Fire risk accurate prediction method and system based on multi-source data fusion knowledge graph

Abstract

The invention discloses a fire risk accurate prediction method and system based on a multi-source data fusion knowledge graph, and belongs to the technical field of machine learning, and the method comprises the following steps: accessing a multi-source heterogeneous data set, integrating sensor data, geographic information, historical data and external data, and obtaining a multi-source heterogeneous data set; a full-size forest three-dimensional model is constructed by means of FDS and SolidWorks, combustion simulation is carried out, environmental parameters are set carefully to ensure that the result is accurate, and a foundation is laid for construction of a multi-source heterogeneous data set; the method comprises the steps of data acquisition, data preprocessing, noise data cleaning, space-time alignment and feature engineering, multi-source heterogeneous data acquisition and preprocessing are carried out, and forest fire related data covering structured, semi-structured and non-structured types are acquired from a multi-source heterogeneous data set. The accuracy and timeliness of fire early warning are remarkably improved, effective fusion of forest fire multi-source heterogeneous data can be achieved, and transparency and traceability in the data fusion process are ensured.

Description

Technical Field

[0001] The present invention belongs to the field of machine learning technology, and in particular relates to a method and system for accurately predicting fire risks based on a multi-source data fusion knowledge graph. Background Art

[0002] Traditional automatic forest fire alarm technology generally collects multiple physical quantities of smoke and heat at the early stage of a fire by arranging smoke and temperature composite detectors and temperature-sensing fire detectors in multiple locations in the forest, and transmits this information to the forest fire alarm controller. The alarm controller has built-in advanced intelligent algorithms that can perform in-depth analysis and real-time comparison of the received data. Once the smoke concentration or temperature is detected to exceed the preset safety threshold, the system will immediately trigger the alarm mechanism. The alarm signal will not only issue a warning in the form of clear and eye-catching sound and light, but will also be synchronously presented in intuitive graphics and text on the human-computer interaction interface, showing in detail the approximate location of the fire, current smoke concentration, temperature values ​​and other key information, ensuring that forest rangers can obtain comprehensive and accurate fire alarm information in the first time, respond quickly and take effective fire-fighting measures.

[0003] The prior art discloses some invention patents in the field of machine learning technology, among which the invention patent with publication number CN119537865B discloses a forest fire risk prediction method. The proposed forest fire risk prediction process includes constructing a forest fire risk assessment dataset, generating a data sequence feature vector, and then sequentially passing through a data feature information interaction module and a spatial neighborhood feature interaction module. The data sequence features are then converted into time series features, and a time series feature capture module and a fully connected neural network are used to obtain prediction values. The data feature information interaction module proposed in the present invention captures the correlation between different types of data, the spatial neighborhood feature interaction module aggregates feature information of adjacent regions, and the time series feature capture module processes a unified time series feature representation to achieve accurate modeling of long time series data. The use of these three modules enables the model to consider information in three dimensions: data, neighborhood, and time, further enhancing the model's prediction accuracy. However, this technical solution still has some shortcomings in its application. The complex and changeable forest environment makes the detector susceptible to interference from external factors, which in turn causes false alarms and data errors. This seriously weakens the reliability of the fire risk prediction system, making it difficult to achieve the goal of real-time and accurate fire monitoring. More seriously, it cannot provide stable and strong support for early fire perception and warning.

[0004] Based on this, the present invention designs a fire risk accurate prediction method and system based on multi-source data fusion knowledge graph to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem that the forest environment is complex and changeable, which makes the detectors susceptible to interference from external factors, thereby causing missed reports and data errors, seriously weakening the reliability of the fire risk prediction system, making it difficult to achieve the goal of real-time and accurate monitoring of fires, and more seriously, unable to stably provide strong support for early fire perception and warning. A fire risk accurate prediction method and system based on multi-source data fusion knowledge graph is proposed.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A fire risk prediction method based on multi-source data fusion knowledge graph includes the following steps:

[0008] Access multi-source heterogeneous data sets, integrate sensor data, geographic information, historical data, and external data, use FDS and SolidWorks to build a full-scale 3D forest model and conduct combustion simulations, carefully set environmental parameters to ensure accurate results, and lay the foundation for the construction of multi-source heterogeneous data sets;

[0009] Data preprocessing, cleaning noisy data, spatiotemporal alignment and feature engineering, multi-source heterogeneous data collection and preprocessing, obtaining forest fire-related data covering structured, semi-structured and unstructured types from multi-source heterogeneous data sets;

[0010] Build reality modeling and define entities and relationships in the fire risk area;

[0011] Dynamic knowledge graph construction, extracting entities from text through NLP, combining rule engines with graph neural networks (GNNs) to infer implicit relationships;

[0012] Knowledge update mechanism, real-time access to new data, dynamic update of the map;

[0013] A multimodal fusion model was constructed, combining GNN-LSTM to process sensor time series data. The knowledge graph was embedded to capture semantic relationships. After the dynamic knowledge graph was constructed, the entity relationships of multi-source heterogeneous data were obtained. On this basis, the knowledge graph was used to comprehensively integrate the multi-source heterogeneous big data of forest fires.

