces the dimension of the data, but also ensures that the key features contained in the feature matrix can most accurately describe the overall operating status and environmental changes of the building. These key features can reflect the main association patterns between equipment and environment, and improve the training efficiency and performance of subsequent models.

[0054] Embodiment 2:

[0055] The difference between this embodiment and Embodiment 1 is that step S300 includes steps S310 to S340.

[0056] Step S310: performing image segmentation processing on the video image according to the third information, segmenting the device area in the video frame and removing background and interference information to obtain a device image sequence;

[0057] Preferably, in order to improve the segmentation effect, the location and boundaries of the device are accurately identified by combining different image features, such as color, texture, shape or motion information. For example, for HVAC (heating, ventilation and air conditioning) systems, lighting equipment or elevators in buildings, the shape and color characteristics of the equipment are more significant, and using this information can help the algorithm to more accurately divide the device area. After image segmentation, the obtained device image sequence can accurately represent the location and state changes of the device in the video frame, eliminating the interference of background clutter information.

[0058] Step S320: perform spatial feature calibration according to the device image sequence and the first information, perform spatial calibration on the device in the image, and establish a corresponding relationship between the image feature and the building space position of the device to obtain the spatial positioning feature of the device;

[0059] In this process, the spatial calibration technology in computer vision is used to extract key features in the device image (such as the edges, corners or specific identifiers of the device) and match them with the actual coordinates in the building topology map to accurately calculate the spatial position of the device. Spatial positioning features can help the system understand the exact location of the device in the building and ensure that subsequent device status analysis (such as anomaly detection or fault diagnosis) can be accurately associated with the location of the device. For example, through spatial calibration, it can be ensured that when a device failure is detected, the system can accurately identify the floor, area or even specific room of the device, thereby optimizing the inspection path and maintenance tasks.

[0060] Step S330: Identify the running state of the device according to the spatial positioning characteristics of the device, model the state changes in the continuous frames through the time series analysis method, extract the switching frequency and stability information of the device, and obtain the state change information of the device;

[0061] Specifically, the change of the operating state of the equipment is usually manifested as a series of time series data. Especially in the complex environment of multi-story buildings, the state of the equipment is affected by many factors (such as usage load, environmental conditions, equipment aging, etc.). By analyzing the continuous frames in the equipment image sequence, the state change of the equipment can be captured. For example, the state of the equipment may be manifested as the switching of the switch state (such as the equipment from off to on, or from normal operation to abnormal state), or the change of the appearance of the equipment (such as the change of the indicator light color, the change of the screen display). These state changes are modeled through time series data, and their switching frequency and stability are further analyzed. The long short-term memory network can effectively capture the time dependence of the equipment state, not only identify the current state of the equipment, but also extract the laws and patterns of state changes from historical data. For example, frequent state switching may mean that there are potential faults or unstable factors in the equipment, while stable operation indicates that the equipment is in a normal state. Furthermore, the model of the long short-term memory network can learn the time series characteristics of the equipment state change, predict the possible state transition of the equipment, and detect potential faults or abnormalities in time.

[0062] Step S340: Identify changes in device appearance based on the state change information and the device image sequence, detect changes in device appearance and perform feature fusion in combination with operating state features to generate a device state feature vector.

[0063] The changes in the operating status and appearance of equipment are often complementary, and changes in the status of equipment often leave traces on the appearance (such as temperature rise causing deformation of the equipment shell, or abnormal operation of mechanical parts causing wear). Therefore, by combining the changes in the appearance of the equipment with its operating status characteristics (such as switch status, temperature changes, energy consumption changes, etc.), a more comprehensive equipment status assessment model can be formed.

[0064] Embodiment 3:

[0065] The difference between this embodiment and Embodiment 1 or 2 is that step S400 includes steps S410 to S440.

[0066] Step S410: Perform feature fusion processing according to the sensor data feature matrix and the device state feature vector, map the features of different modes to a unified space through feature splicing and normalization processing, and dynamically enhance the mutation trend and operation state change features to obtain a multi-dimensional feature description;

[0067] First, the sensor data feature matrix and the device state feature vector come from different data sources. Sensor data involves environmental parameters (such as temperature, humidity, gas concentration, etc.), while device state features represent the operating status, appearance changes, and other operating parameters of the device. Since these two types of data differ in feature dimensions, dimensions, and value ranges, they need to be mapped to a unified feature space through methods such as feature splicing and normalization.

