Urban Internet of Things management platform and method

Abstract

The invention belongs to the technical field of Internet of Things, and particularly relates to an urban Internet of Things management platform and method, and the platform comprises a sensor layer which comprises a temperature sensor, a gas sensor and a monitoring camera, and is used for collecting multi-modal data; the communication gateway layer is used for integrating different network transmission modes according to field scenes to realize data transmission among the devices; and the data decision-making layer comprises an Arduino UNO development board, and the Arduino UNO development board is used for constructing an expert system, performing feature extraction on collected multi-modal data to obtain a multi-modal feature vector, matching the multi-modal feature vector with a preset expert rule base, and judging whether an accident potential exists or not. The expert rule base is based on city safety management regulations and expert experience, the inference rule in the application field is defined by using the multi-modal feature vector set, and an application mechanism for judging whether accident potential exists or not according to multi-modal data collected in real time is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of Internet of Things, and in particular, relates to a city Internet of Things management platform and method. Background Art

[0002] In the wave of smart city 3.0 construction, in order to cope with the complex environment of the city, emergencies such as fire, abnormal gathering of people, pollution leakage, etc. pose a serious threat to urban public safety. The Internet of Things data platform has been widely used in urban safety management. However, although the existing urban management system has partially realized digital transformation, there are still significant technical bottlenecks in multimodal data fusion, accident hazard judgment and decision-making, etc., which makes it difficult to meet the stringent requirements of modern super-large cities for real-time accident hazard prevention and control.

[0003] The current mainstream urban management systems generally adopt a single-modal perception architecture, with environmental monitoring (temperature, humidity, air quality, etc.) and video surveillance systems belonging to independent platforms, resulting in inconsistent time series benchmarks for multi-dimensional data; and the lack of a systematic way to define reasoning rules leads to significant limitations in the system's ability to identify complex accident hazards.

[0004] In addition, the rule base of traditional expert systems is mostly constructed manually, and its reasoning rules are incomplete. There is a lack of methods to complete and discover reasoning rules, making it difficult to adapt to different urban safety scenarios in a timely manner. This leads to insufficient ability to explore potential accident hazards, limiting the universality and intelligence level of the system. Summary of the invention

[0005] In order to solve the technical problems that mainstream urban management systems lack a systematic way to define inference rules and lack methods to complete and discover inference rules, the present invention provides a city Internet of Things management platform and method.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An urban Internet of Things management platform, comprising a sensor layer, a communication gateway layer and a data decision layer;

[0008] The sensor layer includes temperature sensors, gas sensors, and surveillance cameras to collect multi-modal data;

[0009] The communication gateway layer includes UART bus network and wireless communication network, which are used to integrate different network transmission methods according to the actual scene to realize data transmission between devices;

[0010] The data decision layer includes the Arduino UNO development board, which is used to build an expert system. The multimodal feature vector obtained by feature extraction of the collected multimodal data is matched with the preset expert rule library to determine whether there are potential accident hazards.

[0011] Preferably, the specific process of constructing the expert system includes:

[0012] Data collection: Collect multi-modal data through multiple sensors, including environmental numerical data and image data; the environmental numerical data includes temperature and smoke concentration;

[0013] Data preprocessing: Preprocess the collected data, including data filtering and time alignment, to ensure the temporal consistency of multi-modal data;

[0014] Feature extraction and fusion: Use OpenCV image processing technology to perform target entity recognition and annotation on the image data, and capture entity features; and select key numerical features from the environmental numerical data, and fuse the entity features and key numerical features to form a multi-modal feature vector set;

[0015] Expert rule base construction: Establish an expert rule base according to relevant regulations of urban safety management or expert experience, and use the multi-modal feature vector set to define the inference rules within the application domain;

[0016] Expert rule base matching: Match the real-time monitored multi-modal feature vector set with the inference rules in the expert rule base to determine whether there are potential accident hazards.

[0017] Preferably, the inference rules in the expert rule base include:

[0018] When the crowd evacuation is abnormal and accompanied by a sharp rise in temperature or an excessive smoke concentration, it is determined as a potential fire accident hazard.