[0014] Risk classification and prediction: output the probability of fire occurrence and diffusion path, and classify them into three levels: no fire, smoldering and open fire through fuzzy neural network;

[0015] Visual early warning and emergency decision support: GIS maps display high-risk areas and push risk levels and firefighting recommendations in real time. Combined with knowledge graphs, it recommends optimal emergency resources, including water source locations and the nearest fire stations.

[0016] As a further description of the above technical solution:

[0017] The sensor data includes temperature, wind speed, smoke, humidity and gas, the geographic information includes GIS maps and vegetation distribution, the historical data includes a fire case library, and the external data includes weather forecasts and satellite remote sensing data.

[0018] As a further description of the above technical solution:

[0019] The spatiotemporal alignment includes unifying timestamps and coordinate systems, and the feature engineering includes extracting temporal features and spatial features.

[0020] As a further description of the above technical solution:

[0021] The multi-source heterogeneous dataset was constructed using the fire dynamics simulation tool FDS and SolidWorks software to construct a full-scale 3D forest geometric model. Based on this, a combustion process simulation experiment was conducted on the full-scale forest model using FDS. During the simulation process, environmental parameters were carefully set, including the specific location coordinates of the fire point, the change curve of the fire heat release rate over time, the spatial distribution of sensors, and the performance parameters of various sensors. This ensured the accuracy and reliability of the simulation results, providing a solid foundation for the subsequent construction of multi-source heterogeneous datasets.

[0022] The multi-source heterogeneous dataset covers multi-dimensional data samples such as images, temperature, smoke concentration and wind speed. In order to achieve effective judgment of forest fires, manual labeling is first used to carefully label each data sample according to the preset fire judgment criteria, and clearly divide it into two types of judgment results: no fire and fire. Subsequently, according to the scientific principle of proportional division, the labeled dataset is reasonably divided into a training set and a validation set. The training set is used for learning and optimizing model parameters, and the validation set is used to evaluate the performance of the model on unseen data. Finally, the divided training set and validation set are input into a carefully constructed multimodal fusion network model for training. During the training process, the model's hyperparameters and optimization algorithms are continuously adjusted to improve the model's ability to integrate different modal data and the accuracy of fire judgment, and ultimately obtain the network model with the best performance.

[0023] As a further description of the above technical solution:

[0024] The fire case database adopts a layered architecture design and is specifically divided into three functional modules: the account management data sub-library, the structured data sub-library, and the unstructured data sub-library. The account management data sub-library focuses on user identity and permission management, and has built-in user account tables and administrator account tables. The user account table records two key pieces of information, the account number and password of ordinary users, in detail, ensuring that users can access case database resources with legitimate identities. The administrator account table stores the administrator's account number and password, granting them the authority to perform two advanced operations on the case database: data maintenance and permission allocation.

[0025] The fire case database management system adopts a layered architecture design. The overall framework consists of a three-layer structure consisting of a bottom data layer, a background processing layer, and an interface display application layer. Each layer has a clear division of labor and works together to achieve efficient operation and function realization of the system;

[0026] The bottom data layer, as the data storage core of the system, is responsible for storing data on various types of forest fire cases. Specifically, this layer stores detailed data on three different types of forest fires: surface fire, crown fire, and underground fire. The data covers multi-dimensional information such as the time, location, cause, spread speed, burned area, and loss, providing a solid data foundation for subsequent data analysis and processing.

[0027] As a further description of the above technical solution:

[0028] The background processing layer is the core logic processing unit of the system. It is built on the bottom data layer. This layer is developed using the Python language and has powerful data processing and logical operation capabilities. Its main function is to respond to various requests issued by the interface display application layer, including data query and data update, and to efficiently manage the data in the bottom data layer. Through multiple operations such as data screening, integration and calculation, the background processing layer can accurately feed back the processing results to the interface display application layer to ensure the normal operation of the system functions.

