Public emergency plan intelligent generation and dynamic adjustment method and system

Abstract

The invention provides a method and a system for intelligently generating and dynamically adjusting a public emergency plan, which are oriented to the field of public safety emergency. The method comprises the following steps: fusing multi-source heterogeneous data to construct an extensible domain knowledge graph, establishing a standardized emergency instruction library, and carrying out multi-dimensional information labeling; automatic extraction of disaster elements is realized based on an entity recognition model of deep learning; analyzing association rules among the emergency entities through a semantic relationship mining technology; key information such as a disaster chain and resource distribution is rapidly obtained by using a map reasoning mechanism; carrying out cross-department resource collaborative allocation by adopting a multi-objective optimization algorithm to generate an optimal disposal scheme; a high-dynamic adjustment mechanism is constructed, and an emergency plan is dynamically optimized based on situation evolution prediction and real-time monitoring data; and finally, automatic generation and versioning management of the plan are realized. Through multi-mechanism cooperation of knowledge modeling, intelligent element analysis, semantic reasoning, dynamic optimization and predictive adjustment, the emergency plan generation efficiency, situation adaptability and resource allocation rationality are remarkably improved, and support is provided for quick response and scientific decision under emergencies.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent emergency decision support and knowledge graph, and particularly relates to an emergency plan automatic generation and dynamic adjustment method and system based on a knowledge graph, natural language processing and a multi-objective optimization algorithm. The technology faces a public safety emergency management scene, supports task planning, resource collaborative scheduling and real-time decision optimization under a sudden event, realizes rapid adaptation and scientific decision-making of emergency response through a high-dynamic predictive adjustment mechanism, and significantly improves disposal efficiency and intelligent level under a complex emergency environment. BACKGROUND

[0002] With the development of social informatization and intelligentization, public safety emergencies present the characteristics of multi-source concurrency, high dynamic evolution and high uncertainty. The modern emergency management system requires real-time collaborative response across departments, regions and levels, and the traditional pre-plan management mode generally has bottlenecks such as information integration lag, resource allocation imbalance and decision response delay when facing rapidly changing emergencies. Especially in the complex emergency scene of multi-disaster coupling and multi-task concurrency, the pre-plan compilation and adjustment mechanism relying on human experience is difficult to meet the multiple demands of rapid response, precise strategy and dynamic adaptation, resulting in limited emergency decision efficiency and disposal effect.

[0003] In recent years, the application of artificial intelligence technology in the field of emergency management has been continuously deepened, and technologies such as knowledge graph, natural language processing (NLP) and optimization algorithm provide new technical support for intelligent emergency decision-making. The knowledge graph can effectively integrate meteorological, geological, social and other multi-source heterogeneous data through structured modeling of emergency entities and their associated relationships, support disaster chain reasoning and resource demand prediction; the natural language processing technology can realize semantic analysis and key element extraction of emergency plan text, providing structured input for emergency task decomposition; the multi-objective optimization algorithm shows significant collaborative optimization capability in cross-department resource scheduling and rescue force allocation. However, the existing technology still has the following limitations: first, the real-time fusion and dynamic update mechanism of multi-source data is imperfect, which is difficult to adapt to the rapid evolution of emergencies; second, the entity recognition and relationship extraction precision of emergency elements is insufficient, especially in the multi-modal heterogeneous data scene; third, the resource allocation model focuses on a single optimization target, which is difficult to coordinate the multi-dimensional demands of response time, disposal effect and cost control. In addition, the existing systems generally lack the deep cooperation of knowledge graph and deep learning model, which restricts their application efficiency in dynamic scenario deduction and self-adaptive adjustment; fourth, the current technology mostly adopts a passive response mode, lacks a predictive optimization mechanism based on situation prediction, and is difficult to effectively deal with derivative disasters and chain reactions in emergencies.

[0004] Therefore, an intelligent emergency decision support system capable of deeply integrating knowledge graph, natural language processing and multi-objective optimization technology is urgently needed. Through efficient emergency element identification, disaster correlation mining, resource collaborative modeling and dynamic optimization, the system can realize rapid generation and predictive dynamic adjustment of cross-department emergency plans, so as to meet the core needs of timeliness, adaptability and intelligence of modern emergency management. SUMMARY

[0005] Based on this, the present application proposes a public emergency plan intelligent generation and dynamic adjustment method and system, aiming to realize structured analysis of emergency needs, efficient modeling of resource capabilities, multi-objective optimization of task-resource matching and dynamic response to emergencies by constructing a knowledge graph for the public emergency field, integrating natural language processing technology, entity linking technology and multi-objective optimization algorithm. The method realizes rapid adjustment of plans based on situation prediction by real-time updating of the knowledge graph and rapid intelligence acquisition based on Cypher query, combined with dynamic task allocation based on multi-objective optimization algorithm, to adapt to the high dynamic changes of the emergency environment (such as secondary disasters, resource state mutations, task demand changes), significantly improving the adaptability, timeliness and decision support capability of emergency plans in complex emergency environments.