[0019] Preferably, the Arduino UNO development board is also used to predict potential entity relationships through an association rule mining method using historical data, so as to supplement the inference rules in the expert rule base; the specific process includes:

[0020] Data preparation: Collect and organize urban safety regulation documents, historical accident report documents or historical accident monitoring images as the data source for constructing the knowledge graph;

[0021] Information extraction: Use natural language processing technology and OpenCV image processing technology to identify and extract each entity and its relationship from the data source;

[0022] Modeling: Based on each entity and its relationship, construct the graph structure of the knowledge graph, where the entity is used as a node and the relationship is used as an edge;

[0023] Relationship prediction: Adopt an association rule mining method to automatically discover inference rules from the knowledge graph, so as to infer possible but not explicitly recorded inference rules;

[0024] Verification and Iteration: Continuously iterate and optimize the inference rules based on the verification results in the actual application scenarios.

[0025] Preferably, the OpenCV image processing technology is an open-source image annotation tool based on OpenCV, such as LabelImg or VGG ImageAnnotator. By setting the annotation mode, annotation specifications, and annotation formats, each entity and its relationship in the collected image data are automatically annotated.

[0026] Preferably, the specific process of the association rule mining method includes:

[0027] S1) Perform data preprocessing on the constructed knowledge graph, including filling in missing values;

[0028] S2) Pre-define a series of rule templates; based on the rule templates, traverse all possible entity combinations in the knowledge graph to generate a large number of candidate rules;

[0029] S3) For each candidate rule, calculate its support and confidence;

[0030] S4) Set a minimum threshold for both support and confidence respectively, filter out candidate rules with low confidence or insufficient support, and retain high-quality candidate rules as the basis for complementing the expert rule base;

[0031] S5) Perform optimization processing on the retained high-quality candidate rules, including rule merging and experimental verification;

[0032] S6) Complement the optimized candidate rules into the expert rule base.

[0033] Preferably, the specific process of the rule template includes:

[0034] S21) Start from the simplest rule using the AMIE association rule mining method, that is, the initial rule;

[0035] S22) Adopt an extension operation. Based on the initial rule, expand the premise part of the rule by adding additional entities. The process of the extension operation includes:

[0036] S221) Add a new entity z to the initial rule r(x,y) and introduce a new relationship s, thus expanding it to r(x,y) ∧ s(y,z);

[0037] S222) Introduce a new relationship in the current rule r(x,y) ∧ s(y,z) to connect the existing entities, thereby forming a closed structure, that is, forming the rule template: r(x,y) ∧ s(y,z) → g(x,z);

[0038] In the expression of this rule template, r(x, y) ∧ s(y, z) is the premise part of the rule; g(x, z) is the conclusion part of the rule; x, y, and z are variable entities.

[0039] Preferably, before the process S221) of the expansion operation, it further includes:

[0040] Introduce the index values of existing entities into the initial rule r(x, y). The index values include temperature T(x) or smoke concentration C(y), so as to expand it to r(x, y) ∧ T(x) or r(x, y) ∧ C(y).

[0041] Preferably, the specific process of calculating its support and confidence includes:

[0042] S31) Search all entity combinations that satisfy the rule in the knowledge graph;

[0043] S32) For each entity combination that satisfies the premise part, verify whether its conclusion part holds;

[0044] S33) Count the number of instances that simultaneously satisfy the premise part and the conclusion part, denoted as σ(B ∧ H); and then calculate the support, and its calculation formula is:

[0045]

[0046] In the formula, σ(B ∧ H) represents the number of instances that simultaneously satisfy the premise part and the conclusion part; N represents the normalization factor;

[0047] S34) Count the number of instances that satisfy the premise part, denoted as σ(B); and then calculate the confidence, and its calculation formula is:

[0048]

[0049] In the formula, σ(B) represents the number of instances that satisfy the premise part.

[0050] The present invention also provides an urban Internet of Things management method, including the specific process of constructing an expert system in the above-mentioned urban Internet of Things management platform and the specific process of complementing the inference rules in the expert rule base; realizing the prediction of potential entity relationships through the association rule mining method using historical data, so as to complement the inference rules in the expert rule base.

