Environmental emergency aid decision-making method and system based on artificial intelligence

Abstract

The invention discloses an environment emergency aid decision-making method and system based on artificial intelligence, and the method comprises the steps: obtaining a plurality of historical environment disaster event instances, and constructing the multi-modal event features of each historical environment disaster event; analyzing an event triggering rule of each historical environment disaster event and a corresponding disposal scheme triggering rule based on the multi-modal event features, and constructing a disaster environment knowledge graph; environmental condition sensing information is obtained, and environmental emergency event prediction is carried out in combination with the disaster environment knowledge graph; constructing an event situation prediction model, and predicting the situation change of the current environment emergency event; a disaster environment knowledge graph is utilized to obtain candidate disposal schemes of real-time environment emergency events in a target area, and a multi-target grey wolf optimization algorithm is introduced to optimize the disposal schemes to generate an optimal disposal scheme for environment emergency decision assistance, so that the real-time performance, reliability and global optimality of the environment emergency decision are improved; and rapid and accurate disposal of environmental emergency events is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental emergency assistance, and particularly to an environmental emergency assistance decision-making method and system based on artificial intelligence. Background Art

[0002] With the acceleration of the industrialization process and the increase in environmental risk factors, environmental pollution accidents are characterized by suddenness, rapid diffusion, and severe harm. Events such as hazardous chemical leaks, water pollution, and air pollution pose a serious threat to the ecological environment and public safety. Traditional environmental emergency management mainly relies on manual experience judgment and static emergency plans. In recent years, artificial intelligence technology has shown great potential in the fields of environmental monitoring, risk prediction, and decision-making optimization. Machine learning can improve the accuracy of pollutant diffusion prediction; multi-objective optimization algorithms can automatically generate emergency plans that take into account multiple indicators. However, existing methods still have defects such as insufficient data integration ability, low decision-making intelligence level, and insufficient multi-objective collaborative optimization ability.

[0003] Therefore, there is an urgent need for an environmental emergency assistance decision-making method that integrates multi-modal data analysis, dynamic knowledge reasoning, and multi-objective optimization to break through the above technical bottlenecks and improve the real-time, reliability, and global optimality of decision-making. Summary of the Invention

[0004] The present invention overcomes the defects of the prior art and provides an environmental emergency assistance decision-making method and system based on artificial intelligence.

[0005] To achieve the above object, the first aspect of the present invention provides an environmental emergency assistance decision-making method based on artificial intelligence, including: Obtain a number of historical environmental disaster event instances, analyze the early warning characteristics and diffusion evolution characteristics of each historical environmental disaster event, and construct multi-modal event characteristics of each historical environmental disaster event; Analyze the event trigger rules and corresponding disposal plan trigger rules of each historical environmental disaster event based on the multi-modal event characteristics of each historical environmental disaster event, and construct a disaster environment knowledge graph; Build an environmental monitoring sensor network in the target monitoring area to sense the regional environmental conditions and obtain environmental condition sensing information, and combine the disaster environment knowledge graph to predict environmental emergency events and obtain environmental emergency event prediction information; Construct an event situation prediction model based on a generative adversarial network, and combine the environmental condition sensing information and the environmental emergency event prediction information to predict the situation change of the current environmental emergency event and obtain event situation prediction information; Use the disaster environment knowledge graph to obtain candidate disposal plans for real-time environmental emergency events in the target area, and introduce a multi-objective grey wolf optimization algorithm to optimize the disposal plans to generate the optimal disposal plan for environmental emergency decision-making assistance.

[0006] In this solution, obtaining several historical environmental disaster event instances, analyzing the early warning characteristics and diffusion evolution characteristics of each historical environmental disaster event, and constructing the multi-modal event characteristics of each historical environmental disaster event specifically include: Obtaining several historical environmental disaster event instances through a big data network to form a historical environmental disaster event dataset, where the historical environmental disaster event dataset includes environmental disaster event characteristics and disposal plan characteristics when an environmental disaster occurs; Obtaining the historical environmental disaster event characteristics corresponding to each historical environmental disaster event through the historical environmental disaster event dataset, and extracting the historical disaster environment data and corresponding environmental benchmark data when the environmental disaster occurs through the historical environmental disaster event characteristics; Constructing a disaster environment data vector matrix using the historical disaster environment data, constructing an environmental benchmark data vector matrix using the environmental benchmark data, comparing the disaster environment data vector matrix with the environmental benchmark data vector matrix, and obtaining a residual vector matrix; Extracting the historical disaster environment diffusion characteristics when the environmental disaster occurs using the historical environmental disaster event characteristics and performing time series processing to obtain a disaster environment diffusion characteristic sequence, and inputting it into an LSTM network; Analyzing the diffusion evolution relationship of the disaster environment diffusion characteristic sequence through the forget gate and memory cell state update to obtain the diffusion evolution characteristics of each historical environmental disaster event; generating the early warning characteristics of each historical environmental disaster event based on the positive and negative values of each residual vector in the residual vector matrix; Performing data fusion based on the early warning characteristics, diffusion evolution characteristics, environmental disaster event characteristics, and disposal plan characteristics of each historical environmental disaster event to generate the multi-modal event characteristics of each historical environmental disaster event.