[0068] Feature splicing refers to merging the feature vectors of sensor data and device status in sequence to form a unified feature vector. The spliced ​​vector can contain the feature information of both environmental data and device data. Next, through normalization, the features of different data sources are standardized so that they have the same scale. This process ensures that data of different modalities will not affect the training effect of the model due to different scales, and is also convenient for subsequent algorithms (such as machine learning models or deep learning models) to process.

[0069] In particular, the feature fusion process needs to dynamically strengthen the mutation trend and operating state change characteristics. The mutation trend refers to the sudden change in the data, such as abnormal startup of equipment or sudden changes in environmental parameters. Such changes are often signals of potential failures or abnormalities. By strengthening these mutation features, the system can more keenly identify abnormal fluctuations in the equipment or environment, thereby improving the fault warning capability. In addition, the operating state change feature can reflect the transition of the equipment from one state to another, such as the switching of the equipment between normal operation and fault state. Dynamically combining these two types of features can provide richer inputs for subsequent models in multiple dimensions.

[0070] Step S420: perform outlier detection based on the multidimensional feature description, evaluate the degree of isolation of each device feature point by establishing a random segmentation tree of feature distribution, set a dynamic threshold to filter outliers based on the degree of outliers in the feature space, and obtain a preliminary outlier set, which includes a device ID and a preliminary outlier score;

[0071] Random partition tree is an anomaly detection method based on the isolation of data points. Its basic principle is to generate a tree structure by randomly partitioning the data space. The depth and structure of the tree reflect the degree of isolation of the data points. If the position of a feature point (device status feature) in the feature space is far away from other points, it means that the point is highly isolated in the feature space, which usually indicates that the point is an anomaly. By constructing a random partition tree, it is possible to effectively identify those device features that are significantly different from other data points, and further mark these points as potential anomalies.

[0072] On this basis, the degree of outliers refers to the deviation of the position of the device in the feature space from the feature distribution of other devices. By analyzing the difference between the device features and the overall data distribution, it is possible to further evaluate whether it is an anomaly. Combined with the degree of outliers, anomalies can be screened by setting dynamic thresholds. The setting of dynamic thresholds is based on the feature distribution of the device, that is, by learning from the training data, the criteria for determining anomalies are dynamically adjusted. For example, for some devices, the threshold may be loose, while for high-precision or high-load devices, the threshold may be stricter. This flexible dynamic threshold setting can adapt to the characteristics of different devices and environments and improve the accuracy of anomaly detection. Through this process, a preliminary set of anomaly points is generated, including each device ID that is judged to be abnormal and its corresponding preliminary anomaly score. The anomaly score reflects the severity of the device anomaly. The higher the score, the greater the possibility of device anomaly.

[0073] Step S430: Anomaly propagation modeling is performed based on the preliminary abnormal point set. The abnormal nodes are input into the graph neural network, and the abnormal signals are propagated layer by layer in combination with the topological structure of the building. After the feature update and aggregation mechanism of the network layer, an abnormal path diagram of cross-layer propagation is obtained;

[0074] In practical applications, graph neural networks can characterize the physical proximity, functional associations, and data similarities between devices through adjacency matrices, and achieve accurate diffusion modeling of abnormal signals. For example, when an abnormality occurs in a key device on a floor, the signal may spread to other floors through channels such as elevators or ventilation ducts. During the propagation modeling process, graph neural networks can dynamically adjust the propagation path and signal strength according to the device type and its associated weights, thereby accurately locating the chain of abnormal propagation and the affected range. In this way, this step realizes abnormal correlation analysis in complex cross-layer environments and generates a visual abnormal propagation path map, providing patrol personnel with an intuitive display of fault propagation and a clear basis for priority judgment.

[0075] Step S440: classify the abnormalities according to the abnormal path graph, use the spectral clustering algorithm to group the device nodes in the path and mark the abnormal priorities, and obtain the abnormality detection result.