[0006] The first aspect of the embodiment provides a public emergency plan intelligent generation and dynamic adjustment method, comprising: Collecting external data, and constructing / updating an emergency knowledge graph based on the collected data; Building an emergency instruction library and performing entity annotation, and constructing a named entity recognition model based on a bidirectional long short-term memory network BiLSTM model, a conditional random field model CRF and a random forest model Random Forest, inputting emergency instruction text data into the named entity recognition model to obtain emergency entities involved in the instructions; Constructing a relation extraction model based on a spaCy model and a relation database, inputting emergency instruction text into the relation extraction model to obtain the relations between emergency entities; Constructing an entity information query model based on Cypher language, and querying related disposal information of emergency entities involved in the instructions in the emergency knowledge graph; Constructing a task allocation model based on a multi-objective optimization algorithm NSGA-II, inputting the information queried in the knowledge graph and the relations between emergency entities into the task allocation model to obtain a non-dominated task allocation scheme; Storing the entity of the emergency instruction, the relation between the entities, the disposal information of the emergency entity and the task allocation scheme, and generating an emergency plan in combination with related task background and task principles. Compared with the prior art, the public emergency plan intelligent generation and dynamic adjustment method provided by the application realizes efficient extraction of emergency entities and relationships by constructing an emergency knowledge graph with unified semantics, fusing a deep learning model and a rule model, and improves the accuracy of semantic understanding and graph expression; in combination with a Cypher-based query mechanism and a multi-objective optimization algorithm NSGA-II, the method effectively supports task scheduling and resource matching decision-making, and significantly improves the automation and intelligence level of plan construction; at the same time, the method has good dynamic adaptability and can realize high-dynamic predictive task agile adjustment of the plan when the emergency situation changes (such as resource damage and task addition), thereby overcoming the problems of plan response lag, weak model generalization capability and low entity extraction efficiency in the prior art, and improving the overall robustness and practical application value of the plan system.

[0007] As an optional implementation manner of the first aspect, the step of inputting the segmented emergency instruction text data into the named entity recognition model constructed based on the bidirectional long short-term memory network BiLSTM, the conditional random field CRF model and the random forest Random Forest model to obtain emergency entities in the instruction includes: The segmented emergency instruction text data is input into a preprocessing module for standardization processing to obtain a word vector sequence. Each word is converted into a vector representation through a word embedding technology (such as Word2Vec or GloVe), and the obtained word vector sequence is taken as input. The obtained word vector sequence is input into a bidirectional long short-term memory network BiLSTM model, which captures the dependency relationship of the current word and the previous and subsequent words through forward and backward networks to obtain the context semantic feature of each word. The obtained context semantic feature vector is input into a conditional random field CRF model, which uses a global decoding method to obtain an optimal label sequence by considering the transition relationship between labels. The objective of the CRF model is to maximize the conditional probability of a given input sequence and a label sequence, and the formula is as follows: wherein ψ(y t ,y t-1 ,x t ) is a feature function, which describes the relationship between the label y t , the previous label y t-1 and the current input x t . The model selects the optimal label sequence y by maximizing the conditional probability. To enhance the generalization ability and robustness of the model, multiple BiLSTM-CRF models are used to form an ensemble system according to the random forest idea. Each model is trained on a random subset of the training set to increase model diversity. In the prediction stage, each model predicts the label of the input sequence, and then the final label sequence is determined through the majority voting mechanism.

[0008] As an optional implementation of the first aspect, the relation extraction model is constructed based on the spaCy toolkit and a pre-built emergency relation ontology library, including the following steps: First, the input emergency instruction text is subjected to dependency syntax analysis and entity recognition to obtain the dependency path between entity pairs; Typical relations such as "linkage", "dispatch", "rescue", "impact", and "jurisdiction" between emergency entities are recognized and extracted using the integrated rule templates and custom relation dictionaries in spaCy; For complex semantic or implicit relation sentences, a vector similarity matching mechanism based on word embedding is used to supplement the recognition of implicit entity relations; Finally, the extracted triplets <entity1, relation, entity2> are structured and output, and mapped to the predefined emergency knowledge graph architecture. As an optional implementation of the first aspect, the construction of the emergency knowledge graph is based on the graph database Neo4j, which uses a graph structure data model to represent emergency entities and their relationships, specifically including: The collected external data is subjected to data cleaning and standardization processing, and the identified emergency entities such as "rescue team", "disaster point number", "emergency communication vehicle", and "task number" are taken as nodes (Node) in the graph, and the identified emergency entity relations such as "jurisdiction", "support", "dispatch", "protection", and "disposal" are taken as edges (Edge) for linking to form a complete graph structure; Each node contains a set of attributes a1, a2,..., a n Each attribute a i corresponds to a specific attribute value of the entity (such as disposal capacity, deployment location, current state, responsible person number, etc.), and each edge can be represented as e = (v i , r, v j ), where v i and v j are nodes, and r is a semantic relation label between emergency entities.

[0009] As an optional implementation of the first aspect, the graph query model based on the Cypher language performs information query on the emergency entities involved in the emergency instructions by constructing a query template, and the query statement structure is as follows: MATCH (n:Entity {name: "Rescue Team A"})-[:belongs_to]->(b:Emergency Command) RETURN b.name, b.commander, b.location The above query example is used to retrieve the superior "emergency command department" information connected with the "rescue team A" jurisdiction relationship, including the command department name, commander, and command department location. The graph database quickly responds to the query based on the index mechanism, ensuring that the emergency command and dispatch system has high real-time response capability.

[0010] As an optional implementation of the first aspect, the task allocation model is constructed based on the multi-objective optimization algorithm NSGA-II, including the following steps: A task-resource mapping matrix is constructed, and multiple optimization objective functions such as resource execution capability, resource consumption, task importance, priority, and time window are defined; Based on the knowledge graph query result and the logical relationship between entities, the population individuals of task allocation are initialized, and each individual represents a possible task allocation scheme; The NSGA-II algorithm is used for non-dominated sorting, congestion calculation, and elite selection, and a new generation of task allocation solution space is generated through crossover and mutation operations; In each generation evolution, a graph-driven constraint correction mechanism is introduced to ensure that the obtained scheme meets the emergency logical consistency and resource scheduling feasibility; Finally, multiple task allocation solutions on the Pareto optimal front are output to provide the emergency command decision system with options or strategy optimization.