[0007] In this solution, analyzing the event trigger rules and corresponding disposal plan trigger rules of each historical environmental disaster event based on the multi-modal event characteristics of each historical environmental disaster event, and constructing a disaster environment knowledge graph specifically include: Obtaining the multi-modal event characteristics of each historical environmental disaster event, introducing the OPTICS algorithm to divide the scenario modes of each historical environmental disaster event, and defining each historical environmental disaster event as a single sample; Calculating the Mahalanobis distance between each sample, using the calculated Mahalanobis distance as the reachable distance and judging it with a preset reachable distance threshold, and defining the samples smaller than the preset reachable distance threshold as the neighborhood samples of the target sample to construct several sample neighborhoods; Gradually generate a clustering hierarchy through recursive operations until all samples are assigned to their corresponding sample neighborhoods, then output several clusters, calculate the silhouette coefficient of each cluster, select the clusters with a silhouette coefficient greater than the preset threshold to generate environmental disaster scenario patterns, and obtain the first dataset; Use the mRMR algorithm to screen the multi-modal event features corresponding to each environmental disaster scenario pattern in the first dataset, and use the screened multi-modal event features as the principal component direction for principal component projection to obtain a projection scatter plot. Based on the projection scatter plot, select the main disaster features corresponding to each environmental disaster scenario pattern to represent the trigger rules of the corresponding environmental scenario pattern, and obtain the trigger rules for environmental disaster scenario patterns; Extract the disposal plan features corresponding to each historical environmental disaster event through the multi-modal event features of each historical environmental disaster event. Define the disposal plan features as target attributes and the environmental disaster event features of each historical environmental disaster event as decision attributes, and analyze the trigger rules of each disposal measure in the corresponding disposal plan to obtain the trigger rules for the disposal plan; Based on the trigger rules for environmental disaster scenario patterns and the trigger rules for the disposal plan, combine the multi-modal event features of each historical environmental disaster event to construct a disaster environment knowledge graph.

[0008] In this solution, setting up an environmental monitoring sensor network in the target monitoring area to perceive the regional environmental conditions and obtain environmental condition perception information, and combining the disaster environment knowledge graph to predict environmental emergency events to obtain environmental emergency event prediction information specifically includes: Set up an environmental monitoring sensor network in the target monitoring area, and based on the set environmental monitoring sensor network, perceive the environmental conditions of the target area to obtain environmental condition perception information; Perform data preprocessing on the environmental condition perception information using a hierarchical cleaning strategy. Use the isolation forest algorithm to identify outliers by randomly dividing the feature space, and define the outliers with a sampling period greater than the preset period threshold as abnormal values; Eliminate the identified abnormal values, and use the spatio-temporal Kriging interpolation algorithm to analyze the spatial autocorrelation through the semi-variogram, and combine the time series sliding window to dynamically estimate the optimal value of the missing points to generate preprocessed environmental condition perception information; Extract the real-time environmental condition perception features of the target area based on the preprocessed environmental condition perception information, introduce an attention mechanism to map the real-time environmental condition perception features into a query vector, a key matrix, and a value matrix for attention score calculation, and perform feature fusion through the calculated attention analysis and construct an attention fusion feature map; Obtain a disaster environment knowledge graph, import the attention fusion feature map into the disaster environment knowledge graph for environmental emergency event prediction, and calculate and obtain the similarity value between the attention fusion feature map and the environmental disaster scenario pattern trigger rules in the disaster environment knowledge graph; Select the environmental disaster scenario pattern trigger rules with a similarity value greater than a preset similarity threshold based on the calculated similarity value, sort them based on the similarity, and generate environmental emergency event prediction information according to the environmental disaster scenario pattern with the highest similarity based on the sorting result.

[0009] In this solution, the event situation prediction model is constructed based on a generative adversarial network, and the situation change of the current environmental emergency event is predicted by combining the environmental condition perception information and the environmental emergency event prediction information to obtain the event situation prediction information, which specifically includes: Obtain a disaster environment knowledge graph, extract the diffusion evolution features and warning features corresponding to each historical environmental disaster event based on the disaster environment knowledge graph, perform temporal processing on the extracted diffusion evolution features to obtain a diffusion evolution feature sequence, and use the warning features to mark warning feature nodes in the diffusion evolution feature sequence; Introduce the Markov algorithm to construct a state space according to the diffusion evolution feature sequence, and use the state space to calculate the state transition probabilities of the environmental disaster states at different times in each historical environmental disaster event and construct a state transition matrix; Construct an event situation prediction model based on a generative adversarial network, use the diffusion evolution feature sequence to construct a training dataset to train the event situation prediction model, and use the constructed state transition matrix to form a constraint term of the generator through maximum likelihood estimation, and retain the model parameters after iterative training and output an event situation prediction model that meets the expectations; Obtain the environmental emergency event prediction information and the environmental condition perception information, input them into the trained event situation prediction model for analysis, and retrieve the state transition matrix that matches the current environmental state in the preset state space based on the input environmental emergency event prediction information; Generate an environmental condition perception sequence according to the environmental condition perception information and input it into the generator for sequence generation, and use the retrieved state transition matrix as a generation constraint condition to obtain a generated sequence; Input the generated sequence into the discriminator for judgment. If the discriminator does not accept the current generated sequence, obtain the next generated sequence for iterative discrimination to obtain the final generated sequence, and form the event situation prediction information of the current environmental condition based on the final generated sequence.