[0076] It should be noted that the core of the spectral clustering algorithm is to use the Laplace matrix of the graph to perform feature decomposition on the correlation of device nodes, divide device nodes with similar behavior patterns into the same group through clustering, and mark the abnormal priority of the device according to the characteristics of abnormal propagation. In this embodiment, spectral clustering can flexibly respond to the diverse abnormal patterns between devices in complex buildings, such as temperature and humidity anomalies on the same floor and energy consumption anomalies between devices across floors. The key device nodes most relevant to the source of the anomaly can be identified by analyzing the node weights and edge weights of the path graph. In addition, the grouping of nodes by this method can not only effectively narrow the scope of problem investigation, but also determine the processing order of key equipment or areas through priority labeling, avoiding resource waste and missing potential risks.

[0077] Embodiment 4:

[0078] The difference between this embodiment and any one of Embodiments 1 to 3 is that step S500 includes steps S510 to S540.

[0079] Step S510: Perform historical data modeling according to the fourth information, extract the evolution pattern of the equipment operation state through time series analysis, and obtain the historical characteristics of the equipment operation state;

[0080] Preferably, the analysis methods used include autoregressive moving average model (ARIMA), long short-term memory network (LSTM) or wavelet transform, etc. These methods can efficiently process multi-dimensional time series data, and are particularly suitable for complex scenarios where device states frequently switch or change abnormally.

[0081] Step S520: Perform time series anomaly correlation analysis based on anomaly detection results and historical features, use a dynamic Bayesian network to take the historical state of the device as the state node, take the anomaly propagation path as the edge weight, and evaluate the impact of each anomaly point on the state change of other devices to obtain an anomaly causal chain;

[0082] Dynamic Bayesian network is a time series analysis method based on probabilistic graphical models, which can effectively model the temporal and spatial associations between devices. In this step, the historical state of the device is defined as a state node, and the abnormal propagation path is quantified as the edge weight in the network, so as to achieve a quantitative evaluation of how the abnormal signal is transmitted between devices and its impact range. For example, when a device has a high temperature abnormality, the signal may affect the operating status of other devices through physical connections or functional associations. Through the dynamic Bayesian network, the energy consumption abnormalities or alarm behaviors of other devices caused by the high temperature abnormality can be tracked, and their cause-effect relationships can be clarified.

[0083] Step S530: predict the fault type according to the abnormal causal chain, construct a fault feature vector using causal nodes and propagation relationships, and identify the fault mode through a deep learning classification model to obtain a prediction result, which includes the fault type and its probability distribution;

[0084] It is understandable that deep learning classification models, preferably convolutional neural networks or long short-term memory networks, are used to learn the nonlinear mapping relationship between these high-dimensional features and fault modes. The model can accurately identify different types of fault modes (such as sensor failure, equipment aging, or network communication anomalies) through training on a large amount of historical data, and generate a probability distribution of fault types based on the input features. This method can not only output the most likely fault type, but also evaluate the probabilities of other possible types, providing a comprehensive reference for maintenance decisions. For example, when a device shows abnormal high temperature and the propagation chain shows that the energy consumption of its adjacent devices has risen sharply, the classification model may predict that its fault type is a cooling system failure with a probability of 85%, and at the same time suggest other possible causes, such as sensor failure (probability 10%) or external environmental interference (probability 5%).

[0085] Step S540: perform fault impact assessment based on the prediction results, quantify the impact of the fault on building operation through correlation analysis, and generate a fault diagnosis report.

[0086] Specifically, the key steps in the evaluation process include: 1. Determine the main fault types based on the probability distribution of fault prediction, and identify the key equipment and areas that may be affected in combination with the abnormal propagation path; 2. Quantify the scope of impact based on the importance of the equipment in building operation (such as core equipment or non-core equipment) and the transmission effect of its failure on the overall system; 3. Through dynamic evaluation of system performance indicators (such as temperature and humidity stability, energy efficiency, and operational reliability), clarify the secondary problems or long-term risks that may be caused by the failure. For example, the failure of the HVAC system (heating, ventilation and air conditioning) may directly lead to the failure of temperature and humidity control, while indirectly increasing the load of the power system, thereby affecting the overall energy consumption and comfort of the building.

[0087] Further, step S600 includes step S610 to step S640.