[0051] The beneficial effects of the present invention:

[0052] 1. By deploying a variety of environmental value sensors and monitoring cameras in the sensor layer, the comprehensive collection of multi-modal data in the urban environment is realized.

[0053] 2. The expert rule base is based on urban safety management regulations and expert experience, and uses a multi-modal feature vector set to define inference rules within the application domain, implementing an application mechanism for determining whether there are potential accident hazards based on real-time collected multi-modal data.

[0054] 3. Construct a knowledge graph using historical accident texts and accident image data, and combine the association rule mining method of AMIE. By traversing the entities and their relationships recorded in the knowledge graph, automatically discover and complete the inference rules in the expert rule base. After expert verification and iterative update, incorporate them into the expert system built by Arduino UNO to improve the accuracy and robustness of the expert system in judging potential safety accident hazards. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0056] Figure 1 It is a structural framework diagram of a city Internet of Things management platform of the present invention.

[0057] Figure 2 It is a flowchart of the steps for constructing an expert system in a city Internet of Things management platform of the present invention.

[0058] Figure 3 It is a flowchart of the steps of an association rule mining method in a city Internet of Things management platform of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0060] Please refer to Figures 1 - 3 As shown, a city Internet of Things management platform includes a sensor layer, a communication gateway layer, and a data decision layer;

[0061] The sensor layer includes temperature sensors, gas sensors, and monitoring cameras for collecting multi-modal data;

[0062] The communication gateway layer includes a UART bus network and a wireless communication network for integrating different network transmission methods according to the on-site scenario to achieve data transmission between devices;

[0063] The data decision-making layer includes an Arduino UNO development board, which is used to build an expert system. By extracting features from the collected multi-modal data to obtain multi-modal feature vectors, and matching them with a preset expert rule base to determine whether there are potential accident hazards.

[0064] Specifically, the sensor layer is mainly responsible for collecting multi-modal data in the urban environment, including environmental numerical data such as temperature and gas, as well as image data captured by surveillance cameras. This layer ensures rich data sources, covering all data dimensions where potential accident hazards may occur in the urban environment. The communication gateway layer integrates wired (UART bus) and wireless communication networks, and according to the actual situation on site, uses different transmission methods to transmit the data collected by each sensor to the decision-making layer in real time. This layer realizes cross-device and cross-network data interconnection and interoperability, ensuring the stability and low latency of data transmission. The data decision-making layer takes the Arduino UNO development board as the core and deploys an expert system. This layer preprocesses, extracts features from, and fuses the multi-modal data transmitted by the sensors, and then combines with the preset expert rule base for rule matching to determine whether there are potential accident hazards. In addition, the expert rule base is automatically completed and optimized using historical data through association rule mining (such as the AMIE method) to improve the system's adaptability and intelligence level.

[0065] Furthermore, the construction process of the expert system includes:

[0066] Data collection: Collect multi-modal data through multiple sensors, including environmental numerical data and image data; the environmental numerical data includes temperature and smoke concentration;

[0067] Data preprocessing: Preprocess the collected data, including data filtering and time alignment, to ensure the temporal consistency of multi-modal data;

[0068] Feature extraction and fusion: Use OpenCV image processing technology to perform target entity recognition and annotation on image data, capture entity features; and select key numerical features from environmental numerical data, and fuse the entity features and key numerical features to form a multi-modal feature vector set;

[0069] Expert rule base construction: Establish an expert rule base according to relevant regulations of urban safety management or expert experience, and use the multi-modal feature vector set to define inference rules within the application domain;

[0070] Expert rule base matching: Match the real-time monitored multi-modal feature vector set with the inference rules in the expert rule base to determine whether there are potential accident hazards.

[0071] In the specific implementation process, temperature sensors and gas sensors are deployed to collect environmental numerical data, including temperature, smoke concentration, etc. Monitoring cameras are deployed to capture on-site images in real time. In data preprocessing, for environmental numerical data such as temperature and smoke concentration, filtering, denoising, and outlier detection are performed, and the data of each sensor is aligned using timestamps to ensure consistent time series; for image data, image normalization, size adjustment, denoising, and enhancement are performed using OpenCV preprocessing techniques; overall, a clean and synchronized multi-modal dataset is formed.