[0010] In this solution, the candidate disposal plans for real-time environmental emergency events in the target area are obtained by using the disaster environment knowledge graph, and a multi-objective grey wolf optimization algorithm is introduced to optimize the disposal plans to generate the optimal disposal plan for environmental emergency decision-making assistance. Specifically, it includes: Obtain event situation prediction information and environmental condition perception information, extract the change trend characteristics and early warning characteristics of real-time environmental emergency events in the target area based on the event situation prediction information, and define the severity of real-time environmental emergency events in the target area by using the extracted early warning characteristics; Obtain the disaster environment knowledge graph, construct a feature portrait of real-time environmental emergency events in the target area through the event situation prediction information and environmental condition perception information, and import it into the disaster environment knowledge graph for disposal plan analysis; Select candidate disposal plans by calculating the Euclidean distance value between the disposal plan trigger rules corresponding to each disposal plan stored in the disaster environment knowledge graph and the feature portrait of real-time environmental emergency events in the target area, and obtain several candidate disposal plans; Introduce a multi-objective grey wolf optimization algorithm to optimize the disposal plan, generate a search space according to the obtained several candidate disposal plans, and randomly generate an initial grey wolf population through the search space; Preset the objective function and set the constraint conditions, calculate the objective function value of each individual in the initial grey wolf population through the objective function, and divide all individuals in the population into different front levels through the calculated objective function value for non-dominated sorting; Perform congestion calculation on each front level to obtain the crowding distance, select the individual with the largest crowding degree in each level as the leader wolf, regard the remaining individuals as follower wolves, and calculate the direction vector of the leader wolf in the decision space for position update; Output the optimal solution set after repeated iterative optimization until the stop condition is met, generate the optimal disposal plan for real-time environmental emergency events in the target area according to the optimal solution set, and conduct environmental emergency decision-making assistance.

[0011] The second aspect of the present invention provides an environmental emergency auxiliary decision-making system based on artificial intelligence. The system includes: a memory and a processor. The memory contains a program of the environmental emergency auxiliary decision-making method based on artificial intelligence. When the program of the environmental emergency auxiliary decision-making method based on artificial intelligence is executed by the processor, the following steps are implemented: Obtain several historical environmental disaster event instances, analyze the early warning characteristics and diffusion and evolution characteristics of each historical environmental disaster event, and construct multi-modal event characteristics of each historical environmental disaster event; Analyze the event trigger rules and corresponding disposal plan trigger rules of each historical environmental disaster event based on the multi-modal event characteristics of each historical environmental disaster event, and construct a disaster environment knowledge graph; Build an environmental monitoring sensor network in the target monitoring area to perceive the regional environmental conditions, obtain environmental condition perception information, and combine with the disaster environment knowledge graph to predict environmental emergency events, so as to obtain environmental emergency event prediction information; Construct an event situation prediction model based on the generative adversarial network, combine the environmental condition perception information and the environmental emergency event prediction information to predict the situation change of the current environmental emergency event, and obtain the event situation prediction information; Use the disaster environment knowledge graph to obtain candidate disposal plans for real-time environmental emergency events in the target area, and introduce a multi-objective grey wolf optimization algorithm to optimize the disposal plans to generate the optimal disposal plan for environmental emergency decision-making assistance.

[0012] The present invention discloses an artificial intelligence-based environmental emergency auxiliary decision-making method and system, including: obtaining a number of historical environmental disaster event instances, constructing multi-modal event features of each historical environmental disaster event; analyzing the event trigger rules and corresponding disposal plan trigger rules of each historical environmental disaster event based on the multi-modal event features, and constructing a disaster environment knowledge graph; obtaining environmental condition perception information, and combining with the disaster environment knowledge graph to predict environmental emergency events; constructing an event situation prediction model to predict the situation change of the current environmental emergency event; using the disaster environment knowledge graph to obtain candidate disposal plans for real-time environmental emergency events in the target area, and introducing a multi-objective grey wolf optimization algorithm to optimize the disposal plans to generate the optimal disposal plan for environmental emergency decision-making assistance, improving the timeliness, reliability and global optimality of environmental emergency decision-making, and realizing the rapid and accurate disposal of environmental emergency events. Description of the Drawings

[0013] In order to more clearly illustrate the technical solutions in the embodiments or exemplary examples of the present invention, the following will briefly introduce the drawings required for use in the embodiments or exemplary descriptions. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the drawings shown without creative efforts.

[0014] Figure 1 It is a flowchart of an artificial intelligence-based environmental emergency auxiliary decision-making method provided by an embodiment of the present invention; Figure 2 It is a flowchart of an environmental emergency disposal plan optimization method provided by an embodiment of the present invention; Figure 3 It is a block diagram of an artificial intelligence-based environmental emergency auxiliary decision-making system provided by an embodiment of the present invention; The realization, functional characteristics and advantages of the purpose of the present invention will be further described in conjunction with the embodiments with reference to the drawings. Detailed Embodiments

[0015] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0016] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0017] Figure 1 A flowchart of an environment emergency auxiliary decision-making method based on artificial intelligence provided for an embodiment of the present invention; As Figure 1 shown, the present invention provides a flowchart of an environment emergency auxiliary decision-making method based on artificial intelligence, including: S102, obtaining a number of historical environmental disaster event instances, analyzing the early warning characteristics and diffusion evolution characteristics of each historical environmental disaster event, and constructing multi-modal event characteristics of each historical environmental disaster event; S104, analyzing the event trigger rules and corresponding disposal plan trigger rules of each historical environmental disaster event based on the multi-modal event characteristics of each historical environmental disaster event, and constructing a disaster environment knowledge graph; S106, building an environmental monitoring sensor network in the target monitoring area to sense the regional environmental conditions to obtain environmental condition sensing information, and combining the disaster environment knowledge graph to predict environmental emergency events to obtain environmental emergency event prediction information; S108, constructing an event situation prediction model based on a generative adversarial network, combining environmental condition sensing information and environmental emergency event prediction information to predict the situation change of the current environmental emergency event to obtain event situation prediction information; S110, using the disaster environment knowledge graph to obtain candidate disposal plans for real-time environmental emergency events in the target area, and introducing a multi-objective grey wolf optimization algorithm to optimize the disposal plans to generate optimal disposal plans for environmental emergency decision-making assistance.