[0088] Step S610: Prioritize inspection targets according to the first information and the fault diagnosis report, calculate the relative proximity of each abnormal device to the positive ideal solution and the negative ideal solution using an approximate ideal solution algorithm, and generate a priority ranking table;

[0089] It should be noted that this step first extracts multiple evaluation indicators, including the severity of the equipment's fault, propagation range, importance, and maintenance complexity, by combining the building topology information and the fault diagnosis report. The evaluation indicators are normalized to form a unified evaluation system to avoid the influence of unit or magnitude differences on the sorting results. The core of the approximate ideal solution algorithm is to define the positive ideal solution (optimal state) and the negative ideal solution (worst state), which respectively represent the optimal value and worst value of each indicator of the equipment under ideal conditions. The algorithm determines the relative proximity of the equipment by calculating the Euclidean distance between each device and the positive ideal solution and the negative ideal solution. The higher the proximity, the higher the priority of the equipment. This method can scientifically and quantitatively evaluate the urgency and inspection value of each device, and the generated priority ranking table clearly marks the equipment and areas that need priority inspection.

[0090] Step S620: Initialize the path according to the priority sorting table, map the location of the inspection target device to the topology map node, and use the shortest path algorithm to calculate the preliminary inspection path covering all target devices;

[0091] This step ensures the adaptability of the inspection path in complex multi-story buildings by combining priority sorting with spatial layout, avoiding inefficiencies caused by unnecessary repetition or long-distance movement.

[0092] Step S630: Optimize the path according to the preliminary inspection path, dynamically adjust the path selection strategy by building a reinforcement learning model with the goal of minimizing the total energy consumption of the path, and generate an energy-optimized path;

[0093] Step S640: Output the final inspection plan based on the energy-optimal path through path labeling and task allocation.

[0094] Specifically, first, the path is marked according to the optimized energy-efficient path, and the specific location, priority, and operation requirements of each node (target device) in the inspection route are clarified. Path marking is combined with building topology information to intuitively display the key areas and floors that inspectors need to pass through, marking the main nodes in the path and the movement methods between them (such as stairs or elevators). Secondly, based on the task allocation of the inspection plan, the inspection tasks are reasonably decomposed and allocated to different inspectors or groups to make full use of human resources. Task allocation takes into account equipment priority, inspection path coverage, and the workload balance of inspectors to ensure scientific and reasonable resource allocation.

[0095] Embodiment 5:

[0096] The difference between this embodiment and any one of Embodiments 1 to 4 is that step S630 includes steps S631 to S634.

[0097] Step S631: construct a path energy consumption model based on the preliminary inspection path and the building topology in the first information, combined with the energy consumption characteristics of different connection modes such as stair walking and elevator operation, and cumulatively calculate the energy consumption of each path segment based on the path energy consumption model to generate a path energy consumption matrix, where each element in the path energy consumption matrix represents the unit energy consumption between any two abnormal device nodes;

[0098] This step first identifies the connection methods between floors based on the building topology in the first information, including walking up the stairs, taking the elevator, etc. The connection methods have significant differences in energy consumption and time consumption. The path energy consumption model quantifies the differences and assigns corresponding energy consumption weights to each connection method. For example, the energy consumption of walking up the stairs is related to the physical burden of the inspectors, while the energy consumption of elevator operation depends on the elevator load and running distance. The path energy consumption matrix comprehensively considers the path length, connection method and physical location between devices. This matrix provides an accurate energy consumption evaluation standard for further optimization of the inspection path.

[0099] Step S632: construct a reinforcement learning environment according to the path energy consumption matrix, define a reward function by energy consumption and priority, transform the path optimization problem into a path selection problem based on state-action rewards, and obtain an environment model of the reinforcement learning optimization strategy;

[0100] The reinforcement learning environment is based on the building topology and path energy consumption matrix, and maps the spatial position of the inspection target equipment and the energy consumption relationship of the path segment into the state space, where each state represents the position of the inspector at a certain node (equipment). The action space is defined as the possible choices from the current node to the adjacent node, including the direction of the path and the connection method (such as elevator or stairs). The reward function is designed based on the energy consumption of the path selection and the device priority. The higher the device priority and the lower the path energy consumption, the greater the reward value, and vice versa.

