AI intelligent data processing method for fire fighting field

Abstract

The invention belongs to the field of fire-fighting data processing, and particularly relates to an AI intelligent data processing method for the fire-fighting field, and the method comprises the steps: firstly, building a fire-fighting multi-association knowledge graph through collecting fire-fighting multi-field and multi-source data; secondly, analyzing semantic information and a scene background input by a user by adopting a configured multi-input intention reasoning model, and determining a main-collaborative knowledge sub-graph; 3, constructing a multi-graph joint reasoning model, fusing the main-collaborative knowledge sub-graph and semantic information input by the user, generating a user fire-fighting demand plan set, and continuously optimizing the model performance by constructing an optimal reasoning search path and a reasoning approximation loss function; and finally, according to real-time user input and scene feature space, a user-customized fire-fighting demand plan is quickly generated by using the trained multi-graph joint reasoning model, and the information is fed back to the fire-fighting multi-association knowledge graph, so that continuous updating and accumulation of knowledge are realized, and powerful support is provided for intelligent decision making in the fire-fighting field.

Description

Technical Field

[0001] The present invention belongs to the field of fire protection data processing, and in particular relates to an AI intelligent data processing method for the fire protection field. Background Art

[0002] In the field of fire protection, traditional fire management methods mainly rely on manual inspections and basic fire protection facilities such as smoke detectors and sprinkler systems. Although these methods can cope with fire accidents to a certain extent, they have obvious deficiencies in risk prediction, rapid response and optimal resource allocation; for example, traditional methods often rely on simple resampling or a small amount of data expansion technology in fire risk data expansion, which can easily lead to overfitting and insufficient generalization ability; existing technologies may not fully utilize the inherent structure of fire risk data, resulting in time-consuming and inefficient feature extraction and dimensionality reduction processes; when evaluating high-risk fire events, existing technologies may be difficult to achieve high accuracy and sensitivity due to the limitations of model design and loss functions.

[0003] For example, a Chinese patent application with publication number CN118152939A discloses an intelligent fire data processing method and device, which includes: obtaining target fire data to be processed and corresponding data identification parameters; inputting the data identification parameters into a pre-trained neural network model to obtain output predicted data level and predicted data type; determining data processing rules and data forwarding rules corresponding to the target fire data according to the predicted data level and predicted data type, and a preset type-rule correspondence relationship; and performing data processing operations and forwarding operations on the target fire data according to the data processing rules and data forwarding rules.

[0004] For example, a Chinese patent application with publication number CN114764743A discloses a fire data processing method and a smart fire platform; based on the fire management object, the fire data is divided into N preset categories; the categories include: personnel, materials, fire sources, building facilities, fire protection systems, and fire protection facilities; the fire data of enterprises and institutions and their buildings are collected and entered into a network database to initialize the fire data; corresponding data update rules are established based on the categories, and enterprises and institutions dynamically update the fire data according to the data update rules.

[0005] The above-mentioned prior art has the following problems: the prior art is unable to provide professional and accurate fire assessment plans and professional fire fighting strategy overall plans for different scenarios in the fire fighting field, and is unable to quickly find corresponding answers to core issues, resulting in search delays and inability to provide accurate answers. In addition, the fire fighting field is a large field that includes multiple different small fields, and each small field involves huge data. How to coordinate the management of this data is also one of the difficulties of the prior art. For this reason, the present invention provides an AI intelligent data processing method for the fire fighting field. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes an AI intelligent data processing method for the fire protection field. The method first constructs a fire protection multi-association knowledge graph by collecting multi-field and multi-source data in fire protection; secondly, the configured multi-input intention reasoning model is used to parse the semantic information and scene feature space input by the user, and determine the main-collaborative knowledge sub-graph; thirdly, a multi-graph joint reasoning model is constructed to fuse the main-collaborative knowledge sub-graph with the user input semantic information to generate a user fire protection demand plan set, and continuously optimize the model performance by constructing the optimal reasoning search path and reasoning approximation loss function; finally, according to the real-time user input and scene feature space, the trained multi-graph joint reasoning model is used to quickly generate user-customized fire protection demand plans, and this information is fed back to the fire protection multi-association knowledge graph to realize the continuous updating and accumulation of knowledge, and provide strong support for intelligent decision-making in the fire protection field.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] An AI intelligent data processing method for fire protection field, comprising:

[0009] S1. Obtain building fire layout data, fire law data, fire and corresponding processing information data, and fire simulation and management data in different fields, and construct a fire protection multi-association knowledge graph through natural language algorithms and graph databases;

[0010] S2. Obtain user input semantic information and corresponding scene feature space through the configured multi-input intention reasoning model, and determine the main knowledge subgraph and collaborative knowledge subgraph in the fire protection multi-association knowledge graph through the user input semantic information, the corresponding scene feature space and the fire protection multi-association knowledge graph, and perform the main-collaborative knowledge subgraph switching according to the scene feature space through the configured main-cooperative switching node;

[0011] S3. Construct a multi-graph joint reasoning model, input the main knowledge sub-graph and the collaborative knowledge sub-graph information, the user input semantic information and the scene feature space into the multi-graph joint reasoning model, obtain the user fire demand plan set, and use the user fire demand plan set to construct the optimal reasoning search path and reasoning approximation loss function to train the multi-graph joint reasoning model, and obtain the trained multi-graph joint reasoning model;

[0012] S4. Based on the real-time acquired user input semantic information and the corresponding scene feature space, the user fire demand plan is obtained through the trained multi-graph joint reasoning model, and the real-time user input semantic information, scene feature space and the corresponding user fire demand plan are saved in the fire multi-association knowledge graph.