[0018] Further, in a preferred embodiment of the present invention, the obtaining a number of historical environmental disaster event instances, analyzing the early warning characteristics and diffusion evolution characteristics of each historical environmental disaster event, and constructing multi-modal event characteristics of each historical environmental disaster event specifically includes: Obtaining a number of historical environmental disaster event instances through a big data network to form a historical environmental disaster event data set, where the historical environmental disaster event data set includes environmental disaster event characteristics and disposal plan characteristics when environmental disasters occur; Obtain the historical environmental disaster event characteristics corresponding to each historical environmental disaster event from the historical environmental disaster event dataset, and extract the historical disaster environment data and the corresponding environmental baseline data when the environmental disaster occurs through the historical environmental disaster event characteristics; Construct a disaster environment data vector matrix using the historical disaster environment data, construct an environmental baseline data vector matrix using the environmental baseline data, compare the disaster environment data vector matrix with the environmental baseline data vector matrix, and obtain a residual vector matrix; Extract the historical disaster environment diffusion characteristics when the environmental disaster occurs using the historical environmental disaster event characteristics, perform time series processing to obtain a disaster environment diffusion characteristic sequence, and input it into the LSTM network; Analyze the diffusion evolution relationship of the disaster environment diffusion characteristic sequence through the forget gate and the memory cell state update to obtain the diffusion evolution characteristics of each historical environmental disaster event; generate the early warning characteristics of each historical environmental disaster event based on the positive and negative values of each residual vector in the residual vector matrix; Perform data fusion based on the early warning characteristics, diffusion evolution characteristics, environmental disaster event characteristics, and disposal plan characteristics of each historical environmental disaster event to generate the multi-modal event characteristics of each historical environmental disaster event.

[0019] It should be noted that first, historical environmental disaster event instances are crawled based on the big data network to form a structured historical dataset, where each event instance contains dynamic environmental parameters (such as pollutant concentration, meteorological conditions, geospatial information) when the disaster occurs and the corresponding disposal plan records (such as the activated emergency response level, resource scheduling path, implementation measures). These raw data are parsed by the feature extraction module to separate the disaster event characteristics (such as the type of leakage source, diffusion medium) and the disposal plan characteristics (such as response time, material consumption), and then the real-time monitoring data when the disaster occurs and the environmental baseline data are vectorized and encoded to construct a disaster environment data vector matrix and an environmental baseline data vector matrix. During the matrix comparison process, an element-by-element difference operation is used to generate a residual vector matrix, which not only quantifies the deviation degree of the real-time environmental state from the safety baseline (for example, the residual of the VOCs concentration in a certain area is +15 ppm indicating exceeding the standard), but also dynamically triggers a graded early warning signal through the residual symbol (positive / negative) and the amplitude size (for example, when the residual is continuously positive and increasing, a red early warning is activated).

[0020] It should be noted that, in view of the spatio-temporal dynamic characteristics of the disaster diffusion process, environmental diffusion features (such as pollutant diffusion rate, direction deviation angle, concentration attenuation curve) are extracted from historical data, and are subjected to temporal alignment and normalization processing to form a standardized time series and input it into the LSTM network. Through its unique gating mechanism and memory cell state update, LSTM captures the long-term dependence relationship and non-linear pattern in the diffusion features. For example, it identifies the evolution law that "under the condition of a southeast wind of level 3, the diffusion radius of benzene series compounds increases by 200 meters every 10 minutes", so as to output a diffusion evolution feature vector with spatio-temporal coherence. Subsequently, warning features (such as multi-region synchronous exceeding the standard, residual gradient mutation), diffusion evolution features, basic event features (disaster type, occurrence time period), and disposal plan features (resource call priority, measure combination) are fused through a cross-modal attention mechanism to finally generate a unified multi-modal event feature vector. This feature vector not only contains numerical indicators quantifying the environmental situation (diffusion range prediction, resource demand estimation), but also integrates the regularized knowledge of historical disposal experience (such as "in case of chlorine leakage, it is necessary to evacuate the residential areas downwind first").

[0021] Furthermore, in a preferred embodiment of the present invention, the multi-modal event features based on each historical environmental disaster event analyze the event trigger rules of each historical environmental disaster event and the corresponding disposal plan trigger rules, and construct a disaster environment knowledge graph, specifically including: Obtain the multi-modal event features of each historical environmental disaster event, introduce the OPTICS algorithm to divide the scene patterns of each historical environmental disaster event, and define each historical environmental disaster event as a single sample; Calculate the Mahalanobis distance between each sample, use the calculated Mahalanobis distance as the reachable distance and judge it with a preset reachable distance threshold, and define the samples smaller than the preset reachable distance threshold as the neighborhood samples of the target sample to construct several sample neighborhoods; Gradually generate a clustering hierarchy through recursive operations until all samples are assigned to the corresponding sample neighborhoods, then output several clusters and calculate the silhouette coefficient of each cluster, and select the clusters with a silhouette coefficient greater than the preset silhouette coefficient threshold to generate environmental disaster scene patterns, obtaining a first data set; Through the mRMR algorithm, feature screening is performed on the multi-modal event features corresponding to each environmental disaster scene pattern in the first data set, and the screened multi-modal event features are used as the main component direction for principal component projection to obtain a projection scatter plot. Based on the projection scatter plot, the main disaster features corresponding to each environmental disaster scene pattern are selected to represent the trigger rules of the corresponding environmental scene pattern, obtaining the trigger rules of the environmental disaster scene pattern; Extract the disposal plan features corresponding to each historical environmental disaster event through the multi-modal event features of each historical environmental disaster event. Define the disposal plan features as target attributes, and the environmental disaster event features of each historical environmental disaster event as decision attributes. Analyze the triggering rules of each disposal measure in the corresponding disposal plan to obtain the disposal plan triggering rules. Based on the environmental disaster scenario pattern triggering rules and the disposal plan triggering rules, construct a disaster environment knowledge graph by combining the multi-modal event features of each historical environmental disaster event.