[0029] As a further description of the above technical solution:

[0030] The interface display application layer serves as a window for user interaction with the system, providing an intuitive and convenient operation interface. Its main functions include user login and registration to ensure that only authorized users can access the system, import and export of data files to facilitate users to import external data into the system or export data from the system, retrieval of data files and content to enable users to quickly locate required information, and addition and deletion operations of data content to meet users' needs for dynamic data management. When the interface display application layer issues a request, the system will pass the request to the background processing layer. The background processing layer retrieves or updates the information of each table in the database according to the request content, and returns the processing results to the interface display application layer, ultimately realizing multiple functions of data visualization, data import and export, and data addition, deletion and modification, providing users with a comprehensive and efficient data management experience.

[0031] As a further description of the above technical solution:

[0032] The data preprocessing is to collect and preprocess multi-source heterogeneous data, obtain forest fire related data covering structured, semi-structured and unstructured types from multi-source heterogeneous data sets, eliminate duplicate data based on key field comparison, use mean calculation to fill in numerical missing values, ensure data quality, and fill in missing values ​​by calculating the mean of multi-source heterogeneous big data of fire, which is obtained by the following formula:

[0033]

[0034] In the formula, n represents the number of non-missing values, e i Represents each non-missing value in a multi-source heterogeneous data set. After the missing values ​​are filled, the obvious format errors of the data are corrected to ensure that the data format is unified and correct. The multi-source heterogeneous data fields are mapped to ensure that the same or similar fields in different data sources have the same identification and definition. On this basis, the Z-score standardization method is used to convert the original data into standard normal distribution data with a mean of 0 and a standard deviation of 1 by calculating the mean and standard deviation of the data. This effectively eliminates the impact of dimensional and order of magnitude differences between different data sources, provides a high-quality data foundation for subsequent data analysis, mining, and modeling, and performs standardized processing on the multi-source heterogeneous big data of smart cities. See the formula:

[0035]

[0036] Where E represents the original forest data, α represents the mean of the multi-source heterogeneous dataset, and β represents the standard deviation of the multi-source heterogeneous dataset. The original data is transformed into a standard normal distribution by simultaneously eliminating the size differences of different features in the dataset.

[0037] As a further description of the above technical solution:

[0038] After the multimodal fusion model and the dynamic knowledge graph are constructed, the entity relationships of multi-source heterogeneous data are obtained. On this basis, based on the knowledge graph, a comprehensive fusion processing of multi-source heterogeneous big data on forest fires is performed, integrating data from different sources, different types and different structures to achieve more comprehensive, accurate and integrated analysis and decision support;

[0039] Extract temporal, spatial, and semantic features from massive forest fire case data and transform them into feature semantics that can be used for decision support. Leveraging ontology and semantic web technologies, the data relationships are logically represented, giving the data clear semantics for machine processing. The forest fire system entities are divided across scales, a hierarchical structure of information units is constructed, and unique codes are assigned to achieve seamless integration between the digital and physical worlds. This promotes the dynamic integration and connection of multiple heterogeneous data sources, ensuring data transparency and efficient integration with complex environments.

[0040] Identify the key fields used to match data records, namely the matching key, eliminate spaces in the matching key, combine multiple fields as a composite matching key, search for similar entities as candidate entities in the constructed knowledge graph based on similarity measurement technology, and calculate the semantic similarity between entities, as shown in the following formula:

[0041]

[0042] Where x1 and x2 represent two given entity word vectors, α represents the angle between the two vectors, x1 and x2 represent the dot product of the two vectors, |x1| and |x2| represent the modulus of the two vectors respectively, and the value range of W(x1, x2) is [-1, 1]. The closer the value is to 1, the more similar the two data entities are. The closer the value is to -1, the greater the difference between the two data entities is. The value of 0 indicates that the two entities are unrelated.

[0043] A fire risk prediction system based on multi-source data fusion knowledge graph, including:

[0044] Multi-source heterogeneous data set access module, integrating sensor data, geographic information, historical data and external data;

[0045] The sensor data includes temperature, wind speed, smoke, humidity and gas, the geographic information includes GIS maps and vegetation distribution, the historical data includes a fire case library, and the external data includes weather forecasts and satellite remote sensing data;

[0046] Data preprocessing module, cleaning noisy data, spatiotemporal alignment and feature engineering;

[0047] The spatiotemporal alignment includes unifying timestamps and coordinate systems, and the feature engineering includes extracting temporal features and spatial features;

[0048] Build reality modeling and define entities and relationships in the fire risk area;

[0049] Dynamic knowledge graph construction module, extracting entities from text through NLP, combining rule engine and graph neural network (GNN) to infer implicit relationships;

[0050] Knowledge update mechanism module, real-time access to new data, dynamic update of the map;

[0051] Build a multimodal fusion model, combine GNN-LSTM to process sensor time series data, and embed knowledge graphs to capture semantic relationships;

[0052] The risk classification and prediction module outputs the probability of fire occurrence and diffusion path, and classifies them into three levels: no fire, smoldering and open fire through fuzzy neural network;

[0053] The visual early warning and emergency decision support module uses a GIS map to display high-risk areas and push risk levels and firefighting recommendations in real time. It also combines knowledge graphs to recommend optimal emergency resources, including water source locations and the nearest fire stations.