[0011] The second aspect of the embodiment of the application provides a public emergency plan intelligent generation and dynamic adjustment system, which comprises: A data acquisition module is used to collect emergency-related external data such as disaster environment, disaster situation, resource state, and historical emergency instructions, and provide basic data support for graph construction and task planning. A graph construction / update module is used to construct or update the emergency knowledge graph according to the collected emergency data, and combine the emergency ontology standardized modeling of emergency entities and their attributes and relationships to ensure the integrity and dynamic adaptability of the graph structure. A named entity recognition module is used to construct a named entity recognition model based on a bidirectional long short-term memory network BiLSTM model, a conditional random field CRF model, and a random forest Random Forest model, process emergency instruction text, and identify and extract emergency entities such as rescue teams, emergency equipment, and disaster areas. Relation extraction module: based on the spaCy model and the preset relation library, a relation extraction model is constructed to extract entity relations in emergency instructions, such as "jurisdictional relationship", "dispatching relationship", "supporting relationship" and other emergency logical relationships. Entity query module: a query model based on the Cypher language is constructed to query related emergency entity information and background disposal data in the emergency knowledge graph, supporting multi-dimensional and multi-condition emergency data calling. Task allocation module: based on the multi-objective optimization algorithm NSGA-II, a task allocation model is constructed, and the information of emergency resources and target relations queried in the knowledge graph is input to generate a non-dominated task allocation scheme that meets the performance, coverage rate and resource constraints, so as to optimize resource allocation and task execution efficiency and support real-time resource reallocation in a high dynamic environment. Preparation generation module: the identified entities, extracted relations, queried disposal information and optimized task allocation scheme are integrated to automatically generate an emergency plan in combination with the task background, emergency principles and decision logic, supporting version management and flexible adjustment of the plan.

[0012] The present application is applicable to various emergency scenarios, such as typhoon disaster rescue, urban waterlogging disposal or dangerous chemical leakage accident. In actual application, the system can quickly generate an initial plan according to the disaster description, real-time adjust resource allocation to respond to unexpected situations, and present the disaster situation through a visual interface. For example, in a certain typhoon disaster rescue, the system allocates rescue teams and emergency supplies according to disaster needs, monitors the changes in the disaster area in real time, dynamically adjusts the rescue route, and re-plans the task timing when the weather worsens, finally generating an emergency plan containing a backup scheme. The modular design of the present application makes it easy to integrate into existing emergency command systems, supports distributed deployment and cross-departmental collaboration, and provides strong technical support for the intelligent transformation of modern emergency management.

[0013] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0014] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions serve to explain the present application, and do not constitute an improper limitation on the present application. In the drawings: Figure 1 is a flow chart of a public emergency plan intelligent generation and dynamic adjustment method proposed by the first embodiment of the present application. Figure 2 is a framework diagram of the named entity recognition model in the first embodiment of the present application. Figure 3A framework diagram of a relation extraction model in the first embodiment of the present application. Figure 4 A structure schematic diagram of a public emergency plan intelligent generation and dynamic adjustment method and system according to the second embodiment of the present application.

[0015] The following specific embodiments will further illustrate the present application in combination with the above-mentioned drawings. DETAILED DESCRIPTION

[0016] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The drawings show several embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0018] The specific embodiments of the present application are described in detail below in combination with the accompanying drawings, so that those skilled in the art can understand and implement the present application. The embodiments are only exemplary and do not constitute a limitation on the scope of protection of the present application.

[0019] Please refer to Figure 1 , which is a flowchart of a public emergency plan intelligent generation and dynamic adjustment method provided by an embodiment of the present application, and is described in detail as follows:

[0020] Step S01: Collect external data, and build / update an emergency knowledge graph based on the collected data.

[0021] It should be noted that the present application will collect the text data and Internet data of emergency resources such as rescue equipment, communication equipment and transportation tools from the relevant literature database, database and emergency management related materials. The system collects multi-source heterogeneous data through external data interface source, and constructs an emergency knowledge graph by using graph database technology. The specific operation is as follows: the system first acquires various emergency related information through external data interface source, for example, acquires real-time disaster data of the disaster area through disaster monitoring sensors, extracts equipment allocation information of the provincial fire rescue brigade from the emergency resource database, acquires disposal cases of similar disasters from the historical case library, extracts disposal principles and regulations from the emergency expert system, such as hierarchical response principle and humanitarian rescue requirements, acquires emergency resource information of the surrounding area through public information, such as the rescue team and material reserve situation of the adjacent provinces. The system cleanses and converts these data into a unified format, and then connects to the Neo4j graph database through the py2neo library using the Python script, and empties the database to ensure data consistency. The system creates a provincial administrative area node, and creates multiple emergency command center nodes such as command center 1, command center 2 and command center 3, which are associated with the provincial node through the “MANAGES” relationship. The system creates a commander node for each command center, which is connected to the corresponding center through the “LEADS” relationship. The system creates a rescue operation node coordinated by the command center through the “COORDINATES” relationship, and adds an emergency principle node and an emergency regulation node connected through the “GUIDES” and “CONSTRAINTS” relationships respectively. The system creates a rescue target node associated with the rescue operation through the “HAS_TARGET” relationship. The system continues to create rescue team nodes such as the first fire rescue team, which is affiliated to the command center through the “DEPLOYS” relationship, and creates rescue team nodes affiliated to the rescue team. The system creates a team leader node for the rescue team through the “LEADS” relationship. The system creates rescue equipment nodes such as life detection instruments, breaking tools and first aid equipment, and allocates equipment to the rescue team through the “EQUIPS” relationship. The system creates disaster area nodes such as A county and B county, establishes disaster relationship and mutual aid relationship. The system creates emergency facility nodes such as A county emergency shelter, and associates them with A county through the “HAS_FACILITY” relationship. The system creates communication system nodes such as emergency communication vehicles, which are associated with the command center through the “HAS_COMMUNICATION” relationship, and connected with the disaster area through the “DEPLOYS” relationship. The system creates material reserve nodes such as provincial emergency material warehouse, which are associated with the provincial node through the “HAS_WAREHOUSE” relationship, and connected with the rescue equipment through the “SUPPLIES” relationship.The system creates a medical support node subordinate to a provincial command center, connected through a "HAS MEDICAL TEAM" relationship, and containing medical equipment nodes, such as mobile ICU units, connected through a "HAS EQUIPMENT" relationship. The system creates medicine nodes for the medical team, such as first aid medicine kits, connected through a "HAS MEDICINE" relationship. Finally, the knowledge graph contains nodes such as administrative divisions, command centers, rescue teams, rescue groups, rescue equipment, emergency facilities, communication systems, material reserves, medical support, and medical equipment, and their relationships, stored in a Neo4j database, supporting efficient querying.