[0013] Specifically, the fire protection multi-related knowledge graph includes a first knowledge sub-graph, a second knowledge sub-graph, a third knowledge sub-graph and a fourth knowledge sub-graph; the specific steps of constructing the fire protection multi-related knowledge graph include:

[0014] S101, setting the fire protection layout data of buildings in different fields as the first fire protection sub-information stored in the first knowledge sub-graph, the fire protection law data as the second fire protection sub-information stored in the second knowledge sub-graph, the fire and corresponding processing information data as the third fire protection sub-information stored in the third knowledge sub-graph, and the fire protection simulation and management data as the fourth fire protection sub-information stored in the fourth knowledge sub-graph;

[0015] S102, based on the first fire sub-information, the second fire sub-information, the third fire sub-information and the fourth fire sub-information, hierarchically classify each fire sub-information by using a set classification criterion combined with a classification algorithm to obtain multi-level first fire sub-information, second fire sub-information, third fire sub-information and fourth fire sub-information;

[0016] S103, according to the classified fire protection sub-information corresponding to the jth level in the obtained i-th fire protection sub-information, obtain the ;in represents the entity triple set corresponding to the j-th level classification fire protection sub-information in the i-th fire protection sub-information;

[0017] S104, repeating the process of S103 to obtain the entity triple set corresponding to each level in the i-th fire protection sub-information, and marking each entity triple set with a hierarchical label;

[0018] S105. Obtain the knowledge sub-graph corresponding to the i-th fire protection sub-information according to the set of entity triples in the i-th fire protection sub-information annotated with hierarchical labels, and repeat the processes S103-S105 to obtain the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph and the fourth knowledge sub-graph.

[0019] Specifically, the specific steps of constructing the fire protection multi-association knowledge graph also include:

[0020] S106, according to the fire protection sub-information stored in the node corresponding to the jth level in the i-th knowledge sub-graph and the fire protection sub-information stored in the node corresponding to the jth level in the k-th knowledge sub-graph, through a similar matching algorithm, obtain the two nodes with the highest node matching degree and the lowest load in the j-th level. and ;in represents the hth node in the jth level in the i-th knowledge subgraph, represents the g-th node in the j-th level in the k-th knowledge subgraph;

[0021] S107, Utilization and and the corresponding node matching degree, construct the graph connection between the i-th knowledge subgraph and the k-th knowledge subgraph at the j-th level;

[0022] S108, repeat the process of S106-S107 to obtain the graph connection of the same level corresponding to the i-th knowledge sub-graph and the k-th knowledge sub-graph;

[0023] S109, repeating the process of S106-S108 to obtain the corresponding graph connections between any two knowledge sub-graphs in the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph, and the fourth knowledge sub-graph;

[0024] S110, construct a fire protection multi-association knowledge graph based on the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph and the fourth knowledge sub-graph and all corresponding graph connections.

[0025] Specifically, the steps of determining the main knowledge sub-graph and the collaborative knowledge sub-graph include:

[0026] S201, obtaining user historical input semantic information, and obtaining the scene type corresponding to each user input semantic information through a scene classification algorithm;

[0027] S202, according to the semantic information input by the user under the corresponding scene type, the number of historical visits and search frequency of each knowledge sub-graph node, and the fire protection multi-association knowledge graph, obtain the association matching degree corresponding to each scene type and the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph, and the fourth knowledge sub-graph through an association matching algorithm;

[0028] S203: The knowledge sub-graph corresponding to the maximum correlation matching degree is used as the main knowledge sub-graph, and the remaining knowledge sub-graphs are used as collaborative knowledge sub-graphs.

[0029] Specifically, the main knowledge subgraph adopts a full-node search method; the collaborative knowledge subgraph adopts a difference search method; the difference search steps include:

[0030] S2031. According to the number of historical visits and search frequency of each knowledge sub-graph node in the collaborative knowledge sub-graph under each scenario type, obtain the proportion of the number of search nodes corresponding to each collaborative knowledge sub-graph and the proportion of the search frequency of a single node in each collaborative knowledge sub-graph;

[0031] S2032. Obtain the search importance corresponding to each collaborative knowledge sub-graph based on the product of the association matching degree between each collaborative knowledge sub-graph and the corresponding scene type and the proportion of the corresponding search nodes, and use the search importance based on the database index construction principle to construct the first difference index information between each scene type and each collaborative knowledge sub-graph.

[0032] Specifically, the difference search step also includes:

[0033] S2033, obtaining the search importance of the corresponding single node in each collaborative knowledge subgraph according to the product of the association matching degree between each collaborative knowledge subgraph and the corresponding scene type, the proportion of the number of corresponding search nodes and the proportion of the search frequency of a single node, and constructing the second difference index information using the search importance of the single node in each collaborative knowledge subgraph and the corresponding scene type;

[0034] S2034. Configure a main-cooperative switching node in the fire protection multi-association knowledge graph, and save the first difference index information and the second difference index information in the main-cooperative switching node. According to the real-time acquired user input semantic information and the scene type in the corresponding scene feature space, the main-cooperative knowledge sub-graph switching and the fire protection sub-information difference search are performed through the main-cooperative switching node.

[0035] Specifically, the steps for building and training the multi-graph joint reasoning model include:

[0036] S301, constructing a reasoning input sequence according to user input semantic information, corresponding scene feature space, and fire protection multi-association knowledge graph information;

[0037] S302, according to the scene feature space in the reasoning input sequence, determine the main knowledge subgraph and the collaborative knowledge subgraph information through the information call layer, and call the main-cooperative switching node through the determined information to switch the fire protection multi-association knowledge graph between the main and collaborative knowledge subgraphs, and obtain the first difference index information and the second difference index information stored in the main-cooperative switching node;

[0038] S303, input the switching information of the main-collaborative knowledge sub-graph switching, the user input semantic information, the first difference index information and the second difference index information into the search path layer to obtain the full search main path and the first difference search path, the second difference search path, and the third difference search path;

[0039] S304. Input the full search main path, the first difference search path, the second difference search path, the third difference search path and the determined main-collaborative knowledge sub-graph information into the matching search layer to perform the initial fire plan search for the corresponding scene, and obtain the corresponding scene fire plan sequence and the corresponding search matching degree and search matching loss.