[0054] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0055] 1. In this invention, by integrating multi-dimensional data and constructing a semantic knowledge network, the accuracy and timeliness of fire warnings are significantly improved, and the effective fusion of multi-source heterogeneous data on forest fires can be achieved, ensuring transparency and traceability in the data fusion process.

[0056] 2. In the present invention, based on a scientific and systematic multi-source heterogeneous data preprocessing system, a comprehensive and sophisticated preprocessing process is implemented for forest fire multi-source heterogeneous data sets. The process covers multiple key links such as data cleaning, format conversion, field mapping, and outlier processing. Through rigorous data processing strategies and algorithms, it can effectively remove noise in the data, correct format errors, unify field semantics, and eliminate abnormal interference, thereby significantly improving the quality and consistency of forest fire multi-source heterogeneous big data. High-quality and consistent data provides a solid and reliable data foundation for subsequent in-depth data analysis, accurate data fusion, and data-driven forest fire warning, decision-making and other applications, which effectively promotes the intelligent and scientific development of forest fire prevention and control work.

[0057] 3. The present invention has multiple advantages such as data import and export, real-time data update, data visualization and data sharing, which realizes the unified and efficient management of fire case data, is conducive to improving the level of data sharing services, giving full play to the role of fire case data, and providing decision-making support for disaster reduction, emergency response and post-disaster reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of a method and system for accurately predicting fire risks based on a multi-source data fusion knowledge graph proposed by the present invention;

[0059] Figure 2 This is a schematic diagram of a fire risk accurate prediction method based on multi-source data fusion knowledge graph and a system fire case library structure proposed by the present invention;

[0060] Figure 3 This is a schematic diagram of the structure of a fire risk accurate prediction method based on multi-source data fusion knowledge graph and a system fire case library management system proposed by the present invention. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] Please see the attached Figure 1 -Attached Figure 3 The present invention provides a technical solution: a fire risk accurate prediction method based on multi-source data fusion knowledge graph, comprising the following steps:

[0063] Access multi-source heterogeneous data sets, integrate sensor data, geographic information, historical data, and external data, use FDS and SolidWorks to build a full-scale 3D forest model and conduct combustion simulations, carefully set environmental parameters to ensure accurate results, and lay the foundation for the construction of multi-source heterogeneous data sets;

[0064] Data preprocessing, cleaning noisy data, spatiotemporal alignment and feature engineering, multi-source heterogeneous data collection and preprocessing, obtaining forest fire-related data covering structured, semi-structured and unstructured types from multi-source heterogeneous data sets;

[0065] Build reality modeling and define entities and relationships in the fire risk area;

[0066] Dynamic knowledge graph construction, extracting entities from text through NLP, combining rule engines with graph neural networks (GNNs) to infer implicit relationships;

[0067] Knowledge update mechanism, real-time access to new data, dynamic update of the map;

[0068] A multimodal fusion model was constructed, combining GNN-LSTM to process sensor time series data. The knowledge graph was embedded to capture semantic relationships. After the dynamic knowledge graph was constructed, the entity relationships of multi-source heterogeneous data were obtained. On this basis, the knowledge graph was used to comprehensively integrate the multi-source heterogeneous big data of forest fires.

[0069] Risk classification and prediction: output the probability of fire occurrence and diffusion path, and classify them into three levels: no fire, smoldering and open fire through fuzzy neural network;

[0070] Visual early warning and emergency decision support: GIS maps display high-risk areas and push risk levels and firefighting recommendations in real time. Combined with knowledge graphs, it recommends optimal emergency resources, including water source locations and the nearest fire stations.

[0071] Specifically, the sensor data includes temperature, wind speed, smoke, humidity and gas, the geographic information includes GIS maps and vegetation distribution, the historical data includes a fire case library, and the external data includes weather forecasts and satellite remote sensing data.

[0072] Specifically, the spatiotemporal alignment includes unifying timestamps and coordinate systems, and the feature engineering includes extracting temporal features and spatial features.