[0022] It should be noted that first, the multi-modal features of historical disaster events are divided into scenario patterns based on the OPTICS (Ordering Points To Identify the Clustering Structure) algorithm: Each event is regarded as an independent sample, and the Mahalanobis distance between samples is calculated to overcome the influence of feature dimension differences and correlation interference. For example, in a multi-dimensional feature space such as pollutant concentration, diffusion rate, and economic cost, the Mahalanobis distance eliminates the coupling effect between features through the inverse transformation of the covariance matrix, and more accurately measures the similarity of events. After setting the reachable distance threshold, the algorithm recursively traverses the sample space, and samples with distances lower than the threshold are classified into the same neighborhood. For example, a chemical leakage event is classified into the "rapid horizontal diffusion" neighborhood due to the feature combination of "wind speed > 5m / s + limited vertical diffusion", while another similar event is classified into the "vertical penetration dominant" neighborhood due to the features of "low wind speed + high temperature evaporation", forming clusters reflecting different diffusion mechanisms. High-quality clusters are selected by evaluating the cohesion and separation degree of the clusters using the silhouette coefficient to form the first data set. Subsequently, the mRMR (Maximal Relevance Minimal Redundancy) algorithm is used to screen the multi-modal features of each cluster: Calculate the mutual information between each feature and the scenario pattern to measure the correlation, and at the same time evaluate the redundancy between features, and select a feature subset that is highly correlated with the scenario pattern and has strong independence from each other (such as "VOCs concentration gradient" and "emergency response delay time" are retained, and "average humidity" is excluded due to redundancy). Based on the screened features, principal component analysis (PCA) is performed, projected into a low-dimensional space to generate a scatter plot, and the feature combination rules in the direction of the principal component are identified, and the key triggering rules are extracted from them. For example, the triggering rule of a certain leakage scenario pattern is represented by leakage type, leakage degree, and leakage speed, so as to obtain the explicit representation characteristics of the scenario pattern corresponding to the historical environmental disaster event.

[0023] It should be noted that for the mining of the disposal plan rules, the disposal plan features (such as "evacuation range", "amount of neutralizer used") are used as the target attributes, and the environmental features are used as the decision attributes. The condition-conclusion mapping is analyzed through the decision tree rule induction algorithm or the association rule mining algorithm. For example, in the scenario of "chlorine leakage", if "concentration residual > 10 ppm + diffusion direction contains residential areas", it corresponds to the disposal rule of "evacuation radius ≥ 3 km + call the chemical defense force", and the effectiveness of the rule is verified through the confidence and support indicators. Finally, the environmental scenario trigger rules and the disposal plan rules are integrated to construct a knowledge graph with disaster type - environmental conditions - disposal means as nodes, and the edge relationships include attributes such as trigger conditions and execution prerequisites. For example, the triple relationship of "oil spill → (needs to meet) sea current speed > 1 m / s → (trigger) deployment of offshore fences + spraying of oil absorbents" in the graph realizes the semantic reasoning from the environmental situation to the disposal strategy, providing an interpretable rule engine support for real-time emergency decision-making.

[0024] Further, in a preferred embodiment of the present invention, setting up an environmental monitoring sensor network in the target monitoring area to sense the regional environmental conditions and obtain environmental condition perception information, and combining with the disaster environment knowledge graph to predict environmental emergency events, obtaining environmental emergency event prediction information, specifically including: Set up an environmental monitoring sensor network in the target monitoring area, and based on the set environmental monitoring sensor network, sense the environmental conditions of the target area to obtain environmental condition perception information; Adopt a hierarchical cleaning strategy to preprocess the environmental condition perception information, use the isolation forest algorithm to identify outliers by randomly dividing the feature space, and define the outliers with a sampling period greater than the preset period threshold as abnormal values; Eliminate the identified abnormal values, and use the spatio-temporal Kriging interpolation algorithm to analyze the spatial autocorrelation through the semi-variogram, and combine the time series sliding window to dynamically estimate the optimal value of the missing points to generate preprocessed environmental condition perception information; Extract the real-time environmental condition perception features of the target area based on the preprocessed environmental condition perception information, introduce the attention mechanism to map the real-time environmental condition perception features into query vectors, key matrices and value matrices for attention score calculation, and perform feature fusion through the calculated attention analysis and construct an attention fusion feature map; Obtain the disaster environment knowledge graph, import the attention fusion feature map into the disaster environment knowledge graph for environmental emergency event prediction, and calculate and obtain the similarity value between the attention fusion feature map and the environmental disaster scenario pattern trigger rules in the disaster environment knowledge graph; Select the environmental disaster scenario pattern trigger rules greater than the preset similarity threshold based on the calculated similarity values, sort them based on the similarity, and generate environmental emergency event prediction information according to the environmental disaster scenario pattern with the highest similarity based on the sorting result.