[0101] The reinforcement learning environment model learns how to select the optimal path in complex building scenarios by dynamically updating the relationship between state and action. For example, when the inspection path passes through high-priority equipment, the reward function tends to incentivize paths that reach such nodes faster; at the same time, in order to reduce inspection costs, the reward function also penalizes the selection of high-energy consumption paths. With the support of the path energy consumption matrix, the model can dynamically evaluate the energy consumption weight of each path segment, providing a real-time optimization basis for inspection path selection.

[0102] Step S633, performing path strategy training according to the environment model, initializing the Q table using the deep Q learning algorithm to randomly select a path as the initial strategy and updating the path selection strategy by repeatedly simulating the inspection path selection, to obtain a preliminary energy-optimal path;

[0103] Specifically, based on the path optimization problem, the Q-table is initialized using the previously constructed environment model. The Q-table records the reward value of each state-action pair and is used to evaluate the pros and cons of each path selection. In the initial stage, the path selection strategy adopts a random exploration method to encourage the model to try different paths to fully understand all possibilities in the environment. During the training process, the deep Q-learning algorithm continuously simulates the process of inspection path selection and updates the state-action value in the Q-table using environmental feedback (i.e., energy consumption and priority reward function values). Specifically, the algorithm selects an action (path) according to the current state in each iteration, then obtains the reward value from the environment, calculates the maximum Q value for the next step, and updates the Q value of the current state-action pair. This update follows the Bellman equation. By introducing technologies such as experience replay and target network, deep Q-learning can effectively handle the high-dimensional state space and dynamic change characteristics of path selection in complex building environments. Through repeated training, the model gradually converges to the optimal path selection strategy, that is, the path selection scheme that can maximize the reward value under given inspection goals and energy consumption constraints. The initial energy-optimal path covers all high-priority devices while minimizing the total energy consumption required for inspection.

[0104] Step S634: Perform global optimization based on the preliminary energy-optimal path, balance energy consumption and path complexity through a multi-objective optimization algorithm, and generate a final energy-optimal path.

[0105] The multi-objective optimization algorithm first constructs the objective function based on the energy consumption and complexity indicators of each path segment of the preliminary path, and introduces a weight coefficient to reflect the priority balance between the two. Through the iterative optimization process, the algorithm searches for a path solution that optimizes the comprehensive objective function in the candidate path set. For example, a path that prioritizes energy consumption will increase complexity in cross-layer switching, while a path that prioritizes complexity may increase energy consumption. The algorithm gradually approaches the optimal solution that satisfies the trade-off between the two by dynamically adjusting among multiple path solutions. The energy-optimal path finally generated not only covers all high-priority target devices, but also significantly reduces redundant movement and cross-layer frequency during inspections, while maintaining a low energy consumption level. This path design can better adapt to the actual inspection needs of multi-story complex buildings and improve the actual execution efficiency of inspections.

[0106] Embodiment 6:

[0107] like Figure 2 As shown, this embodiment provides a building intelligent inspection system based on an AI algorithm, which can implement the building intelligent inspection method described in any one of Embodiments 1-5. Specifically, the building intelligent inspection system includes:

[0108] The acquisition module 901 is used to acquire first information, second information, third information and fourth information, wherein the first information is a building topology map, which includes the spatial structure, hierarchical interconnection relationship and equipment deployment location of the building, the second information includes the sensor deployment location and sensor data, the third information includes the camera deployment location and image information, and the fourth information includes the historical log data of the device;

[0109] Extraction module 902, used to extract sensor data features according to the first information and the second information, perform multi-scale analysis on the sensor data using wavelet transform, and perform feature aggregation on sensor data of different regions in combination with the spatial position relationship of the building to generate a sensor data feature matrix;

[0110] The identification module 903 is used to identify the device state according to the first information and the third information, identify the operation state and appearance change of the device by using a preset deep learning model, and generate a device state feature vector;

[0111] Detection module 904, used to perform anomaly detection based on the sensor data feature matrix and the device state feature vector, use graph neural network modeling to model the spatial association between devices, identify the anomaly propagation path and impact range, and obtain anomaly detection results;

[0112] A diagnosis module 905 is used to construct a fault diagnosis model according to the fourth information, and use the abnormality detection result as an input value of the fault diagnosis model, and generate a fault diagnosis report by analyzing the time series and spatial association between abnormal points;

[0113] The planning module 906 is used to perform path planning processing according to the first information and the fault diagnosis report, and generate an inspection plan in combination with the priority of abnormal equipment, the interconnection relationship between floors and the energy consumption constraint. The inspection plan includes an optimal inspection path map and corresponding maintenance suggestions.