[0040] Specifically, the steps of building and training the multi-graph joint reasoning model also include:

[0041] S305, setting a search matching degree threshold, when at least one of the search matching degrees of the fire emergency plans is greater than the search matching degree threshold, the fire emergency plan that meets the threshold condition and corresponds to the maximum search matching degree is used as the fire emergency plan corresponding to the current scene;

[0042] S306, setting a fire plan screening threshold. When the search matching degrees of the fire plans are all less than or equal to the search matching degree threshold, the fire plan information with the corresponding search matching degree greater than the fire plan screening threshold in the fire plan sequence of the corresponding scene is used as the fire plan reasoning information sequence through the fire plan screening threshold;

[0043] S307, input the corresponding scene feature space, the fire plan reasoning information sequence and the full search main path, the first difference search path, the second difference search path, and the third difference search path into the fusion path reasoning layer constructed by the convolutional attention algorithm and the path generation algorithm, obtain the corresponding full search reasoning main path, the first difference reasoning path, the second difference reasoning path, and the third difference reasoning path, and build the search matching degree threshold into all the reasoning paths;

[0044] S308, input the full search reasoning main path, the first difference reasoning path, the second difference reasoning path, the third difference reasoning path and the corresponding main-collaborative knowledge sub-graph information into the scene reasoning layer constructed by the multimodal feature fusion scene text recognition algorithm, and obtain each fire plan reasoning information on the full search reasoning main path and each difference reasoning path and the corresponding reasoning search matching degree;

[0045] S309, according to the reasoning search matching degree corresponding to each fire plan reasoning on the full search reasoning main path and each difference reasoning path and the search matching degree threshold built in the corresponding reasoning, obtain the reasoning search loss corresponding to each reasoning process on the full search reasoning main path and each difference reasoning path;

[0046] S310, setting an approximation loss threshold, and constructing an inference approximation loss function using the error between the sum of the inference search loss corresponding to the inference path of each inference process and the approximation loss threshold;

[0047] S311. Set the training loss threshold and training cycle, and use the sum of the inference approximation loss function and the search matching loss to construct the multi-graph joint reasoning model training loss function. When the inference search loss of each inference path in the training cycle is less than the approximation loss threshold and the multi-graph joint reasoning model training loss function is less than the training loss threshold, the trained multi-graph joint reasoning model is obtained.

[0048] Specifically, the full search main path in S303 corresponds to the main knowledge subgraph; the first difference search path corresponds to the knowledge subgraph one contained in the collaborative knowledge subgraph; the second difference search path corresponds to the knowledge subgraph two contained in the collaborative knowledge subgraph; the third difference search path corresponds to the knowledge subgraph three contained in the collaborative knowledge subgraph; the knowledge subgraph one, knowledge subgraph two and knowledge subgraph three contained in the collaborative knowledge subgraph are the remaining three knowledge subgraphs of the first knowledge subgraph, the second knowledge subgraph, the third knowledge subgraph and the fourth knowledge subgraph excluding the remaining three knowledge subgraphs used as the main knowledge subgraph, and the knowledge subgraph one, the knowledge subgraph two and the knowledge subgraph three correspond one-to-one to the remaining three knowledge subgraphs.

[0049] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, an AI intelligent data processing method for the fire protection field is implemented.

[0050] A computer-readable storage medium stores computer instructions, which, when executed, execute an AI intelligent data processing method for the fire protection field.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] In view of the deficiencies of the prior art, the present invention constructs a fire protection multi-association knowledge graph, integrates building fire protection layout, fire protection laws, fire handling information and simulation management data, and adopts natural language processing and graph database technology to achieve in-depth understanding and rapid response to complex fire protection scenarios; with the help of a multi-input intention reasoning model, the user input semantic information and scene features are accurately captured, and the main-cooperative knowledge sub-graph is dynamically switched to ensure the accuracy and timeliness of providing professional fire assessment plans and fire protection strategy plans for different scenarios. At the same time, switching the main-cooperative knowledge sub-graph for different scenarios and combining it with the built-in difference search index can increase the accuracy of the search, reduce the scope and amount of search data, and reduce the search delay; the introduction of the multi-graph joint reasoning model not only optimizes the plan generation path, but also continuously improves the prediction accuracy through machine learning It continuously iterates to solve the problems of search delay and inaccurate answers in the prior art; in addition, the fire protection multi-association knowledge graph and the built-in switching node and difference search method in the present invention are also suitable for large-scale fire protection scenarios covering multiple sub-fields, and can efficiently manage and collaboratively process massive data, provide users with instant, professional and personalized fire protection solutions, and greatly improve the efficiency and reliability of responding to emergencies such as fires. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of an AI intelligent data processing method for fire protection in accordance with Embodiment 1 of the present invention;

[0054] Figure 2 This is a structural diagram of the multi-graph joint reasoning model of Example 1 of the present invention. DETAILED DESCRIPTION

[0055] Example 1

[0056] See also Figure 1 , an embodiment provided by the present invention: an AI intelligent data processing method for fire protection field, comprising the following steps:

[0057] S1. Obtain building fire layout data, fire law data, fire and corresponding processing information data, and fire simulation and management data in different fields, and construct a fire protection multi-association knowledge graph through natural language algorithms and graph databases;

[0058] Furthermore, the fire protection multi-related knowledge graph in this embodiment includes a first knowledge sub-graph, a second knowledge sub-graph, a third knowledge sub-graph and a fourth knowledge sub-graph; the specific steps of constructing the fire protection multi-related knowledge graph include:

[0059] S101, setting the fire protection layout data of buildings in different fields as the first fire protection sub-information stored in the first knowledge sub-graph, the fire protection law data as the second fire protection sub-information stored in the second knowledge sub-graph, the fire and corresponding processing information data as the third fire protection sub-information stored in the third knowledge sub-graph, and the fire protection simulation and management data as the fourth fire protection sub-information stored in the fourth knowledge sub-graph;

[0060] S102, based on the first fire sub-information, the second fire sub-information, the third fire sub-information and the fourth fire sub-information, hierarchically classify each fire sub-information by using a set classification criterion combined with a classification algorithm to obtain multi-level first fire sub-information, second fire sub-information, third fire sub-information and fourth fire sub-information;

[0061] Furthermore, the classification criteria specifically include:

[0062] Data analysis is performed on the fire protection layout data of buildings in different fields, fire protection law data, fire and corresponding processing information data, and fire simulation and management data, respectively, to obtain the second-level classification data corresponding to each major category, and so on, to obtain the hierarchical data corresponding to each level, among which the first-level stratification is the fire protection layout data of buildings in different fields, fire protection law data, fire and corresponding processing information data, and fire simulation and management data.