[0072] In the feature extraction and fusion process, LabelImg or VGG Image Annotator based on OpenCV is used to annotate image data and identify target entities, and entity features are extracted, including flames, smoke, crowd dynamics, or entity relationships, etc. Key numerical features such as temperature and smoke concentration are selected from the environmental data; furthermore, the features extracted from the images are fused with the numerical features to form a unified feature vector set, providing a comprehensive judgment basis for subsequent rule matching.

[0073] In the construction and matching process of the expert rule library, based on urban safety management regulations, national standards, and expert on-site experience, preliminary inference rules for potential safety hazards are sorted out, such as "when the crowd evacuation is abnormal and accompanied by a sharp rise in temperature or an excessive smoke concentration, it is judged as a potential fire accident hazard". Then, the inference rules are encoded for expression so that they can be matched with the feature vectors. Specifically, a multi-modal feature vector set is used to define the inference rules within the application domain. After the multi-modal feature vector encoding is completed, all the rules are stored in the expert rule library to form a structured inference rule. Thus, the judgment of potential safety hazards is realized: the fused multi-modal feature vectors received in real time are matched with the inference rules in the expert rule library, and the system can determine whether there are corresponding accident hazards and trigger an alarm or notification mechanism.

[0074] Furthermore, the inference rules in the expert rule library include:

[0075] When the crowd evacuation is abnormal and accompanied by a sharp rise in temperature or an excessive smoke concentration, it is judged as a potential fire accident hazard.

[0076] Furthermore, the Arduino UNO development board is also used to predict potential entity relationships through association rule mining methods using historical data, thereby complementing the inference rules in the expert rule library; the specific process includes:

[0077] Data preparation: Collect and organize urban safety regulation documents, historical accident report documents, or historical accident monitoring images as data sources for constructing a knowledge graph;

[0078] Information extraction: Using natural language processing technology and OpenCV image processing technology, identify and extract various entities (such as crowds, flames, smoke) and their relationships (such as "far from", "contain") from the data source;

[0079] Modeling: Based on each entity and its relationship, construct the graph structure of the knowledge graph, where the entity is used as a node and the relationship is used as an edge;

[0080] Relationship prediction: Adopt the method of association rule mining (such as AMIE), automatically discover inference rules from the knowledge graph, so as to infer the inference rules that may exist but are not explicitly recorded;

[0081] Verification and iteration: Continuously iterate and optimize the inference rules according to the verification results in the actual application scenario.

[0082] In the specific implementation process, data collection includes text data and image data; among them, entity information and the relationships between entities are extracted from the text data by using natural language processing (NLP) technology (such as word segmentation, named entity recognition, relationship extraction). Use OpenCV and related image annotation tools (such as LabelImg or VGG Image Annotator) to perform object detection and annotation on the monitoring images, and extract the entity information and the relationships between entities in the images. Furthermore, organize the extracted entities and relationships into triples, and construct a graph structure through the triples, with nodes representing entities and edges representing relationships, to form a preliminary knowledge graph.

[0083] Then, based on the knowledge graph, adopt the method of association rule mining (such as AMIE), traverse all possible entity combinations in the knowledge graph, and automatically generate a large number of candidate rules. Finally, after being reviewed by domain experts, verified and iterated; realize the complement and optimization of the inference rules in the expert rule base.

[0084] Furthermore, the OpenCV image processing technology is an open-source image annotation tool based on OpenCV, such as LabelImg or VGG Image Annotator, which automatically annotates each entity and its relationship in the collected image data by setting the annotation mode, annotation specification and annotation format.

[0085] In the specific implementation process, the annotation mode improves the model's recognition ability for complex scenarios by defining the scene information, target entities, relationships between entities, etc. of the accident monitoring images.