[0114] Embodiment 6:

[0115] Corresponding to the above method embodiment, this embodiment also provides an intelligent building inspection device based on AI algorithm. The intelligent building inspection device based on AI algorithm described below and the intelligent building inspection method based on AI algorithm described above can refer to each other.

[0116] Figure 3 is a block diagram of an AI algorithm-based building intelligent inspection device 800 according to an exemplary embodiment. Figure 3As shown, the AI ​​algorithm-based building intelligent inspection device 800 may include: a processor 801, a memory 802. The AI ​​algorithm-based building intelligent inspection device 800 may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0117] The processor 801 is used to control the overall operation of the AI ​​algorithm-based building intelligent inspection device 800 to complete all or part of the steps in the above-mentioned AI algorithm-based building intelligent inspection method. The memory 802 is used to store various types of data to support the operation of the AI ​​algorithm-based building intelligent inspection device 800. These data may include, for example, instructions for any application or method operating on the AI ​​algorithm-based building intelligent inspection device 800, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the AI ​​algorithm-based building intelligent inspection device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0118] In an exemplary embodiment, an AI algorithm-based intelligent building inspection device 800 can be implemented by one or more application-specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned AI algorithm-based intelligent building inspection method.

[0119] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned intelligent building inspection method based on an AI algorithm are implemented. For example, the computer-readable storage medium may be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions may be executed by a processor 801 of an intelligent building inspection device 800 based on an AI algorithm to complete the above-mentioned intelligent building inspection method based on an AI algorithm.

[0120] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A building intelligent inspection method based on AI algorithm, characterized in that: include: Acquire first information, second information, third information and fourth information, wherein the first information is a building topology map, the building topology map includes the spatial structure of the building, the hierarchical interconnection relationship and the equipment deployment location, the second information includes the sensor deployment location and sensor data, the third information includes the camera deployment location and image information, and the fourth information includes the historical log data of the device; Extract sensor data features based on the first information and the second information, perform multi-scale analysis on the sensor data using wavelet transform, and perform feature aggregation on sensor data of different regions in combination with the spatial position relationship of the building to generate a sensor data feature matrix; Identify the device state based on the first information and the third information, use a preset deep learning model to identify the operating state and appearance changes of the device, and generate a device state feature vector; Anomaly detection is performed based on the sensor data feature matrix and the device state feature vector, and the spatial association between devices is modeled using a graph neural network to identify the anomaly propagation path and impact range, thereby obtaining anomaly detection results; constructing a fault diagnosis model according to the fourth information, and using the abnormality detection result as an input value of the fault diagnosis model, and generating a fault diagnosis report by analyzing the time series and spatial association between abnormal points; Path planning is performed based on the first information and the fault diagnosis report, and an inspection plan is generated by combining the priority of abnormal equipment, the connectivity between floors and energy consumption constraints.

2. The intelligent building inspection method according to claim 1, characterized in that: The sensor data feature extraction is performed according to the first information and the second information, the sensor data is multi-scale analyzed by wavelet transform, and the sensor data of different regions are feature aggregated in combination with the spatial position relationship of the building to generate a sensor data feature matrix, including: Performing time series smoothing processing according to the second information, and obtaining preliminary multi-scale features by multi-scale decomposition and denoising of the data; Performing hierarchical feature decomposition according to the preliminary multi-scale features and the first information, enhancing the features of sensor data in different regions by spatial position association, and obtaining a spatially weighted hierarchical feature matrix, wherein each row in the hierarchical feature matrix represents a comprehensive feature of a level; Performing feature aggregation according to the hierarchical feature matrix, capturing feature distribution patterns and identifying feature distributions of similar regions through a dynamic clustering algorithm, and obtaining regional feature aggregation results; The cross-layer features are globally modeled according to the regional feature aggregation results, and the cross-layer features are dimensionally reduced and key features are extracted through principal component analysis to generate a final sensor data feature matrix.