[0022] Step S02: Constructing an emergency instruction library and performing entity annotation, and based on a bidirectional long short-term memory network BiLSTM model, a conditional random field model CRF, and a random forest model Random Forest, a named entity recognition model is constructed, and the emergency instruction text data is input into the named entity recognition model to obtain the emergency entities involved in the instructions.

[0023] It should be noted that the emergency instruction text data is annotated with entities and relationships to improve the pre-training efficiency and accuracy of the emergency instruction named entity recognition model.

[0024] For example, the BIO annotation method is used to annotate the entities in the emergency instruction text data. For non-entity characters, a uniform annotation of "0" is used. For entity characters, a composite tagging system is used, which consists of two main parts: first, the "B" is used to mark the starting character of the named entity, and the "I" is used to identify the characters within the named entity (including the end character), to indicate the specific position of the character in the entity; second, the second part of the label is used to indicate the type of the entity to which the character belongs. The BIO annotation label mainly includes: command center (CMD), rescue equipment (EQU), transport vehicle (VEH), emergency facility (FAC), medical resource (MED), location (LOC), person name (PER), organization (ORG), time (TIM), and emergency event (EVE). Taking "Tomorrow morning at 8 o'clock, the first fire rescue team went to the disaster area in A County for rescue" as an example, the annotation result is shown in Table 1.

[0025] Table 1: BIO annotation example Today Morning Time Fire 8 B-TIM ， I-TIM I-TIM I-TIM I-TIM I-TIM B-ORG I-ORG Rescue O First Team Forward I-ORG I-ORG I-ORG I-ORG I-ORG I-ORG B-LOC A County Affected Area Carry out Rescue O O I-LOC I-LOC I-LOC I-LOC Figure 2 Figure 3 Figure 4 ​ ​ ​ ​ ​ ​ ​ O O O O O

[0026] Specifically, a named entity recognition model is constructed based on a bidirectional long short-term memory network BiLSTM model, a conditional random field model CRF, and a random forest model Random Forest. For details, please refer to ​A framework graph of the named entity recognition model. The model is constructed based on a BiLSTM-CRF forest structure, has an integrated learning capability, and is used for high-precision extraction of key entity information in emergency instructions.

[0027] In a specific implementation, the system first constructs a training corpus set wherein X (i) ={x1,x2,...,x T} represents a word sequence of an input sentence, and Y (i) ={y1,y2,...,y T} represents a corresponding entity label sequence. To improve the generalization ability of the model, the system generates N sub-data sets D (1) ,D (2) ,...,D (N) from D by using a Bootstrap resampling method, and respectively trains N BiLSTM-CRF models f (1) ,f (2) ,...,f (N) that are of the same structure but have independent parameters, to form a model forest: Each sub-model internally includes the following main components: Word embedding layer: maps each word x t to a low-dimensional vector representation BiLSTM layer: uses a bidirectional LSTM network to model context information to obtain a context feature representation of each word: CRF layer: defines a transition matrix to model the entire label sequence. The score function of a given sentence is: wherein, is the score of the t-th word being labeled as label y t , and the output from the previous BiLSTM layer. Finally, the model outputs an optimal label path: In the testing stage, multiple sub-models are used in parallel to infer the input text to obtain multiple label sequences The system uses a majority voting method to fuse the prediction results, and finally outputs a label sequence wherein mode(·) represents the mode operation, that is, the label with the highest proportion in the N model predictions.

[0028] Exemplarily, the system inputs the instruction sentence "Tomorrow morning 8 o'clock, the first fire rescue team and the second medical rescue team go to the landslide disaster area of A County, the dangerous chemical leakage point of B area and the waterlogging area of C city caused by the consecutive disasters of typhoon to carry out joint rescue" into the entity extraction model. First, the system encodes each word into a vector through the word vector layer to obtain its corresponding low-dimensional dense representation, and then inputs it into the bidirectional long short-term memory network for context modeling. The forward LSTM extracts the forward semantic features in the natural order of the word sequence, and the backward LSTM obtains the backward semantic features in the reverse order. Finally, the system splices the hidden state vectors of the forward and backward to form a word vector representation that fuses the context information. Then, the system inputs the word vector representation into the conditional random field (CRF) model, models and decodes the entire label sequence by combining the label transition probability, and outputs the optimal entity label sequence. To improve the stability and accuracy of the prediction, the system fuses the output results of multiple BiLSTM-CRF sub-models, and determines the final label by using the majority voting mechanism to complete the accurate identification of the key entities in the instruction text.