[0073] Specifically, the multi-source heterogeneous dataset was constructed using the fire dynamics simulation tool FDS and SolidWorks software to build a full-scale three-dimensional forest geometric model. Based on this, a combustion process simulation experiment of the full-scale forest model was carried out based on FDS. During the simulation process, the environmental parameters were carefully set, covering the specific location coordinates of the fire point, the change curve of the fire heat release rate over time, the spatial distribution of sensors, and the performance parameters of various sensors to ensure the accuracy and reliability of the simulation results, providing a solid foundation for the subsequent construction of multi-source heterogeneous datasets.

[0074] The multi-source heterogeneous dataset covers multi-dimensional data samples such as images, temperature, smoke concentration and wind speed. In order to achieve effective judgment of forest fires, manual labeling is first used to carefully label each data sample according to the preset fire judgment criteria, and clearly divide it into two types of judgment results: no fire and fire. Subsequently, according to the scientific principle of proportional division, the labeled dataset is reasonably divided into a training set and a validation set. The training set is used for learning and optimizing model parameters, and the validation set is used to evaluate the performance of the model on unseen data. Finally, the divided training set and validation set are input into a carefully constructed multimodal fusion network model for training. During the training process, the model's hyperparameters and optimization algorithms are continuously adjusted to improve the model's ability to integrate different modal data and the accuracy of fire judgment, and ultimately obtain the network model with the best performance.

[0075] Specifically, the fire case database adopts a layered architecture design and is divided into three functional modules: the account management data sub-library, the structured data sub-library, and the unstructured data sub-library. The account management data sub-library focuses on user identity and authority management, and has built-in user account tables and administrator account tables. The user account table records two key pieces of information, the account number and password of ordinary users, in detail, ensuring that users can access case database resources with legitimate identities. The administrator account table stores the administrator's account number and password, granting them the authority to perform two advanced operations on the case database: data maintenance and authority allocation.

[0076] The fire case database management system adopts a layered architecture design. The overall framework consists of a three-layer structure consisting of a bottom data layer, a background processing layer, and an interface display application layer. Each layer has a clear division of labor and works together to achieve efficient operation and function realization of the system;

[0077] The bottom data layer, as the data storage core of the system, is responsible for storing data on various types of forest fire cases. Specifically, this layer stores detailed data on three different types of forest fires: surface fire, crown fire, and underground fire. The data covers multi-dimensional information such as the time, location, cause, spread speed, burned area, and loss, providing a solid data foundation for subsequent data analysis and processing.

[0078] Specifically, the background processing layer is the core logic processing unit of the system. It is built on the bottom data layer. This layer is developed using the Python language and has powerful data processing and logical operation capabilities. Its main function is to respond to various requests issued by the interface display application layer, including data query and data update, and to efficiently manage the data in the bottom data layer. Through multiple operations such as data screening, integration and calculation, the background processing layer can accurately feedback the processing results to the interface display application layer to ensure the normal operation of the system functions.

[0079] Specifically, the interface display application layer serves as a window for user interaction with the system, providing an intuitive and convenient operation interface. Its main functions include user login and registration to ensure that only authorized users can access the system, import and export of data files to facilitate users to import external data into the system or export data from the system, retrieval of data files and content to enable users to quickly locate required information, as well as addition and deletion operations of data content to meet users' needs for dynamic data management. When the interface display application layer issues a request, the system will pass the request to the background processing layer. The background processing layer will retrieve or update the information of each table in the database according to the request content, and return the processing results to the interface display application layer, ultimately realizing multiple functions such as data visualization, data import and export, and data addition, deletion and modification, providing users with a comprehensive and efficient data management experience.

[0080] Specifically, the data preprocessing involves collecting and preprocessing multi-source heterogeneous data, obtaining forest fire-related data covering structured, semi-structured, and unstructured types from multi-source heterogeneous data sets, eliminating duplicate data based on key field comparison, and filling in missing numerical values ​​using mean calculation to ensure data quality. The missing values ​​are filled by calculating the mean of the multi-source heterogeneous big data on fire, which is obtained by the following formula:

[0081]

[0082] In the formula, n represents the number of non-missing values, e i Represents each non-missing value in a multi-source heterogeneous data set. After the missing values ​​are filled, the obvious format errors of the data are corrected to ensure that the data format is unified and correct. The multi-source heterogeneous data fields are mapped to ensure that the same or similar fields in different data sources have the same identification and definition. On this basis, the Z-score standardization method is used to convert the original data into standard normal distribution data with a mean of 0 and a standard deviation of 1 by calculating the mean and standard deviation of the data. This effectively eliminates the impact of dimensional and order of magnitude differences between different data sources, provides a high-quality data foundation for subsequent data analysis, mining, and modeling, and performs standardized processing on the multi-source heterogeneous big data of smart cities. See the formula:

[0083]

[0084] Where E represents the original forest data, α represents the mean of the multi-source heterogeneous dataset, and β represents the standard deviation of the multi-source heterogeneous dataset. The original data is transformed into a standard normal distribution by simultaneously eliminating the size differences of different features in the dataset.