3. The intelligent building inspection method according to claim 2 is characterized in that: Performing hierarchical feature decomposition according to the preliminary multi-scale features and the first information, enhancing the features of sensor data in different regions by spatial position association, and obtaining a spatially weighted hierarchical feature matrix, wherein each row in the hierarchical feature matrix represents a comprehensive feature of a level, including: The hierarchical feature matrix is ​​expressed as: Among them, S l,m represents the comprehensive feature value of floor l in the mth feature dimension; T :,m represents the multi-scale eigenvalues ​​of all sensors in the mth feature dimension; W i,: represents the spatial association weight between sensor i and other sensors; P l,: represents the cross-layer propagation weight between floor l and other floors; represents the sensor set on floor l; i represents the sensor number; l represents the floor number; m represents the feature dimension number; represents the eigenvalue of floor k in the mth feature dimension in the previous iteration; P l,k represents the information propagation weight between floor l and floor k; L represents the total number of floors; t-1 represents the previous iteration; α represents the cross-layer propagation adjustment coefficient; σ is the nonlinear activation function.

4. The intelligent building inspection method according to claim 3 is characterized in that: Perform feature aggregation according to the hierarchical feature matrix, capture feature distribution patterns through a dynamic clustering algorithm and identify feature distributions of similar regions, and obtain regional feature aggregation results, including: The dynamic clustering algorithm formula is as follows: Among them, R c represents the comprehensive feature vector of the cth cluster; C c represents the floor index set belonging to the cth cluster; c represents the cluster number; S l,: represents the feature vector of floor l; S k,: represents the feature vector of floor k; λ is the feature balance factor; K represents the total number of clusters.

5. The intelligent building inspection method according to claim 1, characterized in that: The device state recognition is performed according to the first information and the third information, and a preset deep learning model is used to recognize the operation state and appearance change of the device, and a device state feature vector is generated, including: Performing image segmentation processing on the video image according to the third information, obtaining a device image sequence by segmenting the device area in the video frame and removing background and interference information; Performing spatial feature calibration according to the device image sequence and the first information, performing spatial feature calibration on the device in the image, and establishing a corresponding relationship between the image feature and the building space position of the device, thereby obtaining the spatial positioning feature of the device; The running state of the device is identified according to the spatial positioning characteristics of the device, the state changes in the continuous frames are modeled by the time series analysis method, the switching frequency and stability information of the device are extracted, and the state change information of the device is obtained; The appearance change of the device is identified according to the state change information and the device image sequence, and a device state feature vector is generated by detecting the appearance change of the device and combining it with the running state feature for feature fusion.

6. The intelligent building inspection method according to claim 1, characterized in that: Anomaly detection is performed based on the sensor data feature matrix and the device state feature vector, and the spatial association between devices is modeled using a graph neural network to identify the abnormal propagation path and impact range, and obtain anomaly detection results, including: Performing feature fusion processing according to the sensor data feature matrix and the device state feature vector, mapping the features of different modes to a unified space through feature splicing and normalization processing, and dynamically strengthening the mutation trend and operation state change features to obtain a multi-dimensional feature description; Perform outlier detection according to the multidimensional feature description, evaluate the degree of isolation of each device feature point by establishing a random segmentation tree of feature distribution, set a dynamic threshold to filter outliers based on the degree of outliers of the device in the feature space, and obtain a preliminary outlier set, which includes a device ID and a preliminary outlier score; Anomaly propagation modeling is performed based on the preliminary anomaly point set, by inputting the anomaly nodes into the graph neural network, propagating the anomaly signals layer by layer in combination with the topological structure of the building, and obtaining an anomaly path graph of cross-layer propagation through the feature update and aggregation mechanism of the network layer; Anomalies are classified according to the abnormal path graph, and the device nodes in the path are grouped and anomaly priorities are marked using a spectral clustering algorithm to obtain an anomaly detection result.