[0029] Step S03: Constructing a relationship extraction model based on the spaCy model and the relationship library, inputting the emergency instruction text into the relationship extraction model to obtain the relationship between the emergency entities.

[0030] See ​ , the framework diagram of the relationship extraction model.

[0031] Specifically, the relationship extraction model includes an instruction preprocessing module, a named entity recognition (NER) module, a relationship recognition module, and a knowledge graph disambiguation module. After receiving the instruction text "A county emergency management bureau reports that at 14:30 today, a dangerous chemical storage tank leakage accident occurred in B industrial park. Now instruct the city fire rescue team special squad leader Zhang Ming to immediately lead the heavy chemical defense rescue team, carry professional equipment such as flammable gas detector and chemical adsorption pad, and ride the chemical defense rescue vehicle to dispose, at the same time, coordinate the city medical emergency center to send an ambulance to carry emergency medicine and medical team to the scene, and set up a front command post 500 meters southeast of the accident point to ensure real-time communication with the emergency communication vehicle", the system first completes word segmentation and part-of-speech tagging through the instruction preprocessing module. Then, the NER module uses a deep learning model based on BiLSTM-CRF to perform multi-level semantic analysis on the instruction text, accurately identifying the following key entity information: A county-LOC, 14:30-TIM, B industrial park-LOC, dangerous chemical storage tank leakage accident-EVE, city fire rescue team special squad-ORG, Zhang Ming-PER, flammable gas detector-VEH, emergency medicine-MED, medical team-MED, front command post-CMD, and emergency communication vehicle-VEH.

[0032] Next, the relationship identification module identifies the multi-dimensional semantic relationship between entities based on the spaCy dependency syntax analysis and multi-modal keyword matching rules. The system extracts the emergency response relationship five-tuple <EVE: Chemical Spill Accident, ORG: City Fire Rescue Brigade, disposition, LOC: B Industrial Park, [VEH: Rescue Vehicle + EQU: Flammable Gas Detector]> in the instruction by constructing the "event-subject-action-object-resource" complex syntactic structure, combined with the emergency response keyword library (including "disposition", "rescue", "repair", "evacuation", "containment" and other professional terms), and identifies the auxiliary relationship triple <CMD: Frontline Command, coordination, ORG: City Medical Emergency Center>.

[0033] To avoid entity ambiguity, the system further calls the knowledge graph disambiguation module to accurately disambiguate "B Industrial Park Chemical Tank" by combining the four-dimensional disambiguation feature system (spatial topology feature, time series feature, organizational affiliation feature, and event association feature) in the emergency knowledge graph through a multi-dimensional feature fusion algorithm.

[0034] Finally, the system outputs the above structured information in JSON-LD format for subsequent emergency plan generation and resource scheduling modules, with output fields including subject, object, predicate, resources, spatiotemporal_constraints, and other core dimensions, as well as priority and coordination_units and other extended attributes.

[0035] Step S04: Based on the Cypher language, build an entity information query model to query the relevant disposition information of the emergency entities involved in the instruction in the emergency knowledge graph.

[0036] Specifically, based on the key entities extracted from the instruction sentence, the system locates the corresponding node in the emergency knowledge graph through entity linking results, and deeply queries the equipment configuration information of the rescue team. For example, the system can retrieve the equipment details of the City Fire Rescue Brigade Special Operations Unit (ORG) node, obtain its equipped professional disposition equipment, including EX580 flammable gas detector, RHA-02 heavy chemical protective clothing, CA500 chemical adsorption pad, and the key performance parameters and operation specifications of each equipment, forming a complete equipment capability profile.

[0037] Meanwhile, the system can also query multi-dimensional feature information of the accident site (such as the B industrial park hazardous chemical storage tank area), including infrastructure (such as tank type, volume parameters, MSDS data of stored substances), risk elements (such as leakage substance CAS number, diffusion simulation parameters, heat radiation influence range), emergency resources (such as the location of the nearest fire hydrant, the capacity of the emergency collection pool), and environmental attribute information (such as the prevailing wind direction, real-time weather data such as temperature and humidity). Based on the pre-built "disposal scheme" knowledge network in the graph, the system intelligently matches the optimal disposal process combination under this accident scenario, and dynamically calculates the required rescue force scale combined with real-time monitoring data.

[0038] Finally, the system integrates the query results into a multi-dimensional data matrix conforming to the ICS standard, including the following core elements: rescue force formation, equipment matrix, disposal scheme, and decision support. This output provides full-element, quantifiable data support for subsequent emergency response decisions, supporting the optimization of the entire chain from tactical-level disposal to strategic-level resource scheduling.

[0039] To achieve the above query process, the system uses the graph database query language Cypher to construct standardized query templates, and specific examples are shown in the following table:

[0040] Table 2: Query template examples

[0041] Among them, $facility_names is the set of emergency facility names extracted from the named entity recognition results. The system parses the structured format attribute field response_requirements in the returned results and converts it into a two-dimensional matrix form as the resource demand matrix of each emergency facility for each type of rescue equipment. $teams is the set of rescue team names identified (such as "city fire rescue detachment"). The system summarizes the number of each type of equipment equipped by each team and constructs a resource availability matrix for subsequent task scheduling and resource matching calculations. The query template can be automatically filled by the system according to the emergency input parameters (such as the name of the affected facility or the identification of the rescue team), thereby realizing efficient access and extraction of structured node information in the knowledge graph, ensuring the accuracy and timeliness of emergency task planning.