[0063] For example, the first level is:

[0064] The first level of classification is directly based on four categories of basic data sources, each of which constitutes the basic structure of the knowledge sub-graph;

[0065] Fire protection layout data of buildings in different fields (first knowledge sub-graph):

[0066] Data types: including fire protection design drawings, safety passage layouts, fire protection facility locations (such as fire extinguishers, fire hydrants, sprinkler systems, etc.), emergency exit instructions, etc. of various types of buildings (such as residential, commercial, industrial, etc.); specifically, the specific types corresponding to the second level can be different types of building categories, such as residential, commercial, industrial, etc.; the third level further refines residential, commercial, and industrial, such as residential housing is divided into high-end residential, mid-range residential, and low-end residential, or densely populated or sparsely populated, according to residential prices, geographical locations, and the degree of population concentration; industrial buildings are divided into different levels according to the safety level of planned building requirements, such as the higher the fire protection level of chemical plant buildings, the better, while the fire protection level of ordinary small factories is lower; classification level division, specific division according to specific actual needs; the following fire protection law data, fire and corresponding processing information data, and fire protection simulation and management data are also divided similarly according to this process;

[0067] Data information: fire safety requirements, evacuation route planning, fire zoning, etc. specific to each building type.

[0068] Fire law data (second knowledge sub-graph):

[0069] Data type: national or regional fire regulations, industry standards, building codes, etc.

[0070] Data information: specific provisions of laws and regulations, update records, scope of application, legal responsibilities, etc.

[0071] Fire and corresponding processing information data (third knowledge sub-graph):

[0072] Data types: historical fire cases, fire cause analysis, fire fighting action reports, casualties, etc.

[0073] Data information: time and location of the fire, type of fire source, damage assessment, key decision points in the rescue process, etc.

[0074] Fire simulation and management data (the fourth knowledge sub-graph):

[0075] Data types: fire simulation experiment results, emergency plan drill records, risk management strategies, etc.

[0076] Data information: setting of simulation conditions, comparison between expected and actual results, discovery of problems during the drill and improvement measures, etc.

[0077] In the process of formulating classification criteria corresponding to fire law data, fire and corresponding processing information data, and fire simulation and management data, reference is made to the classification process of building fire layout data in different fields. Technical personnel in this field set major categories according to this embodiment for specific detailed stratification. The specific number of stratification levels is set by technical personnel in this field based on the actual data storage requirements and hardware facility configuration. The more detailed the data division, the more accurate the corresponding answer to the subsequent search.

[0078] S103, according to the classified fire protection sub-information corresponding to the jth level in the obtained i-th fire protection sub-information, obtain the ;in represents the entity triple set corresponding to the j-th level classification fire protection sub-information in the i-th fire protection sub-information;

[0079] S104, repeating the process of S103 to obtain the entity triple set corresponding to each level in the i-th fire protection sub-information, and marking each entity triple set with a hierarchical label; further, the hierarchical label marking is to mark each entity triple set with its corresponding hierarchical level;

[0080] S105. Obtain the knowledge sub-graph corresponding to the i-th fire protection sub-information according to the entity triple set in the i-th fire protection sub-information annotated with the hierarchical labels, and repeat the processes S103-S105 to obtain the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph, and the fourth knowledge sub-graph;

[0081] S106, according to the fire protection sub-information stored in the node corresponding to the jth level in the i-th knowledge sub-graph and the fire protection sub-information stored in the node corresponding to the jth level in the k-th knowledge sub-graph, through a similar matching algorithm, obtain the two nodes with the highest node matching degree and the lowest load in the j-th level. and ;in represents the hth node in the jth level in the i-th knowledge subgraph, represents the g-th node in the j-th level in the k-th knowledge subgraph;

[0082] S107, Utilization and and the corresponding node matching degree, construct the graph connection between the i-th knowledge subgraph and the k-th knowledge subgraph at the j-th level;

[0083] S108, repeat the process of S106-S107 to obtain the graph connection of the same level corresponding to the i-th knowledge sub-graph and the k-th knowledge sub-graph;

[0084] S109, repeating the process of S106-S108 to obtain the corresponding graph connections between any two knowledge sub-graphs in the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph, and the fourth knowledge sub-graph;

[0085] S110, construct a fire protection multi-association knowledge graph based on the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph and the fourth knowledge sub-graph and all corresponding graph connections.

[0086] By carefully constructing a multi-association fire protection knowledge graph, we have achieved efficient organization and utilization of complex information in the fire protection field. Specifically, we first ensured the systematicness and orderliness of the data by classifying the building fire protection layout, fire protection laws, fire handling information, and simulation management data in different fields at multiple levels. This structured classification method enables each level of data to be accurately refined into the most needed application scenarios, such as formulating corresponding fire protection strategies for different types of buildings or safety levels; in further data processing, an entity extraction algorithm is used to obtain entity triple sets from each level of classification and annotate them with hierarchical labels, which not only improves the readability of the data, but also provides a solid foundation for the subsequent knowledge graph construction. Through similarity matching algorithms, the best matching nodes are found and connections are established between different knowledge sub-graphs. This process effectively integrates scattered information resources, forms an organic whole, and enhances the connectivity and practicality of the knowledge graph. The fire protection multi-association knowledge graph finally constructed can achieve cross-domain and cross-level data collaboration, greatly improving the efficiency and accuracy of user queries. It can quickly locate core issues, provide professional and accurate fire assessment plans and fire protection strategy plans, and reduce search delays and inaccurate answers. In addition, the graph is highly flexible and scalable, and can be continuously updated and improved with the addition of new data, ensuring the timeliness and reliability of information. This efficient resource integration and intelligent analysis capability provides strong support for fire management and emergency response, and significantly improves the ability and effectiveness of responding to emergencies such as fires.

[0087] S2. Obtain user input semantic information and corresponding scene feature space through the configured multi-input intention reasoning model, and determine the main knowledge subgraph and collaborative knowledge subgraph in the fire protection multi-association knowledge graph through the user input semantic information, the corresponding scene feature space and the fire protection multi-association knowledge graph, and perform the main-collaborative knowledge subgraph switching according to the scene feature space through the configured main-cooperative switching node;

[0088] Furthermore, in this embodiment, the multi-input intention reasoning model is constructed through the pre-trained scene reasoning model and the user demand reasoning library constructed by the reasoning process data such as voice, text, gesture, and lip shape in the historical reasoning process; the user demand reasoning library includes the voice, text, gesture, lip shape and other information input by the user and the corresponding inferred user demand text information and the corresponding scene space; for example: the user asks: Please output a fire prevention plan based on the characteristics of the building. That is, based on the text information, it is inferred that the user's demand is a fire prevention plan, and the corresponding scene is a building scene; the pre-trained scene reasoning model is constructed by combining the existing voice, text, gesture, lip shape recognition algorithm and the deep reasoning algorithm, and is used to obtain the user demand intention and the corresponding scene space characteristics according to the user input information.