7. The intelligent building inspection method according to claim 1, characterized in that: Constructing a fault diagnosis model according to the fourth information, and using the abnormality detection result as an input value of the fault diagnosis model, and generating a fault diagnosis report by analyzing the time series and spatial association between abnormal points, including: Performing historical data modeling based on the fourth information, extracting the evolution pattern of the equipment operation state through time series analysis, and obtaining the historical characteristics of the equipment operation state; Perform time series anomaly association analysis based on the anomaly detection results and the historical features, use a dynamic Bayesian network to take the historical state of the device as the state node, take the anomaly propagation path as the edge weight, and evaluate the impact of each anomaly point on the state change of other devices to obtain an anomaly causal chain; Predicting the fault type according to the abnormal causal chain, constructing a fault feature vector using causal nodes and propagation relationships, and identifying the fault mode through a deep learning classification model to obtain a prediction result, which includes the fault type and its probability distribution; The fault impact assessment is performed based on the prediction results, the impact range of the fault on the building operation is quantified through correlation analysis, and a fault diagnosis report is generated.

8. The intelligent building inspection method according to claim 1, characterized in that: Performing path planning processing according to the first information and the fault diagnosis report, and generating an inspection plan in combination with the priority of abnormal equipment, the interconnection relationship between floors and energy consumption constraints, including: Prioritize inspection targets according to the first information and the fault diagnosis report, calculate the relative proximity of each abnormal device to the positive ideal solution and the negative ideal solution using an approximate ideal solution algorithm, and generate a priority ranking table; Initialize the path according to the priority sorting table, map the location of the inspection target device to the topology map node, and use the shortest path algorithm to calculate the preliminary inspection path covering all the target devices; Optimizing the path according to the preliminary inspection path, dynamically adjusting the path selection strategy by constructing a reinforcement learning model with the goal of minimizing the total energy consumption of the path, and generating an energy-optimal path; According to the energy-optimal path, the final inspection plan is output through path labeling and task allocation.

9. The intelligent building inspection method according to claim 8, characterized in that: Path optimization is performed according to the preliminary inspection path, and the path selection strategy is dynamically adjusted by constructing a reinforcement learning model with the goal of minimizing the total energy consumption of the path to generate an energy-optimal path, including: According to the preliminary inspection path and the building topology in the first information, a path energy consumption model is constructed in combination with energy consumption characteristics of different connection modes such as stair walking and elevator operation, and the energy consumption of each path segment is cumulatively calculated based on the path energy consumption model to generate a path energy consumption matrix, wherein each element in the path energy consumption matrix represents a unit energy consumption between any two abnormal device nodes; A reinforcement learning environment is constructed according to the path energy consumption matrix, a reward function is defined by energy consumption and priority, the path optimization problem is transformed into a path selection problem based on state-action rewards, and an environment model of a reinforcement learning optimization strategy is obtained; Performing path strategy training according to the environmental model, initializing the Q table using a deep Q learning algorithm to randomly select a path as the initial strategy, and updating the path selection strategy by repeatedly simulating inspection path selection to obtain a preliminary energy-optimal path; A global optimization is performed based on the preliminary energy-optimal path, and a multi-objective optimization algorithm is used to balance energy consumption and path complexity to generate a final energy-optimal path.

10. An intelligent building inspection system based on AI algorithm, characterized in that: include: an acquisition module, configured to acquire first information, second information, third information, and fourth information, wherein the first information is a building topology map, the building topology map including the spatial structure, hierarchical interconnection relationship, and device deployment location of the building, the second information including the sensor deployment location and sensor data, the third information including the camera deployment location and image information, and the fourth information including historical log data of the device; an extraction module, configured to extract sensor data features according to the first information and the second information, perform multi-scale analysis on the sensor data using wavelet transform, and perform feature aggregation on sensor data of different regions in combination with the spatial position relationship of the building to generate a sensor data feature matrix; an identification module, configured to identify the device state according to the first information and the third information, identify the operating state and appearance change of the device using a preset deep learning model, and generate a device state feature vector; A detection module, used to perform anomaly detection based on the sensor data feature matrix and the device state feature vector, use a graph neural network to model the spatial association between devices, identify the anomaly propagation path and impact range, and obtain anomaly detection results; A diagnosis module, configured to construct a fault diagnosis model according to the fourth information, and use the abnormality detection result as an input value of the fault diagnosis model, and generate a fault diagnosis report by analyzing the time series and spatial association between abnormal points; A planning module is used to perform path planning processing according to the first information and the fault diagnosis report, and generate an inspection plan in combination with the priority of abnormal equipment, the interconnection relationship between floors and energy consumption constraints. The inspection plan includes an optimal inspection path map and corresponding maintenance suggestions.

Images