[0042] Step S05: Based on the multi-objective optimization algorithm NSGA-II, a task allocation model is constructed, and the information queried from the knowledge graph and the relationship between emergency entities are input into the task allocation model to obtain a non-dominated task allocation scheme.

[0043] Specifically, the system first extracts structured data containing elements such as rescue resource capabilities, emergency disposal needs, and response time windows, and constructs a mathematical modeling of the task allocation problem. With the dual optimization goals of maximizing task completion efficiency and minimizing resource scheduling cost, the following dual objective function is defined: maximize the resource adaptation degree and disposal effect of all tasks, and minimize the coordination cost introduced by emergency resource rescheduling.

[0044] The task completion efficiency is defined by the resource adaptation degree function, which measures the matching degree between the professional capabilities of the rescue team and the disposal needs; while the resource scheduling cost is described by the task transfer matrix, which describes the coordination cost generated by the redistribution of emergency resources from the original task to the new task. This task allocation problem has multiple resource type dimensions, task-resource combination space, and emergency response time limit constraints, and can be mathematically classified as a mixed integer nonlinear multi-objective optimization problem.

[0045] To solve the above problem, the system uses the non-dominated sorting genetic algorithm NSGA-II and performs constraint enhancement processing. During the algorithm execution process, the system constructs an initial population and defines the fitness function, where individuals that violate the constraint conditions (such as task completion quality not meeting standards, task platform quantity exceeding limits, etc.) will be subjected to a penalty function for fitness punishment, with the constraint form as follows: where ρ is the penalty factor, Q i is the actual completion quality of task i, Q min is the minimum rescue response requirement threshold.

[0046] The system solves the task allocation problem in parallel under the Apache Spark distributed computing framework, performs selection, crossover, mutation, and other operations to generate new solutions in the population iteration process, continuously approaches the Pareto frontier, and finally outputs a set of non-dominated solutions. Each non-dominated solution represents a set of optional task allocation plans, and the system can select the optimal or suboptimal scheme from them according to the rescue effect priority, resource utilization efficiency, or rescue timeliness strategy.

[0047] Exemplarily, for the scene of "the first branch of the city fire rescue handles the B industrial park hazardous chemical leakage accident" in the instruction, the system calculates the optimal resource allocation and generates multiple disposal schemes according to the professional equipment such as EX580 combustible gas detector, RHA-02 heavy chemical protective clothing and CA500 chemical adsorption pad equipped by the branch, and the degree of harm and diffusion risk of the accident scene. The finally determined scheme can be: the first branch of the city fire rescue dispatches 3 groups of personnel carrying 2 sets of EX580 detectors and 50 pieces of CA500 adsorption pads to implement emergency disposal, the estimated disposal window is 08:00-10:00, the task completion efficiency reaches 92.3%, and the coordination cost is controlled within the acceptable range. In order to reflect the high dynamic predictive task agile adjustment, for example, when the system monitors that part of the rescue personnel needs to evacuate due to the excessive concentration of dangerous gas in the scene, resulting in insufficient disposal power, the system queries the standby resources through the knowledge graph, and combines the multi-objective optimization algorithm (NSGA-II) to predict the development trend of the accident, and quickly adjusts the allocation scheme to: dispatch 2 groups of personnel and 30 pieces of CA500 adsorption pads from the second branch of the city fire rescue to assist, adjust the disposal window to 08:30-10:30, and maintain the task completion efficiency at 90.8%, ensuring the disposal effect and response timeliness.

[0048] The above task allocation results, including the scheme after high dynamic predictive task agile adjustment, will be stored in a structured format to drive the subsequent emergency resource scheduling matrix generation and task execution module, realizing intelligent and high-robustness emergency decision support driven by the knowledge graph.

[0049] Step S06: Store the information of the entities of the emergency instruction, the relationships between the entities, the disposal information of the emergency entities, and the task allocation scheme, and generate an emergency plan in combination with related task background and emergency principles.

[0050] Specifically, after completing the entity recognition of the emergency instruction, the emergency relationship extraction between the entities, the emergency resource information query, and the task allocation scheme optimization, the system unifies the above information into a structured format to form a standard plan information template. The template includes the following key element fields: Event element information: including the core entities identified in the instruction text, such as rescue units, disaster targets, disposal behaviors, and task execution times; Entity relationship information: such as the emergency triple in the form of "city fire rescue branch → disposal → B industrial park hazardous chemical leakage"; Resource capacity information: including the equipment type, professional equipment model and available quantity equipped by the rescue team, risk factors, key facilities and protection needs of the disaster area, etc.; Task allocation scheme: including the specific correspondence between task and rescue power, the matching situation of equipment type and quantity, the disposal time window and the estimated completion time, etc. optimization results.

[0051] The above information is uniformly stored in the preplan database and organized in a structured format (such as JSON, XML or RDF) to form a complete preplan metadata set. On this basis, the system further introduces related task background knowledge (such as the disaster level of this event, the regional impact range, the linkage response unit, etc.) and emergency disposal principles (such as personnel safety priority, key facility protection priority, environmental pollution control priority, etc.), combines rule templates or expert system modules for semantic reorganization and reasoning, and finally outputs an emergency preplan document with complete structure and execution guidance significance.

[0052] The generated emergency preplan includes rescue team execution sequence, disposal node and path, time scheduling information, emergency resource calling plan and possible alternative solutions, has readability, analyzability and deliverability, and can be directly provided for superior emergency command system scheduling execution or further man-machine joint review.