[0025] It should be noted that multi-type environmental monitoring sensor networks are deployed in the target area (such as chemical industrial parks, river basins), including gas sensors (detecting VOCs, SO 2 etc.), water quality sensors (monitoring pH value, heavy metal ion concentration) and weather stations (collecting wind speed, temperature and humidity data), forming a three-dimensional monitoring grid covering the target area. The sensor nodes upload real-time data every 5 minutes to generate environmental condition perception information containing spatio-temporal tags (longitude, latitude, timestamp). For the noise and missing problems existing in the original data, a hierarchical cleaning strategy is adopted: using the Isolation Forest algorithm to construct a random decision tree forest, identifying outliers (such as instantaneous zero values or saturation values caused by sensor failures) by calculating the path length required for data points to be isolated, and removing the outliers that last for more than 3 sampling periods; for local data missing caused by equipment offline or signal interference, the spatio-temporal Kriging interpolation algorithm is used to model the spatial autocorrelation through the semi-variogram and dynamically calculate the optimal estimate of the missing points in combination with the time sliding window to ensure that the repaired data conforms to the spatio-temporal continuity law. Input the preprocessed environmental condition perception information into the feature extraction module, which can be constructed by a Convolutional Neural Network (CNN) and a Long Short-Term Memory Network (LSTM). To further strengthen the representation ability of key features, a multi-head attention mechanism is introduced: mapping the real-time feature vector into a query vector, a key matrix and a value matrix, calculating the dot product attention score between the query vector and the key matrix, dynamically aggregating highly relevant features, and generating an attention fusion feature map. Subsequently, call the pre-constructed disaster environment knowledge graph, and perform similarity matching between the attention fusion feature map and the scenario pattern trigger rules in the knowledge graph. The cosine similarity algorithm is used to calculate the matching degree between the feature encoding and the rule vector (such as the diffusion rate threshold and wind direction sensitivity coefficient corresponding to the "chlorine leakage pattern"), and a threshold of 0.75 is set to screen high-confidence candidate rules, and a priority list is generated in descending order of similarity. Finally, select the scenario pattern with the highest similarity as the prediction result and output the environmental emergency event prediction information.

[0026] Furthermore, in a preferred embodiment of the present invention, the event situation prediction model is constructed based on a generative adversarial network, and the situation change of the current environmental emergency event is predicted by combining the environmental condition perception information and the environmental emergency event prediction information to obtain the event situation prediction information, specifically including: Obtain a disaster environment knowledge graph, extract the diffusion evolution characteristics and early warning characteristics corresponding to each historical environmental disaster event based on the disaster environment knowledge graph, perform temporal processing on the extracted diffusion evolution characteristics to obtain a diffusion evolution characteristic sequence, and use the early warning characteristics to mark early warning characteristic nodes in the diffusion evolution characteristic sequence; Introduce the Markov algorithm to construct a state space according to the diffusion evolution characteristic sequence, calculate the state transition probability of the environmental disaster state at different times in each historical environmental disaster event using the state space, and construct a state transition matrix; Construct an event situation prediction model based on a generative adversarial network, use the diffusion evolution characteristic sequence to construct a training dataset to train the event situation prediction model, and constitute a constraint term for the generator through maximum likelihood estimation according to the constructed state transition matrix. After iterative training, retain the model parameters and output an event situation prediction model that meets the expectations; Obtain environmental emergency event prediction information and environmental condition perception information, input them into the trained event situation prediction model for analysis, and retrieve in a preset state space based on the input environmental emergency event prediction information to obtain a state transition matrix that matches the current environmental state; Generate an environmental condition perception sequence according to the environmental condition perception information and input it into the generator for sequence generation, and use the retrieved state transition matrix as a generation constraint condition to obtain a generated sequence; Input the generated sequence into the discriminator for judgment. If the discriminator does not accept the current generated sequence, obtain the next generated sequence for iterative discrimination to obtain the final generated sequence, and constitute the event situation prediction information of the current environmental condition based on the final generated sequence.

[0027] It should be noted that first, extract the historical event diffusion evolution characteristics and early warning characteristics (such as the moment when the exceedance threshold is triggered, the emergency response start node) from the pre-constructed disaster environment knowledge graph, align the diffusion characteristics along the time axis and standardize them into a time series sequence of a fixed length. At the same time, use the early warning characteristics as labels to embed the corresponding timestamps to form an annotated diffusion evolution characteristic sequence. Based on this sequence, use K-means clustering to discretize the continuous characteristics into a finite state space, count the transition frequencies between states in historical events, and construct a state transition probability matrix to describe the statistical law of disaster evolution.

[0028] It is worth mentioning that a generative adversarial network is used to construct an event situation prediction model. The generator adopts a sequence generation architecture with gated recurrent units (GRUs). The input layer receives a real-time environmental condition perception sequence (such as pollutant concentration and wind speed data within the current 3 hours). The hidden layer state is associated with the historical diffusion evolution feature sequence through an attention mechanism, and a Markov constraint is superimposed on the output layer: the KL divergence between the transition probability of adjacent state pairs in the generated sequence and the preset state transition matrix is calculated and added to the generator loss function as a regularization term. The discriminator is designed as a spatio-temporal convolutional network (STCN) to capture the temporal continuity and spatial correlation of the generated sequence and output a sequence authenticity score. During the training process, a curriculum learning strategy is adopted. Initially, only the Markov constraint is used to guide the generator to learn the basic state transition rules. Later, the intensity of adversarial training is gradually increased to force the generator to generate sequences with both statistical compliance and spatio-temporal rationality.

[0029] Furthermore, in the real-time prediction stage, the current environmental emergency event prediction information is mapped to a subset of the state transition matrix retrieved and matched in the state space, and an initial environmental condition sequence is generated in combination with the real-time perception data and input into the generator. The generator iteratively generates future multi-step state sequences under the constraint of the state transition matrix. Each generation result is input into the discriminator for confidence evaluation: if the discriminator score is lower than the threshold of 0.7, the hidden layer state weights of the generator are adjusted and regenerated until the score meets the standard or the maximum number of iterations is reached. The finally generated sequence is restored to specific environmental parameters (such as pollutant concentration and diffusion radius) through inverse state mapping, and the warning rules in the knowledge graph are superimposed to output event situation prediction information including time nodes and risk levels. For example, in a certain oil depot leakage event, it is predicted that the core pollution area will spread 1.2 kilometers to the southeast direction in the next 2 hours (confidence level 0.85), triggering a secondary response plan.