[0089] Furthermore, in this embodiment, the steps of determining the main knowledge sub-graph and the collaborative knowledge sub-graph include:

[0090] S201, obtaining user historical input semantic information, and obtaining the scene type corresponding to each user input semantic information through a scene classification algorithm;

[0091] S202, according to the semantic information input by the user under the corresponding scene type, the number of historical visits and search frequency of each knowledge sub-graph node, and the fire protection multi-association knowledge graph, obtain the association matching degree corresponding to each scene type and the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph, and the fourth knowledge sub-graph through an association matching algorithm;

[0092] S203, taking the knowledge sub-graph corresponding to the maximum correlation matching degree as the main knowledge sub-graph, and taking the remaining knowledge sub-graphs as collaborative knowledge sub-graphs;

[0093] Furthermore, in this embodiment, the main knowledge subgraph adopts a full-node search method; the collaborative knowledge subgraph adopts a difference search method; further, in this embodiment, the difference search step includes:

[0094] S2031. According to the number of historical visits and search frequency of each knowledge sub-graph node in the collaborative knowledge sub-graph under each scenario type, obtain the proportion of the number of search nodes corresponding to each collaborative knowledge sub-graph and the proportion of the search frequency of a single node in each collaborative knowledge sub-graph;

[0095] S2032, obtaining the search importance corresponding to each collaborative knowledge subgraph according to the product of the association matching degree between each collaborative knowledge subgraph and the corresponding scene type and the proportion of the corresponding search node number, and constructing the first difference index information between each scene type and each collaborative knowledge subgraph based on the construction principle of the search importance based on the database index;

[0096] S2033, obtaining the search importance of the corresponding single node in each collaborative knowledge subgraph according to the product of the association matching degree between each collaborative knowledge subgraph and the corresponding scene type, the proportion of the number of corresponding search nodes and the proportion of the search frequency of a single node, and constructing the second difference index information using the search importance of the single node in each collaborative knowledge subgraph and the corresponding scene type;

[0097] S2034. Configure a main-cooperative switching node in the fire protection multi-association knowledge graph, and save the first difference index information and the second difference index information in the main-cooperative switching node. According to the real-time acquired user input semantic information and the scene type in the corresponding scene feature space, the main-cooperative knowledge sub-graph switching and the fire protection sub-information difference search are performed through the main-cooperative switching node.

[0098] This process significantly improves the accuracy and response speed of fire information query through the multi-input intention reasoning model and the main-collaborative knowledge sub-graph switching mechanism. Specifically, the multi-input intention reasoning model combined with the pre-trained scene reasoning model and the user demand reasoning library can accurately infer the user's actual needs and the corresponding scene feature space from multiple input methods such as voice, text, gestures, and lip shape, ensuring a more comprehensive and accurate understanding of the user's intentions; in the process of determining the main knowledge sub-graph and the collaborative knowledge sub-graph, the system first identifies the scene type corresponding to the user's input semantic information through the scene classification algorithm, and then uses the association matching algorithm to calculate the association matching degree between each scene type and each knowledge sub-graph, and selects the one with the highest matching degree as the main knowledge sub-graph, and the rest as collaborative knowledge sub-graphs. This design enables the system to quickly focus on the most relevant fire information for a specific scene and provide professional and targeted solutions; in particular, the main knowledge sub-graph adopts full-node search to ensure that key information is not missed; while the collaborative knowledge sub-graph uses differential search optimization to build the first and second differential index information based on the number of historical visits and search frequency, improving search efficiency. The existence of the master-cooperator switching node enables the system to flexibly switch between different scenarios, ensuring the consistency and accuracy of information retrieval; in summary, this process not only solves the problem that the existing technology cannot provide accurate solutions for specific scenarios, but also greatly shortens the search delay and provides instant and personalized fire prevention plans. By continuously updating the user demand reasoning library and optimizing the difference index, the system can continuously improve its performance and adapt to new firefighting challenges, thereby providing users with more intelligent and efficient fire safety services. This approach effectively promotes the scientific and intelligent development of firefighting work and enhances the ability to respond to emergencies such as fires.

[0099] S3. Construct a multi-graph joint reasoning model, input the main knowledge sub-graph and the collaborative knowledge sub-graph information, the user input semantic information and the scene feature space into the multi-graph joint reasoning model, obtain the user fire demand plan set, and use the user fire demand plan set to construct the optimal reasoning search path and reasoning approximation loss function to train the multi-graph joint reasoning model, and obtain the trained multi-graph joint reasoning model;

[0100] For further information, see Figure 2 In this embodiment, the steps of building and training the multi-graph joint reasoning model include:

[0101] S301, constructing a reasoning input sequence according to user input semantic information, corresponding scene feature space, and fire protection multi-association knowledge graph information;

[0102] S302, according to the scene feature space in the reasoning input sequence, determine the main knowledge subgraph and the collaborative knowledge subgraph information through the information call layer, and call the main-collaborative switching node through the determined information to switch the fire protection multi-association knowledge graph between the main and collaborative knowledge subgraphs, and obtain the first difference index information and the second difference index information saved in the main-collaborative switching node; further, the information call layer in this embodiment is constructed by the SAC algorithm;

[0103] S303, input the switching information of the main-collaborative knowledge sub-graph switching, the user input semantic information, the first difference index information and the second difference index information into the search path layer, and obtain the full search main path and the first difference search path, the second difference search path, and the third difference search path; the search path layer in this embodiment is constructed by the path generation algorithm;

[0104] Furthermore, the full search main path in this step corresponds to the main knowledge subgraph; the first difference search path corresponds to the knowledge subgraph one contained in the collaborative knowledge subgraph; the second difference search path corresponds to the knowledge subgraph two contained in the collaborative knowledge subgraph; the third difference search path corresponds to the knowledge subgraph three contained in the collaborative knowledge subgraph; the knowledge subgraph one, knowledge subgraph two and knowledge subgraph three contained in the collaborative knowledge subgraph are the remaining three knowledge subgraphs of the first knowledge subgraph, the second knowledge subgraph, the third knowledge subgraph and the fourth knowledge subgraph excluding the remaining three knowledge subgraphs that are used as the main knowledge subgraph, and the knowledge subgraph one, the knowledge subgraph two and the knowledge subgraph three correspond one-to-one to the remaining three knowledge subgraphs.