[0086] The annotation specification uses the COCO format, which supports more complex annotation structures (such as segmentation, multi-class, multi-object entities). The COCO annotation format is a widely used image annotation data format, originally proposed by the COCO dataset, and is used to describe the data structures in tasks such as object detection, segmentation, and keypoint detection. It uses JSON files as the storage method, has powerful expressive ability, and supports the annotation requirements of multi-object, multi-class, and complex structures. The COCO format is widely supported, and many deep learning frameworks (such as PyTorch, SSD) provide tools to directly load COCO data. The standardized data format of the COCO format facilitates developers to manage and exchange annotation data in a unified format. It supports various computer vision tasks such as object entity detection, instance segmentation, and keypoint detection. And through detailed and diverse annotations, it improves the adaptability of the model to complex scenarios. It can also quickly adapt to new tasks or scenario requirements by adding new categories or fields.

[0087] Furthermore, the specific process of the association rule mining (such as AMIE) method includes:

[0088] S1) Perform data preprocessing on the constructed knowledge graph, including filling missing values;

[0089] S2) Pre-define a series of rule templates; based on the rule templates, traverse all possible entity combinations in the knowledge graph to generate a large number of candidate rules;

[0090] S3) For each candidate rule, calculate its support (i.e., the frequency of the rule appearing in the knowledge graph) and confidence (i.e., the probability of the rule conclusion appearing when the rule premise is satisfied);

[0091] S4) Set a minimum threshold for support and confidence respectively, filter out candidate rules with low confidence or insufficient support, and retain high-quality candidate rules as the basis for complementing the expert rule library;

[0092] S5) Perform optimization processing on the retained high-quality candidate rules, including rule merging and experimental verification;

[0093] S6) Complement the optimized candidate rules into the expert rule library.

[0094] Furthermore, the specific process of the rule template includes:

[0095] S21) Use the association rule mining method of AMIE to start from the simplest rule, that is, the initial rule;

[0096] S22) Use the expansion operation. Based on the initial rule, expand the premise part of the rule by adding additional entities; the process of the expansion operation includes:

[0097] S221) Add a new entity z to the initial rule r(x, y) and introduce a new relationship s, thus expanding it to r(x, y) ∧ s(y, z);

[0098] S222) Introduce a new relationship in the current rule r(x, y) ∧ s(y, z) to connect the existing entities, thus forming a closed structure, that is, forming a rule template: r(x, y) ∧ s(y, z) → g(x, z);

[0099] In the expression of this rule template, r(x, y) ∧ s(y, z) is the premise part of the rule; g(x, z) is the conclusion part of the rule; x, y, z are variable entities.

[0100] Specifically, in the application field of urban safety management of the present invention, the knowledge graph is composed of a large number of triples'relationship (entity x, entity y)', reflecting the explicit relationships between entities in the field of urban safety management. The AMIE method is specifically designed for incomplete knowledge graphs, aiming to automatically mine interpretable Horn rules from a large amount of triple data.

[0101] Due to the often missing real-world data, the knowledge graph is not complete. The association rule mining method based on AMIE is exactly in such an incomplete data environment, and mines the potential semantic relationships and structural relationships in the knowledge graph through statistical patterns. The AMIE method starts from the simplest Horn rule template (for example, r(x, y)) as the initial rule, and then generates more complex Horn rule templates through expansion operations, such as r(x, y) ∧ s(y, z) → g(x, z). Furthermore, based on the combinations of various entities and relationships in the knowledge graph, all possible entity combinations are traversed in the form of Horn rule templates to generate a large number of candidate rules. The expansion operation can also introduce numerical indicators (such as temperature, smoke concentration, etc.) to enhance the rule expression ability.

[0102] Furthermore, before step 1) of the expansion operation, it also includes:

[0103] Introduce the index values of existing entities into the initial rule r(x, y), and the index values include temperature T(x) or smoke concentration C(y), thus expanding it to r(x, y) ∧ T(x) or r(x, y) ∧ C(y).

[0104] Furthermore, the specific process of calculating its support and confidence includes:

[0105] Search for all entity combinations that satisfy the rule in the knowledge graph;

[0106] For each entity combination that satisfies the premise part, verify whether its conclusion part holds;

[0107] Count the number of instances that satisfy both the premise part and the conclusion part, denoted as σ(B∧H); then calculate the support degree, and its calculation formula is:

[0108]

[0109] In the formula, σ(B∧H) represents the number of instances that satisfy both the premise part and the conclusion part; N represents the normalization factor;

[0110] Count the number of instances that satisfy the premise part, denoted as σ(B); then calculate the confidence degree, and its calculation formula is:

[0111]

[0112] In the formula, σ(B) represents the number of instances that satisfy the premise part.