[0053] To sum up, the present application provides a public emergency preplan intelligent generation and dynamic adjustment method aiming at the problems of scattered information sources, difficult structured data processing, low instruction analysis efficiency and low entity relationship extraction accuracy in the existing emergency preplan generation process. The method specifically comprises the following steps: first, acquiring emergency related information from literature, database, sensor and public information channels through a multi-source data acquisition interface, and constructing an emergency knowledge graph in a unified format, and using a Neo4j graph database for structured modeling and efficient storage; then, based on a BiLSTM-CRF and random forest fusion model, a named entity recognition model with integrated learning ability is constructed to accurately label key entities such as disaster points, rescue teams, equipment, locations, etc. in emergency instruction texts; then, combined with the dependency syntax analysis technology of spaCy and the rule matching mechanism, emergency disposal triples in the instructions are extracted, and the accuracy of relationship extraction is improved through a knowledge graph disambiguation module; on this basis, based on the entity relationships in the knowledge graph and the emergency resource constraint conditions, a multi-objective optimization model of task-resource matching is constructed, and combined with evolutionary algorithms such as NSGA-II, intelligent allocation and resource scheduling of emergency tasks are realized, taking into account multiple indicators such as task completion rate, resource utilization rate and risk control; finally, according to the task allocation results and the current disaster situation, a preplan set covering multiple emergency situations is automatically generated, and under the disturbance of resource state changes, task additions, etc., the agile adjustment and real-time reconstruction of the preplan are realized based on the dynamic updating mechanism of the graph, providing fast response support for emergency decision-making.

[0054] Referring to ​ , a structure schematic diagram of a public emergency preplan intelligent generation and dynamic adjustment system in an embodiment of the present application is shown, and the system comprises: Data acquisition module 10: for collecting external data related to emergencies, including disaster environment information, disaster situation, resource status, historical emergency instructions, etc., to provide basic data support for subsequent graph construction and task planning. Graph construction / update module 20: for constructing or updating the emergency knowledge graph according to the collected emergency data, combining a unified emergency ontology to standardize modeling of emergency entities and their attributes and relationships, ensuring the integrity and dynamic adaptability of the graph structure. Named entity recognition module 30: for constructing a named entity recognition model based on a bidirectional long short-term memory network BiLSTM model, a conditional random field CRF model, and a random forest Random Forest model, processing input emergency instruction text, and identifying and extracting emergency entities, including rescue teams, emergency equipment, disaster areas, and other elements. Relationship extraction module 40: for constructing a relationship extraction model based on the spaCy model and a pre-set relationship library, extracting entity relationships in emergency instructions, and obtaining important emergency logical relationships such as "jurisdictional relationship", "dispatch relationship", and "support relationship". Entity query module 50: for constructing a query model based on the Cypher language, querying emergency entity information and background disposal materials related to the current emergency instruction in the constructed emergency knowledge graph, and supporting multi-dimensional and multi-condition emergency data calling. Task allocation module 60: for constructing a task allocation model based on the multi-objective optimization algorithm NSGA-II, inputting emergency resources, target relationships, and other information queried from the knowledge graph into the model, combining a high-dynamic predictive task agile adjustment mechanism, and generating a non-dominated task allocation scheme that meets the efficiency, coverage rate, and resource constraints. This mechanism dynamically optimizes task allocation by predicting disaster situation changes to quickly respond to sudden changes in resource status or disposal demand, ensuring the adaptability and timeliness of the scheme. Plan generation module 70: for integrating the identified entities, extracted relationships, queried disposal information, and optimized task allocation scheme, and automatically generating an emergency plan combining task background, emergency principles, and decision logic, supporting version management and flexible adjustment of the plan.

[0055] Through the collaborative work of the above modules, the system can realize full-process automated processing from emergency data acquisition, knowledge modeling, intelligent analysis, to plan generation and adjustment, effectively improving the intelligent command and rapid response capability in the emergency response environment.

[0056] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0057] The above-described embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but cannot be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for intelligent generation and dynamic adjustment of public emergency plans, characterized in that: The method comprises: Collect external data and build / update the emergency knowledge graph based on the collected data; Build an emergency instruction library and perform entity annotation, and build a named entity recognition model based on the bidirectional long short-term memory network BiLSTM model, conditional random field model CRF model and random forest model Random Forest. Input the emergency instruction text data into the named entity recognition model to obtain the emergency entities involved in the instruction; Building a relation extraction model based on the spaCy model and the relation library, and inputting the emergency instruction text into the relation extraction model to obtain the relationship between emergency entities; An entity information query model is constructed based on the Cypher language to query the emergency knowledge graph for relevant disposal information of the emergency entity involved in the instruction; A task allocation model is constructed based on the multi-objective optimization algorithm NSGA-II, and the information retrieved from the knowledge graph and the relationship between emergency entities are input into the task allocation model to obtain a non-dominated task allocation solution; Highly dynamic and predictive task agile adjustment, through real-time monitoring of disaster situations (such as secondary disasters, changes in meteorological conditions, and resource damage) and prediction model outputs, dynamically adjusts task allocation plans and resource scheduling strategies to generate optimized plans that adapt to future situations.

2. The method according to claim 1, wherein: The construction and updating of the emergency knowledge graph involves dynamically updating the rescue resources, mission information, command structure, and other content within the graph based on collected external data, such as the disaster environment, emergency resources, affected areas, and weather information. A version control mechanism is used to record the graph's evolution to support multi-tasking concurrency and long-term maintenance. The graph update mechanism ensures data timeliness through real-time data streams (such as sensor data and monitoring feedback), supporting highly dynamic and predictive adjustments.