[0105] S304, inputting the full search main path, the first difference search path, the second difference search path, the third difference search path and the determined main-collaborative knowledge sub-graph information into the matching search layer to search for the initial fire plan of the corresponding scene, and obtaining the fire plan sequence of the corresponding scene and the corresponding search matching degree and search matching loss; further, in this embodiment, the matching search layer is constructed by the attention matching algorithm;

[0106] S305, setting a search matching degree threshold, when at least one of the search matching degrees of the fire emergency plans is greater than the search matching degree threshold, the fire emergency plan that meets the threshold condition and corresponds to the maximum search matching degree is used as the fire emergency plan corresponding to the current scene;

[0107] S306, setting a fire plan screening threshold. When the search matching degrees of the fire plans are all less than or equal to the search matching degree threshold, the fire plan information with the corresponding search matching degree greater than the fire plan screening threshold in the fire plan sequence of the corresponding scene is used as the fire plan reasoning information sequence through the fire plan screening threshold;

[0108] S307, input the corresponding scene feature space, the fire plan reasoning information sequence and the full search main path, the first difference search path, the second difference search path, and the third difference search path into the fusion path reasoning layer constructed by the convolutional attention algorithm and the path generation algorithm, obtain the corresponding full search reasoning main path, the first difference reasoning path, the second difference reasoning path, and the third difference reasoning path, and build the search matching degree threshold into all the reasoning paths;

[0109] Furthermore, in this embodiment, the full search reasoning main path is obtained by the full search main path through the fusion path reasoning layer; the first difference reasoning path is obtained by the first difference search path through the fusion path reasoning layer; the second difference reasoning path is obtained by the second difference search path through the fusion path reasoning layer; and the third difference reasoning path is obtained by the third difference search path through the fusion path reasoning layer.

[0110] S308, input the full search reasoning main path, the first difference reasoning path, the second difference reasoning path, the third difference reasoning path and the corresponding main-collaborative knowledge sub-graph information into the scene reasoning layer constructed by the multimodal feature fusion scene text recognition algorithm, and obtain each fire plan reasoning information on the full search reasoning main path and each difference reasoning path and the corresponding reasoning search matching degree;

[0111] Furthermore, in this embodiment, each fire plan reasoning information and the corresponding reasoning search matching degree on each difference reasoning path refer to the reasoning search matching degree obtained by a single reasoning of information reasoning on the first difference reasoning path, the second difference reasoning path, and the third difference reasoning path respectively.

[0112] S309, according to the reasoning search matching degree corresponding to each fire plan reasoning on the full search reasoning main path and each difference reasoning path and the search matching degree threshold built in the corresponding reasoning, obtain the reasoning search loss corresponding to each reasoning process on the full search reasoning main path and each difference reasoning path;

[0113] S310, setting an approximation loss threshold, and constructing an inference approximation loss function using the error between the sum of the inference search loss corresponding to the inference path of each inference process and the approximation loss threshold;

[0114] S311. Set the training loss threshold and training cycle, and use the sum of the inference approximation loss function and the search matching loss to construct the multi-graph joint reasoning model training loss function. When the inference search loss of each inference path in the training cycle is less than the approximation loss threshold and the multi-graph joint reasoning model training loss function is less than the training loss threshold, the trained multi-graph joint reasoning model is obtained.

[0115] This process significantly improves the efficiency, accuracy and intelligence level of fire demand plan generation by constructing a multi-graph joint reasoning model. Specifically, the system first constructs a reasoning input sequence based on the user input semantic information and scene feature space, and uses the SAC algorithm to determine the main knowledge subgraph and the collaborative knowledge subgraph in the information call layer to realize the main-cooperative switching node configuration, thereby obtaining differential index information. Then, the full search main path and multiple differential search paths are generated through the path generation algorithm to ensure comprehensive coverage of different knowledge subgraphs. The matching search layer uses the attention matching algorithm to evaluate the matching degree and loss of the initial fire plan, sets the threshold to screen the optimal plan, and improves the accuracy of plan selection; the fusion path reasoning layer combines the convolutional attention algorithm and the path generation algorithm to optimize the reasoning path, and has a built-in search matching degree threshold to improve accuracy, ensuring that the most suitable solution can be found in each reasoning process. The scene reasoning layer further refines the fire plan reasoning information on each path through the scene text recognition algorithm with multimodal feature fusion, enhancing the adaptability and flexibility of the system. Based on this, the system can provide personalized and professional fire plans according to the characteristics of different scenarios, greatly shortening the response time and reducing human errors; in addition, by constructing an inference approximation loss function and setting a training loss threshold, the system can automatically adjust parameters, optimize model performance, and ensure the quality of the output plan. This method not only improves the speed and accuracy of fire plan generation, but also enhances the robustness and learning ability of the system; continuous training allows the model to continuously adapt to new data and maintain its advancement and practicality; ultimately, the invention provides users with an intelligent and efficient fire safety service platform, significantly improving the ability to respond to emergencies such as fires, and protecting the lives and property of the public. This efficient, accurate, and intelligent fire plan generation mechanism provides strong technical support for fire management and emergency response, and promotes the scientific and intelligent development of firefighting work.

[0116] S4. Based on the real-time acquired user input semantic information and the corresponding scene feature space, the user fire demand plan is obtained through the trained multi-graph joint reasoning model, and the real-time user input semantic information, scene feature space and the corresponding user fire demand plan are saved in the fire multi-association knowledge graph.

[0117] Furthermore, the fire data processing method proposed in this embodiment can quickly and accurately provide fire emergency plans and related legal and regulatory issues related to corresponding fields and corresponding scenarios.

[0118] For example: User asks: Please output a fire prevention plan based on the characteristics of the building. According to the content of the question, for a building fire plan, the main corresponding domain data is the first fire sub-information content in the first knowledge sub-graph, that is, the first knowledge sub-graph is the main knowledge sub-graph, and the main knowledge sub-graph is used to comprehensively obtain the corresponding building fire layout, personnel gathering distribution and building pattern. Through the third fire sub-information content in the collaborative third knowledge sub-graph and the fourth fire sub-information content in the fourth knowledge sub-graph, the corresponding building fire prevention plan is collaboratively and quickly generated and corresponding specific examples can be given;

[0119] Answer: According to statistics, there are 8 households living in the building, and each household has a smoke-triggered sprinkler device in their room. Monitoring shows that the smoke-triggered sprinkler devices are in good health and can be triggered and used at any time. The specific usage methods and corresponding distribution of the smoke-triggered sprinkler devices are as follows.