[0085] Specifically, after the multimodal fusion model and the dynamic knowledge graph are constructed, the entity relationships of multi-source heterogeneous data are obtained. On this basis, based on the knowledge graph, the multi-source heterogeneous big data of forest fires is comprehensively fused and processed, and data from different sources, different types and different structures are integrated together to achieve more comprehensive, accurate and integrated analysis and decision support;

[0086] Extract temporal, spatial, and semantic features from massive forest fire case data and transform them into feature semantics that can be used for decision support. Leveraging ontology and semantic web technologies, the data relationships are logically represented, giving the data clear semantics for machine processing. The forest fire system entities are divided across scales, a hierarchical structure of information units is constructed, and unique codes are assigned to achieve seamless integration between the digital and physical worlds. This promotes the dynamic integration and connection of multiple heterogeneous data sources, ensuring data transparency and efficient integration with complex environments.

[0087] Identify the key fields used to match data records, namely the matching key, eliminate spaces in the matching key, combine multiple fields as a composite matching key, search for similar entities as candidate entities in the constructed knowledge graph based on similarity measurement technology, and calculate the semantic similarity between entities, as shown in the following formula:

[0088]

[0089] Where x1 and x2 represent two given entity word vectors, α represents the angle between the two vectors, x1 and x2 represent the dot product of the two vectors, |x1| and |x2| represent the modulus of the two vectors respectively, and the value range of W(x1, x2) is [-1, 1]. The closer the value is to 1, the more similar the two data entities are. The closer the value is to -1, the greater the difference between the two data entities is. The value of 0 indicates that the two entities are unrelated.

[0090] A fire risk prediction system based on multi-source data fusion knowledge graph, including:

[0091] Multi-source heterogeneous data set access module, integrating sensor data, geographic information, historical data and external data;

[0092] The sensor data includes temperature, wind speed, smoke, humidity and gas, the geographic information includes GIS maps and vegetation distribution, the historical data includes a fire case library, and the external data includes weather forecasts and satellite remote sensing data;

[0093] Data preprocessing module, cleaning noisy data, spatiotemporal alignment and feature engineering;

[0094] The spatiotemporal alignment includes unifying timestamps and coordinate systems, and the feature engineering includes extracting temporal features and spatial features;

[0095] Build reality modeling and define entities and relationships in the fire risk area;

[0096] Dynamic knowledge graph construction module, extracting entities from text through NLP, combining rule engine and graph neural network (GNN) to infer implicit relationships;

[0097] Knowledge update mechanism module, real-time access to new data, dynamic update of the map;

[0098] Build a multimodal fusion model, combine GNN-LSTM to process sensor time series data, and embed knowledge graphs to capture semantic relationships;

[0099] The risk classification and prediction module outputs the probability of fire occurrence and diffusion path, and classifies them into three levels: no fire, smoldering and open fire through fuzzy neural network;

[0100] The visual early warning and emergency decision support module uses a GIS map to display high-risk areas and push risk levels and firefighting recommendations in real time. It also combines knowledge graphs to recommend optimal emergency resources, including water source locations and the nearest fire stations.

[0101] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A fire risk accurate prediction method based on multi-source data fusion knowledge graph, characterized by: The following steps are involved: Access multi-source heterogeneous data sets, integrate sensor data, geographic information, historical data, and external data, use FDS and SolidWorks to build a full-scale 3D forest model and conduct combustion simulations, carefully set environmental parameters to ensure accurate results, and lay the foundation for the construction of multi-source heterogeneous data sets; Data preprocessing, cleaning noisy data, spatiotemporal alignment and feature engineering, multi-source heterogeneous data collection and preprocessing, obtaining forest fire-related data covering structured, semi-structured and unstructured types from multi-source heterogeneous data sets; Build reality modeling and define entities and relationships in the fire risk area; Dynamic knowledge graph construction, extracting entities from text through NLP, combining rule engines with graph neural networks (GNNs) to infer implicit relationships; Knowledge update mechanism, real-time access to new data, dynamic update of the map; A multimodal fusion model was constructed, combining GNN-LSTM to process sensor time series data. The knowledge graph was embedded to capture semantic relationships. After the dynamic knowledge graph was constructed, the entity relationships of multi-source heterogeneous data were obtained. On this basis, the knowledge graph was used to comprehensively integrate the multi-source heterogeneous big data of forest fires. Risk classification and prediction: output the probability of fire occurrence and diffusion path, and classify them into three levels: no fire, smoldering and open fire through fuzzy neural network; Visual early warning and emergency decision support: GIS maps display high-risk areas and push risk levels and firefighting recommendations in real time. Combined with knowledge graphs, it recommends optimal emergency resources, including water source locations and the nearest fire stations.

2. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 1 is characterized in that: The sensor data includes temperature, wind speed, smoke, humidity and gas, the geographic information includes GIS maps and vegetation distribution, the historical data includes a fire case library, and the external data includes weather forecasts and satellite remote sensing data.

3. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 2 is characterized in that: The spatiotemporal alignment includes unifying timestamps and coordinate systems, and the feature engineering includes extracting temporal features and spatial features.

4. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 3 is characterized in that: The multi-source heterogeneous dataset was constructed using the fire dynamics simulation tool FDS and SolidWorks software to construct a full-scale three-dimensional forest geometric model. Based on this, a combustion process simulation experiment was conducted on the full-scale forest model using FDS. During the simulation process, environmental parameters were carefully set, including the specific location coordinates of the fire starting point, the change curve of the fire heat release rate over time, the spatial distribution of sensors, and the performance parameters of various sensors. This ensured the accuracy and reliability of the simulation results, providing a solid foundation for the subsequent construction of multi-source heterogeneous datasets. The multi-source heterogeneous dataset covers multi-dimensional data samples such as images, temperature, smoke concentration and wind speed. In order to achieve effective judgment of forest fires, manual labeling is first used to carefully label each data sample according to the preset fire judgment criteria, and clearly divide it into two types of judgment results: no fire and fire. Subsequently, according to the scientific principle of proportional division, the labeled dataset is reasonably divided into a training set and a validation set. The training set is used for learning and optimizing model parameters, and the validation set is used to evaluate the performance of the model on unseen data. Finally, the divided training set and validation set are input into a carefully constructed multimodal fusion network model for training. During the training process, the model's hyperparameters and optimization algorithms are continuously adjusted to improve the model's ability to integrate different modal data and the accuracy of fire judgment, and ultimately obtain the network model with the best performance.

5. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 4 is characterized in that: The fire case database adopts a layered architecture design and is specifically divided into three functional modules: the account management data sub-library, the structured data sub-library, and the unstructured data sub-library. The account management data sub-library focuses on user identity and permission management, and has built-in user account tables and administrator account tables. The user account table records two key pieces of information, the account number and password of ordinary users, in detail, ensuring that users can access case database resources with legitimate identities. The administrator account table stores the administrator's account number and password, granting them the authority to perform two advanced operations on the case database: data maintenance and permission allocation. The fire case database management system adopts a layered architecture design. The overall framework consists of a three-layer structure consisting of a bottom data layer, a background processing layer, and an interface display application layer. Each layer has a clear division of labor and works together to achieve efficient operation and function realization of the system; The bottom data layer, as the data storage core of the system, is responsible for storing data on various types of forest fire cases. Specifically, this layer stores detailed data on three different types of forest fires: surface fire, crown fire, and underground fire. The data covers multi-dimensional information such as the time, location, cause, spread speed, burned area, and loss, providing a solid data foundation for subsequent data analysis and processing.

6. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 5 is characterized in that: The background processing layer is the core logic processing unit of the system. It is built on the bottom data layer. This layer is developed using the Python language and has powerful data processing and logical operation capabilities. Its main function is to respond to various requests issued by the interface display application layer, including data query and data update, and to efficiently manage the data in the bottom data layer. Through multiple operations such as data screening, integration and calculation, the background processing layer can accurately feed back the processing results to the interface display application layer to ensure the normal operation of the system functions.

7. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 6 is characterized in that: The interface display application layer serves as a window for user interaction with the system, providing an intuitive and convenient operation interface. Its main functions include user login and registration to ensure that only authorized users can access the system, import and export of data files to facilitate users to import external data into the system or export data from the system, retrieval of data files and content to enable users to quickly locate required information, and addition and deletion operations of data content to meet users' needs for dynamic data management. When the interface display application layer issues a request, the system will pass the request to the background processing layer. The background processing layer retrieves or updates the information of each table in the database according to the request content, and returns the processing results to the interface display application layer, ultimately realizing multiple functions of data visualization, data import and export, and data addition, deletion and modification, providing users with a comprehensive and efficient data management experience.

8. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 7 is characterized in that: The data preprocessing is to collect and preprocess multi-source heterogeneous data, obtain forest fire related data covering structured, semi-structured and unstructured types from multi-source heterogeneous data sets, eliminate duplicate data based on key field comparison, use mean calculation to fill in numerical missing values, ensure data quality, and fill in missing values ​​by calculating the mean of multi-source heterogeneous big data of fire, which is obtained by the following formula: In the formula, n represents the number of non-missing values, e i Represents each non-missing value in a multi-source heterogeneous data set. After the missing values ​​are filled, the obvious format errors of the data are corrected to ensure that the data format is unified and correct. The multi-source heterogeneous data fields are mapped to ensure that the same or similar fields in different data sources have the same identification and definition. On this basis, the Z-score standardization method is used to convert the original data into standard normal distribution data with a mean of 0 and a standard deviation of 1 by calculating the mean and standard deviation of the data. This effectively eliminates the impact of dimensional and order of magnitude differences between different data sources, provides a high-quality data foundation for subsequent data analysis, mining, and modeling, and performs standardized processing on the multi-source heterogeneous big data of smart cities. See the formula: Where E represents the original forest data, α represents the mean of the multi-source heterogeneous dataset, and β represents the standard deviation of the multi-source heterogeneous dataset. The original data is transformed into a standard normal distribution by simultaneously eliminating the size differences of different features in the dataset.

9. The fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 8 is characterized in that: After the multimodal fusion model and the dynamic knowledge graph are constructed, the entity relationships of multi-source heterogeneous data are obtained. On this basis, based on the knowledge graph, a comprehensive fusion processing of multi-source heterogeneous big data on forest fires is performed, integrating data from different sources, different types and different structures to achieve more comprehensive, accurate and integrated analysis and decision support; Extract temporal, spatial, and semantic features from massive forest fire case data and transform them into feature semantics that can be used for decision support. Leveraging ontology and semantic web technologies, the data relationships are logically represented, giving the data clear semantics for machine processing. The forest fire system entities are divided across scales, a hierarchical structure of information units is constructed, and unique codes are assigned to achieve seamless integration between the digital and physical worlds. This promotes the dynamic integration and connection of multiple heterogeneous data sources, ensuring data transparency and efficient integration with complex environments. Identify the key fields used to match data records, namely the matching key, eliminate spaces in the matching key, combine multiple fields as a composite matching key, search for similar entities as candidate entities in the constructed knowledge graph based on similarity measurement technology, and calculate the semantic similarity between entities, as shown in the following formula: Where x1 and x2 represent two given entity word vectors, α represents the angle between the two vectors, x1 and x2 represent the dot product of the two vectors, |x1| and |x2| represent the modulus of the two vectors respectively, and the value range of W(x1, x2) is [-1, 1]. The closer the value is to 1, the more similar the two data entities are. The closer the value is to -1, the greater the difference between the two data entities is. The value of 0 indicates that the two entities are unrelated.

10. A fire risk accurate prediction system based on multi-source data fusion knowledge graph, the fire risk accurate prediction method based on multi-source data fusion knowledge graph according to claim 9, characterized in that: include: Multi-source heterogeneous data set access module, integrating sensor data, geographic information, historical data and external data; The sensor data includes temperature, wind speed, smoke, humidity and gas, the geographic information includes GIS maps and vegetation distribution, the historical data includes a fire case library, and the external data includes weather forecasts and satellite remote sensing data; Data preprocessing module, cleaning noisy data, spatiotemporal alignment and feature engineering; The spatiotemporal alignment includes unifying timestamps and coordinate systems, and the feature engineering includes extracting temporal features and spatial features; Build reality modeling and define entities and relationships in the fire risk area; Dynamic knowledge graph construction module, extracting entities from text through NLP, combining rule engine and graph neural network (GNN) to infer implicit relationships; Knowledge update mechanism module, real-time access to new data, dynamic update of the map; Build a multimodal fusion model, combine GNN-LSTM to process sensor time series data, and embed knowledge graphs to capture semantic relationships; The risk classification and prediction module outputs the probability of fire occurrence and diffusion path, and classifies them into three levels: no fire, smoldering and open fire through fuzzy neural network; The visual early warning and emergency decision support module uses a GIS map to display high-risk areas and push risk levels and firefighting recommendations in real time. It also combines knowledge graphs to recommend optimal emergency resources, including water source locations and the nearest fire stations.

Images