[0113] Specifically, for support degree calculation: count the number of triple combinations in the knowledge graph that satisfy the premise part of the candidate rule, which reflects the occurrence frequency of the rule in the data.

[0114] For confidence degree calculation: under the condition of satisfying the premise, count the proportion of instances that also satisfy the rule conclusion. To cope with the incompleteness of the knowledge graph, AMIE usually introduces PCA confidence degree, adjusting the denominator to only consider those known parts, so as to more truly reflect the reliability of the rule.

[0115] By setting the minimum thresholds of support degree and confidence degree, filter out the candidate rules with higher quality and filter out the candidate rules with low frequency or low reliability.

[0116] Finally, merge the rules screened by the thresholds to form a clearer and more interpretable inference rule expression; use the optimized rules as supplementary information and add them to the expert rule base to expand and improve the original inference rules. At the same time, load the supplemented expert rule base into the expert system constructed by Arduino UNO.

[0117] The present invention also provides a method for managing urban Internet of Things, including the specific process of constructing an expert system in the above-mentioned urban Internet of Things management platform and the specific process of complementing the inference rules in the expert rule base. Use the association rule mining method of AMIE to automatically generate candidate rules on the knowledge graph constructed from historical data, and perform screening and optimization through support degree and confidence degree calculations to form high-quality inference rules. After expert verification and iterative update, these rules complement the expert rule base, thereby improving the accuracy and intelligence level of the urban Internet of Things security management platform in real-time accident hidden danger identification and risk warning.

[0118] In summary, the present invention constructs a knowledge graph using historical accident texts and accident image data, and combines the association rule mining method of AMIE to automatically analyze and complete the inference rules in the expert rule base. This mechanism discovers potential implicit associations by traversing the entities and their relationships in the knowledge graph, generates candidate rules, and after expert verification and iterative update, incorporates the optimized inference rules into the expert system built with Arduino UNO. This significantly improves the comprehensiveness, accuracy, and self-adaptability of the urban Internet of Things management platform in accident hazard judgment, providing more intelligent and efficient technical support for urban safety management.

[0119] In the description of specific embodiments, the descriptions referring to terms such as "specifically", "concretely", "in the specific implementation process", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. The above content is only an example and explanation of the structure of the present invention. Those skilled in the art of the present technology can make various modifications or supplements to the described specific embodiments or use similar methods for substitution, as long as they do not deviate from the structure of the invention or exceed the scope defined by the claims of the present invention, they should fall within the protection scope of the present invention.

Claims

1. A city Internet of Things management platform, characterized by: Includes sensor layer, communication gateway layer and data decision layer; The sensor layer includes temperature sensors, gas sensors, and surveillance cameras to collect multi-modal data; The communication gateway layer includes UART bus network and wireless communication network, which are used to integrate different network transmission methods according to the actual scene to realize data transmission between devices; The data decision layer includes an Arduino UNO development board, which is used to build an expert system. The multimodal feature vector obtained by feature extraction of the collected multimodal data is matched with the preset expert rule library to determine whether there are potential accident hazards.

2. According to claim 1, a city Internet of Things management platform is characterized by: The specific process of building the expert system includes: Data collection: Collect multi-modal data through multiple sensors, including environmental numerical data and image data; environmental numerical data includes temperature and smoke concentration; Data preprocessing: preprocess the collected data, including data filtering and time alignment, to ensure the temporal consistency of multimodal data; Feature extraction and fusion: Use OpenCV image processing technology to identify and annotate target entities in image data and capture entity features; select key numerical features from environmental numerical data and fuse entity features and key numerical features to form a multimodal feature vector set; Expert rule base construction: Establish an expert rule base based on relevant regulations of urban safety management or expert experience, and use a multimodal feature vector set to define the inference rules in the application field; Expert rule base matching: Match the multimodal feature vector set monitored in real time with the inference rules in the expert rule base to determine whether there are potential accident hazards.