[0120] Question: What is the latest fire protection policy for chemical plants, and what should we do? According to this question, the main area is still the fire protection plan for the corresponding building field, that is, the first knowledge sub-graph is the main knowledge sub-graph, and the corresponding building fire protection layout is fully obtained through the main knowledge sub-graph; through the fire protection layout, building pattern and safety requirements information about chemical plant buildings in the subordinate categories within the first fire protection sub-information content in the main knowledge sub-graph, in coordination with the fire protection laws and regulations information in the second fire protection sub-information in the second knowledge sub-graph and the historical fire and corresponding handling information in the third fire protection sub-information in the third knowledge sub-graph, the corresponding chemical plant building fire protection plan is quickly generated and corresponding specific examples can be given;

[0121] Answer: According to the relevant policies issued on November 28, 2024, the latest chemical fire protection policy shows a tightening trend and is more stringent than in the past. Based on this trend change, we can layout fire protection products for our company.

[0122] Example 2

[0123] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, an AI intelligent data processing method for the fire protection field is implemented.

[0124] A computer-readable storage medium stores computer instructions, which, when executed, execute an AI intelligent data processing method for the fire protection field.

[0125] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are within the protection of the present invention.

[0126] If the disclosed technical solution involves personal information, the product using the disclosed technical solution has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the disclosed technical solution involves sensitive personal information, the product using the disclosed technical solution has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. An AI intelligent data processing method for fire protection, characterized in that the steps include: S1. Obtain building fire layout data, fire law data, fire and corresponding processing information data, and fire simulation and management data in different fields, and construct a fire protection multi-association knowledge graph through natural language algorithms and graph databases; S2. Obtain user input semantic information and corresponding scene feature space through the configured multi-input intention reasoning model, and determine the main knowledge subgraph and collaborative knowledge subgraph in the fire protection multi-association knowledge graph through the user input semantic information, the corresponding scene feature space and the fire protection multi-association knowledge graph, and perform the main-collaborative knowledge subgraph switching according to the scene feature space through the configured main-cooperative switching node; S3. Construct a multi-graph joint reasoning model, input the main knowledge sub-graph and the collaborative knowledge sub-graph information, the user input semantic information and the scene feature space into the multi-graph joint reasoning model, obtain the user fire demand plan set, and use the user fire demand plan set to construct the optimal reasoning search path and reasoning approximation loss function to train the multi-graph joint reasoning model, and obtain the trained multi-graph joint reasoning model; S4. Based on the real-time acquired user input semantic information and the corresponding scene feature space, the user fire demand plan is obtained through the trained multi-graph joint reasoning model, and the real-time user input semantic information, scene feature space and the corresponding user fire demand plan are saved in the fire multi-association knowledge graph.

2. The AI ​​intelligent data processing method for fire protection field according to claim 1, characterized in that: The fire protection multi-related knowledge graph includes a first knowledge sub-graph, a second knowledge sub-graph, a third knowledge sub-graph and a fourth knowledge sub-graph; the specific steps of constructing the fire protection multi-related knowledge graph include: S101, setting the fire protection layout data of buildings in different fields as the first fire protection sub-information stored in the first knowledge sub-graph, the fire protection law data as the second fire protection sub-information stored in the second knowledge sub-graph, the fire and corresponding processing information data as the third fire protection sub-information stored in the third knowledge sub-graph, and the fire protection simulation and management data as the fourth fire protection sub-information stored in the fourth knowledge sub-graph; S102, based on the first fire sub-information, the second fire sub-information, the third fire sub-information and the fourth fire sub-information, hierarchically classify each fire sub-information by using a set classification criterion combined with a classification algorithm to obtain multi-level first fire sub-information, second fire sub-information, third fire sub-information and fourth fire sub-information; S103, according to the classified fire protection sub-information corresponding to the jth level in the obtained i-th fire protection sub-information, obtain the ;in represents the entity triple set corresponding to the j-th level classification fire protection sub-information in the i-th fire protection sub-information; S104, repeating the process of S103 to obtain the entity triple set corresponding to each level in the i-th fire protection sub-information, and marking each entity triple set with a hierarchical label; S105. Obtain the knowledge sub-graph corresponding to the i-th fire protection sub-information according to the set of entity triples in the i-th fire protection sub-information annotated with hierarchical labels, and repeat the processes S103-S105 to obtain the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph and the fourth knowledge sub-graph.

3. The AI ​​intelligent data processing method for fire protection field as claimed in claim 2, characterized in that: The specific steps of constructing the fire protection multi-association knowledge graph also include: S106, according to the fire protection sub-information stored in the node corresponding to the jth level in the i-th knowledge sub-graph and the fire protection sub-information stored in the node corresponding to the jth level in the k-th knowledge sub-graph, through a similar matching algorithm, obtain the two nodes with the highest node matching degree and the lowest load in the j-th level. and ;in represents the hth node in the jth level in the i-th knowledge subgraph, represents the g-th node in the j-th level in the k-th knowledge subgraph; S107, Utilization and and the corresponding node matching degree, construct the graph connection between the i-th knowledge subgraph and the k-th knowledge subgraph at the j-th level; S108, repeat the process of S106-S107 to obtain the graph connection of the same level corresponding to the i-th knowledge sub-graph and the k-th knowledge sub-graph; S109, repeating the process of S106-S108 to obtain the corresponding graph connections between any two knowledge sub-graphs in the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph, and the fourth knowledge sub-graph; S110, construct a fire protection multi-association knowledge graph based on the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph and the fourth knowledge sub-graph and all corresponding graph connections.

4. The AI ​​intelligent data processing method for fire protection field as claimed in claim 3, characterized in that: The steps of determining the main knowledge sub-graph and the collaborative knowledge sub-graph include: S201, obtaining user historical input semantic information, and obtaining the scene type corresponding to each user input semantic information through a scene classification algorithm; S202, according to the semantic information input by the user under the corresponding scene type, the number of historical visits and search frequency of each knowledge sub-graph node, and the fire protection multi-association knowledge graph, obtain the association matching degree corresponding to each scene type and the first knowledge sub-graph, the second knowledge sub-graph, the third knowledge sub-graph, and the fourth knowledge sub-graph through an association matching algorithm; S203: The knowledge sub-graph corresponding to the maximum correlation matching degree is used as the main knowledge sub-graph, and the remaining knowledge sub-graphs are used as collaborative knowledge sub-graphs.