[0030] Figure 2 It is a flowchart of an environmental emergency response plan optimization method provided by an embodiment of the present invention; As Figure 2 shown, the present invention provides a flowchart of an environmental emergency response plan optimization method, including: S202, obtaining event situation prediction information and environmental condition perception information, extracting the change trend features and warning features of real-time environmental emergency events in the target area based on the event situation prediction information, and defining the severity of real-time environmental emergency events in the target area by using the extracted warning features; S204, obtaining a disaster environment knowledge graph, constructing a feature portrait of real-time environmental emergency events in the target area through the event situation prediction information and environmental condition perception information, and importing it into the disaster environment knowledge graph for disposal plan analysis; S206. Select candidate disposal plans by calculating the Euclidean distance values between the disposal plan trigger rules corresponding to each disposal plan stored in the disaster environment knowledge graph and the feature portrait of the real-time environmental emergency event in the target area, and obtain a number of candidate disposal plans; S208. Introduce a multi-objective grey wolf optimization algorithm to optimize the disposal plan. Generate a search space based on the obtained number of candidate disposal plans, and randomly generate an initial grey wolf population through the search space; S210. Preset an objective function and set constraint conditions. Calculate the objective function values of each individual in the initial grey wolf population through the objective function, and divide all individuals in the population into different front levels through the calculated objective function values for non-dominated sorting; S212. Calculate the crowding distance for each front level to obtain the crowding degree distance. Select the individual with the largest crowding degree in each level as the leading wolf, regard the remaining individuals as following wolves, and calculate the direction vector of the leading wolf in the decision space for position update; S214. Repeatedly iterate and optimize until the stop condition is met, then output the optimal solution set. Generate the optimal disposal plan for the real-time environmental emergency event in the target area according to the optimal solution set, and assist in environmental emergency decision-making.

[0031] It should be noted that through the collaborative mechanism of event situation prediction and multi-objective optimization decision-making, the dynamic generation and optimization of environmental emergency plans are realized. First, based on event situation prediction information (such as the pollutant diffusion path and the concentration peak time point), extract the real-time change trend characteristics of the target area (such as the diffusion rate growth rate and the thermal radiation intensity change rate), and quantify the severity of the event in combination with the warning characteristics. At the same time, fuse the environmental condition perception information to construct the feature portrait of the target event, and form a multi-dimensional feature vector including "pollution type - diffusion mode - resource availability" (for example, the portrait of a benzene leakage event is: diffusion rate 2.5 km / h, with residential areas downwind). Import this feature portrait into the disaster environment knowledge graph, and retrieve associated nodes through the graph traversal algorithm to generate a set of candidate disposal plans. Calculate the Euclidean distance between the feature portrait and the disposal plan trigger rules in the knowledge graph: for each plan rule (such as "diffusion rate > 2 km / h and population density > 100 people / km² → evacuation radius ≥ 3 km"), align its conditional parameters (such as rate threshold and density threshold) with the real-time feature values, calculate the distance value in the multi-dimensional space, and screen out the plans with distance values less than the threshold as the candidate set to ensure that the plans not only conform to historical rules but also adapt to the real-time scenario.

[0032] Furthermore, in the multi-objective grey wolf optimization (MOGWO) stage, the decision variables of the candidate disposal plans (such as evacuation scope, material scheduling volume, response time window) are encoded as the position vectors of grey wolf individuals. An objective function including safety, timeliness, and economy is constructed, and hard constraints (such as material inventory limits, traffic accessibility time) are set. After initializing the grey wolf population, the population individuals are divided into multiple front levels through non-dominated sorting: individuals in the first level are not dominated by other individuals in all objectives (such as a plan achieving both the minimum casualties and the lowest cost), individuals in the second level are only dominated by the first level, and so on. The crowding distance (measuring the distribution density of solutions in the objective space) is calculated for each front level, and the individual with the largest crowding degree in each level is selected as the leading wolf, and the remaining individuals are used as following wolves.

[0033] During the grey wolf position update process, the direction vector of the following wolves moving towards the leading wolf is dynamically adjusted by the objective function weights. After each iteration, the population is combined with the external archive set for non-dominated sorting and crowding degree screening, and high-quality solutions are retained. When the maximum number of iterations (such as 100 times) is reached or convergence occurs (the optimal solutions do not change for 10 consecutive generations), the optimization stops, and the Pareto optimal solution set is output. Finally, combined with the decision-maker's preference (such as safety first), the TOPSIS method is used to select the plan with the highest comprehensive score for environmental emergency event decision-making assistance. Through the cascading mechanism of rule screening and multi-objective optimization, the scientific nature, real-time nature, and multi-dimensional balance of emergency decision-making are realized.

[0034] Figure 3 An environment emergency assistance decision-making system 3 based on artificial intelligence provided by an embodiment of the present invention includes: a memory 31 and a processor 32. The memory 31 contains a program of an environment emergency assistance decision-making method based on artificial intelligence. When the program of the environment emergency assistance decision-making method based on artificial intelligence is executed by the processor 32, the following steps are implemented: Obtain several historical environmental disaster event instances, analyze the early warning characteristics and diffusion and evolution characteristics of each historical environmental disaster event, and construct the multi-modal event characteristics of each historical environmental disaster event; Analyze the event trigger rules and corresponding disposal plan trigger rules of each historical environmental disaster event based on the multi-modal event characteristics of each historical environmental disaster event, and construct a disaster environment knowledge graph; Build an environmental monitoring sensor network in the target monitoring area to sense the regional environmental conditions and obtain environmental condition sensing information, and combine the disaster environment knowledge graph to predict environmental emergency events and obtain environmental emergency event prediction information; Construct an event situation prediction model based on a generative adversarial network, and combine the environmental condition sensing information and the environmental emergency event prediction information to predict the situation change of the current environmental emergency event and obtain event situation prediction information; The candidate disposal plans for real-time environmental emergency events in the target area are obtained by using the disaster environment knowledge graph, and a multi-objective grey wolf optimization algorithm is introduced to optimize the disposal plans to generate the optimal disposal plan for environmental emergency decision-making assistance.