3. A city Internet of Things management platform according to claim 2, characterized in that: The inference rules in the expert rule base include: When the evacuation of people is abnormal and accompanied by a sharp rise in temperature or excessive smoke concentration, it is judged as a potential fire accident hazard.

4. The urban Internet of Things management platform according to claim 1, characterized in that: The Arduino UNO development board is also used to use historical data to predict potential entity relationships through association rule mining methods, thereby completing the inference rules in the expert rule base; the specific process includes: Data preparation: Collect and organize city safety regulations documents, historical accident report documents, or historical accident monitoring images as data sources for building knowledge graphs; Information extraction: Use natural language processing technology and OpenCV image processing technology to identify and extract entities and their relationships from data sources; Modeling: Based on each entity and its relationship, a graph structure of the knowledge graph is constructed, where entities are nodes and relationships are edges; Relationship prediction: Use association rule mining methods to automatically discover inference rules from the knowledge graph, thereby inferring inference rules that may exist but are not explicitly recorded; Verification and iteration: Based on the verification results in actual application scenarios, the inference rules are continuously iterated and optimized.

5. The urban Internet of Things management platform according to claim 4 is characterized in that: The OpenCV image processing technology is an open source image annotation tool based on OpenCV's Label Img or VGG Image Annotator, which automatically annotates various entities and their relationships in the collected image data by setting the annotation mode, annotation specification and annotation format.

6. The urban Internet of Things management platform according to claim 4 is characterized by: The specific process of the association rule mining method includes: S1) Perform data preprocessing on the constructed knowledge graph, including missing value filling; S2) predefine a series of rule templates; based on the rule templates, traverse all possible entity combinations in the knowledge graph to generate a large number of candidate rules; S3) For each candidate rule, calculate its support and confidence; S4) setting a minimum threshold for support and confidence respectively, filtering out candidate rules with low confidence or insufficient support, and retaining high-quality candidate rules as the basis for completing the expert rule base; S5) optimizing the retained high-quality candidate rules, including rule merging and experimental verification; S6) Completing the optimized candidate rules into the expert rule base.

7. The urban Internet of Things management platform according to claim 6, characterized in that: The specific process of the rule template includes: S21) The association rule mining method using AMIE starts from the simplest rule, i.e., the initial rule; S22) adopting an extension operation, based on the initial rule, by adding additional entities to extend the premise part of the rule; the process of the extension operation includes: S221) Add a new entity z to the initial rule r(x,y) and introduce a new relation s, thereby expanding it to r(x,y)∧s(y,z); S222) Introduce a new relationship into the current rule r(x,y)∧s(y,z) to connect the existing entities, so as to form a closed structure, that is, form a rule template: r(x,y)∧s(y,z)→g(x,z); In the expression of the rule template, r(x,y)∧s(y,z) is the premise part of the rule; g(x,z) is the conclusion part of the rule; x, y, z are variable entities.

8. The urban Internet of Things management platform according to claim 7, characterized in that: Before the process S221 of the expansion operation, it also includes: The index values ​​of existing entities are introduced into the initial rule r(x,y), and the index values ​​include temperature T(x) or smoke concentration C(y), thereby expanding to r(x,y)∧T(x) or r(x,y)∧C(y).

9. The urban Internet of Things management platform according to claim 6, characterized in that: The specific process of calculating its support and confidence includes: S31) searching the knowledge graph for all entity combinations that satisfy the rules; S32) for each entity combination that satisfies the premise part, verify whether its conclusion part is established; S33) Count the number of instances that satisfy both the premise and conclusion, denoted as σ(B∧H); and then calculate the support, the calculation formula is: In the formula, σ(B∧H) represents the number of instances that satisfy both the premise and conclusion; N represents the normalization factor; S34) Count the number of instances that meet the premise part, recorded as σ(B); then calculate the confidence, the calculation formula is: Where σ(B) represents the number of instances that satisfy the premise.

10. A city Internet of Things management method, characterized by: The specific process of a city Internet of Things management platform as described in any one of claims 4 to 9; realizing the use of historical data to predict potential entity relationships through association rule mining methods, thereby completing the inference rules in the expert rule library.

Images