5. The AI ​​intelligent data processing method for fire protection field as claimed in claim 4, characterized in that: The main knowledge subgraph adopts a full-node search method; the collaborative knowledge subgraph adopts a difference search method; the difference search steps include: S2031. According to the number of historical visits and search frequency of each knowledge sub-graph node in the collaborative knowledge sub-graph under each scenario type, obtain the proportion of the number of search nodes corresponding to each collaborative knowledge sub-graph and the proportion of the search frequency of a single node in each collaborative knowledge sub-graph; S2032. Obtain the search importance corresponding to each collaborative knowledge sub-graph based on the product of the association matching degree between each collaborative knowledge sub-graph and the corresponding scene type and the proportion of the corresponding search nodes, and use the search importance based on the database index construction principle to construct the first difference index information between each scene type and each collaborative knowledge sub-graph.

6. The AI ​​intelligent data processing method for fire protection field as claimed in claim 5, characterized in that: The step of difference searching also includes: S2033, obtaining the search importance of the corresponding single node in each collaborative knowledge subgraph according to the product of the association matching degree between each collaborative knowledge subgraph and the corresponding scene type, the proportion of the number of corresponding search nodes and the proportion of the search frequency of a single node, and constructing the second difference index information using the search importance of the single node in each collaborative knowledge subgraph and the corresponding scene type; S2034. Configure a main-cooperative switching node in the fire protection multi-association knowledge graph, and save the first difference index information and the second difference index information in the main-cooperative switching node. According to the real-time acquired user input semantic information and the scene type in the corresponding scene feature space, the main-cooperative knowledge sub-graph switching and the fire protection sub-information difference search are performed through the main-cooperative switching node.

7. The AI ​​intelligent data processing method for fire protection field according to claim 6, characterized in that: The steps of constructing and training the multi-graph joint reasoning model include: S301, constructing a reasoning input sequence according to user input semantic information, corresponding scene feature space, and fire protection multi-association knowledge graph information; S302, according to the scene feature space in the reasoning input sequence, determine the main knowledge subgraph and the collaborative knowledge subgraph information through the information call layer, and call the main-cooperative switching node through the determined information to switch the fire protection multi-association knowledge graph between the main and collaborative knowledge subgraphs, and obtain the first difference index information and the second difference index information stored in the main-cooperative switching node; S303, input the switching information of the main-collaborative knowledge sub-graph switching, the user input semantic information, the first difference index information and the second difference index information into the search path layer to obtain the full search main path and the first difference search path, the second difference search path, and the third difference search path; S304. Input the full search main path, the first difference search path, the second difference search path, the third difference search path and the determined main-collaborative knowledge sub-graph information into the matching search layer to perform the initial fire plan search for the corresponding scene, and obtain the corresponding scene fire plan sequence and the corresponding search matching degree and search matching loss.

8. The AI ​​intelligent data processing method for fire protection field as claimed in claim 7, characterized in that: The steps of constructing and training the multi-graph joint reasoning model also include: S305, setting a search matching degree threshold, when at least one of the search matching degrees of the fire emergency plans is greater than the search matching degree threshold, the fire emergency plan that meets the threshold condition and corresponds to the maximum search matching degree is used as the fire emergency plan corresponding to the current scene; S306, setting a fire plan screening threshold. When the search matching degrees of the fire plans are all less than or equal to the search matching degree threshold, the fire plan information with the corresponding search matching degree greater than the fire plan screening threshold in the fire plan sequence of the corresponding scene is used as the fire plan reasoning information sequence through the fire plan screening threshold; S307, input the corresponding scene feature space, the fire plan reasoning information sequence and the full search main path, the first difference search path, the second difference search path, and the third difference search path into the fusion path reasoning layer constructed by the convolutional attention algorithm and the path generation algorithm, obtain the corresponding full search reasoning main path, the first difference reasoning path, the second difference reasoning path, and the third difference reasoning path, and build the search matching degree threshold into all the reasoning paths; S308, input the full search reasoning main path, the first difference reasoning path, the second difference reasoning path, the third difference reasoning path and the corresponding main-collaborative knowledge sub-graph information into the scene reasoning layer constructed by the multimodal feature fusion scene text recognition algorithm, and obtain each fire plan reasoning information on the full search reasoning main path and each difference reasoning path and the corresponding reasoning search matching degree; S309, according to the reasoning search matching degree corresponding to each fire plan reasoning on the full search reasoning main path and each difference reasoning path and the search matching degree threshold built in the corresponding reasoning, obtain the reasoning search loss corresponding to each reasoning process on the full search reasoning main path and each difference reasoning path; S310, setting an approximation loss threshold, and constructing an inference approximation loss function using the error between the sum of the inference search loss corresponding to the inference path of each inference process and the approximation loss threshold; S311. Set the training loss threshold and training cycle, and use the sum of the inference approximation loss function and the search matching loss to construct the multi-graph joint reasoning model training loss function. When the inference search loss of each inference path in the training cycle is less than the approximation loss threshold and the multi-graph joint reasoning model training loss function is less than the training loss threshold, the trained multi-graph joint reasoning model is obtained.

9. The AI ​​intelligent data processing method for fire protection field as claimed in claim 8, characterized in that: The full search main path in S303 corresponds to the main knowledge subgraph; the first difference search path corresponds to the knowledge subgraph one contained in the collaborative knowledge subgraph; the second difference search path corresponds to the knowledge subgraph two contained in the collaborative knowledge subgraph; the third difference search path corresponds to the knowledge subgraph three contained in the collaborative knowledge subgraph; the knowledge subgraph one, knowledge subgraph two and knowledge subgraph three contained in the collaborative knowledge subgraph are the remaining three knowledge subgraphs of the first knowledge subgraph, the second knowledge subgraph, the third knowledge subgraph and the fourth knowledge subgraph excluding the remaining three knowledge subgraphs used as the main knowledge subgraph, and the knowledge subgraph one, the knowledge subgraph two and the knowledge subgraph three correspond one-to-one to the remaining three knowledge subgraphs.

10. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, an AI intelligent data processing method for the fire protection field as described in any one of claims 1 to 9 is executed.

Images