3. The method according to claim 1, wherein: The named entity recognition model performs word segmentation and standardization preprocessing on the emergency instruction text, and combines the BiLSTM-CRF model with the Random Forest model to accurately identify and classify key entities such as rescue teams, disaster locations, time information, task requirements, and equipment types. Specifically, the implementation of the model The following steps are involved: First, in the word embedding stage, each word in the text is mapped to a corresponding low-dimensional vector representation, providing basic features for subsequent deep model processing. Next, a bidirectional long short-term memory (BiLSTM) network is used to perform contextual modeling on each word in the sequence. The forward and backward information flows jointly capture the semantic relationships between words in context, generating an intermediate representation that incorporates contextual features. Subsequently, a conditional random field (CRF) model is used to decode the feature sequence output by the bidirectional long short-term memory (BiLSTM) network. By modeling the dependencies between labels, CRF further optimizes the determination of entity boundaries and categories. The core of this stage is to maximize the conditional probability and select the optimal label sequence to improve overall sequence labeling accuracy. To improve recognition robustness and generalization, we introduced the Random Forest model, building on the previous work. This model fuses and optimizes the entity recognition results obtained from training multiple BiLSTM-CRF models. By integrating a voting mechanism across multiple decision trees, Random Forest makes more robust judgments about the entity category of each word. Finally, through the collaborative work of the BiLSTM-CRF model and the random forest model, key named entities in the emergency instruction text are accurately identified and classified, providing solid support for subsequent information extraction and intelligent reasoning.

4. The method according to claim 1, wherein: The relationship extraction model is constructed based on the spaCy framework and a predefined relationship library, and includes an instruction preprocessing module, a named entity recognition (NER) module, a relationship recognition module, and a knowledge graph disambiguation module, which is used to extract the semantic relationship between emergency entities in emergency instructions. Specifically, after the system receives the instruction text from the emergency command center, it first completes word segmentation and part-of-speech tagging through the instruction preprocessing module; then, the NER module uses the BiLSTM-CRF and random forest collaborative named entity recognition model described in claim 3 to identify key emergency entities in the text, including rescue organizations, disaster-stricken locations, task types, etc.; the relationship recognition module is based on spaCy's dependency syntactic analysis capabilities, combined with keyword matching rules, to extract the semantic relationship between entities from the subject-verb-object structure and construct an emergency triple <subject entity, disposal behavior, object entity>; in order to eliminate entity ambiguity, the system further calls the knowledge graph disambiguation module, combines the hierarchical structure and geographic information in the knowledge graph, unifies the entity reference, and ensures that the extracted entities accurately correspond to the nodes in the knowledge graph. Finally, the system outputs the above structured entity and relationship information in a structured format for emergency plan generation and resource scheduling module calls.

5. The method according to claim 1, wherein: The task allocation model uses the multi-objective optimization algorithm NSGA-II to calculate and optimize task allocation plans based on task requirements, resource constraints, and emergency response requirements, minimizing resource allocation costs and ensuring task completion quality. The model supports highly dynamic predictive adjustments, dynamically adjusting allocation plans by predicting changes in disaster situations (such as the occurrence of secondary disasters and resource depletion), generating non-dominated allocation plans that adapt to future situations.

6. The method according to any one of claims 1 to 5, characterized in that The method supports multi-department joint response, covering collaborative task planning in fields such as fire protection, medical care, and transportation. It can handle cross-departmental and cross-regional resource sharing needs. The generated emergency plans are stored in a structured format and support two-dimensional or three-dimensional visualization to assist emergency command decision-making.

7. The method according to claim 5, characterized in that: The multi-objective optimization algorithm of this method aims to maximize the task completion quality and minimize the adjustment cost, where the task completion quality is expressed by the formula Calculate z il is the satisfaction of the first type of resources, R i For the set of resource types required for the task, the cost is adjusted by the formula Calculation, c jnm is the cost of the jth platform transferring from task n to task m, t jnm To transfer matrix elements, the optimization process refers to the node information of emergency resources, organizational structures, and scope of application in the knowledge graph to ensure balanced resource allocation and meet the time and geographical constraints of the emergency response process.

8. A public emergency plan intelligent generation and dynamic adjustment system, characterized by: The system includes the following modules: Data acquisition module: used to collect external data, including disaster environment, emergency equipment, command instructions and other information, to provide basic data for the generation of emergency plans. Graph construction / update module: Build or update the emergency knowledge graph based on the collected external data, map rescue resources, mission information, command instructions and other data into the knowledge graph, and ensure the dynamic update and accuracy of the knowledge graph. Named entity recognition module: Based on the bidirectional long short-term memory network BiLSTM, conditional random field CRF and random forest model, a named entity recognition model is constructed to analyze the input emergency instruction text data and extract the involved emergency entities, such as rescue equipment, emergency facilities, rescue personnel, etc. Relationship Extraction Module: Based on the spaCy model and a relationship library containing multiple predefined emergency behavior grammar patterns, this module builds a relationship extraction model to extract relationships between emergency entities from emergency instruction texts, such as command and dispatch relationships, resource demand relationships, and task execution relationships. Entity query module: Build an entity information query model based on the Cypher language to query the emergency entities and their disposal information related to the instructions in the emergency knowledge graph, ensuring that the generated plan is based on the latest and most accurate resources and intelligence. Task allocation module: Based on the multi-objective optimization algorithm NSGA-II, a task allocation model is constructed. The information retrieved from the knowledge graph and the relationship between emergency entities are input into the model to generate a non-dominated task allocation plan to optimize resource allocation and task execution efficiency, and support real-time resource reallocation in highly dynamic environments. The emergency plan generation module automatically generates a complete emergency plan based on the generated task allocation plan, the relationships between emergency entities, disposal information, and other relevant task background information. This plan can be flexibly adjusted to respond to various emergencies during the disposal process, improving the scientific nature of emergency decision-making and responsiveness.

Images