[0035] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed with each other can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical, or other forms.

[0036] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0037] In addition, in each embodiment of the present invention, the various functional units can all be integrated in one processing unit, or each unit can be separately used as a unit, or two or more units can be integrated in one unit; the above-mentioned integrated units can be implemented in the form of hardware, or in the form of a combination of hardware and software functional units.

[0038] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media that can store program codes such as mobile storage devices, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs.

[0039] Alternatively, if the above-integrated units of the present invention are implemented in the form of software function modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as removable storage devices, ROM, RAM, magnetic disks, or optical discs.

[0040] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An artificial intelligence-based environmental emergency decision-making assistance method, characterized in that: include: Obtain several historical environmental disaster event examples, analyze the warning characteristics and diffusion evolution characteristics of each historical environmental disaster event, and construct the multimodal event characteristics of each historical environmental disaster event; Based on the multimodal event characteristics of each historical environmental disaster event, the event triggering rules of each historical environmental disaster event and the corresponding disposal plan triggering rules are analyzed, and a disaster environment knowledge graph is constructed; An environmental monitoring sensor network is built in the target monitoring area to sense the regional environmental conditions and obtain environmental condition perception information, and environmental emergency events are predicted in combination with the disaster environment knowledge graph to obtain environmental emergency event prediction information; An event situation prediction model is constructed based on a generative adversarial network, and the situation changes of the current environmental emergency events are predicted by combining environmental condition perception information and environmental emergency event prediction information to obtain event situation prediction information; The disaster environment knowledge graph is used to obtain candidate disposal plans for real-time environmental emergency events in the target area, and a multi-objective grey wolf optimization algorithm is introduced to optimize the disposal plans to generate the optimal disposal plan for environmental emergency decision-making assistance.

2. The method for assisting environmental emergency decision-making based on artificial intelligence according to claim 1 is characterized in that: The method of obtaining several historical environmental disaster event instances, analyzing the early warning characteristics and diffusion evolution characteristics of each historical environmental disaster event, and constructing multimodal event characteristics of each historical environmental disaster event specifically includes: Acquire several historical environmental disaster event instances through a big data network to form a historical environmental disaster event data set, wherein the historical environmental disaster event data set includes environmental disaster event characteristics and disposal plan characteristics when an environmental disaster occurs; Acquire historical environmental disaster event features corresponding to each historical environmental disaster event through the historical environmental disaster event data set, and extract historical disaster environmental data and corresponding environmental baseline data when the environmental disaster occurred through the historical environmental disaster event features; Constructing a disaster environment data vector matrix using historical disaster environment data, constructing an environmental benchmark data vector matrix using environmental benchmark data, and comparing the disaster environment data vector matrix with the environmental benchmark data vector matrix to obtain a residual vector matrix; The historical environmental disaster event characteristics are used to extract the historical disaster environment diffusion characteristics when the environmental disaster occurs, and the time series processing is performed to obtain the disaster environment diffusion feature sequence, which is input into the LSTM network; The diffusion evolution relationship of the disaster environment diffusion feature sequence is analyzed through the forget gate and memory cell state update to obtain the diffusion evolution characteristics of each historical environmental disaster event; the warning characteristics of each historical environmental disaster event are generated based on the positive and negative values ​​of each residual vector in the residual vector matrix; Based on the warning characteristics, diffusion evolution characteristics, environmental disaster event characteristics and disposal plan characteristics of each historical environmental disaster event, data fusion is performed to generate multimodal event characteristics of each historical environmental disaster event.

3. The method for assisting environmental emergency decision-making based on artificial intelligence according to claim 1 is characterized in that: The event triggering rules of each historical environmental disaster event and the corresponding disposal plan triggering rules are analyzed based on the multimodal event characteristics of each historical environmental disaster event, and a disaster environment knowledge graph is constructed, specifically including: Obtain the multimodal event features of each historical environmental disaster event, introduce the OPTICS algorithm to divide the scene mode of each historical environmental disaster event, and define each historical environmental disaster event as a single sample; Calculate the Mahalanobis distance between each sample, use the calculated Mahalanobis distance as the reachable distance and compare it with the preset reachable distance threshold, define the samples smaller than the preset reachable distance threshold as the neighborhood samples of the target sample to construct several sample neighborhoods; A clustering hierarchy is gradually generated through recursive operations until all samples are classified into corresponding sample neighborhoods, and then a number of clusters are output and the silhouette coefficients of each cluster are calculated. Clusters with a silhouette coefficient greater than a preset threshold are selected to generate environmental disaster scene patterns, thereby obtaining a first data set. The multimodal event features corresponding to each environmental disaster scene mode in the first data set are screened by the mRMR algorithm, and the screened multimodal event features are used as the principal component direction for principal component projection to obtain a projection scatter plot, and the main disaster features corresponding to each environmental disaster scene mode are selected based on the projection scatter plot to characterize the triggering rules of the corresponding environmental scene mode, so as to obtain the environmental disaster scene mode triggering rules; Through the multimodal event characteristics of each historical environmental disaster event, the disposal plan characteristics corresponding to each historical environmental disaster event are extracted, the disposal plan characteristics are defined as the target attribute, and the environmental disaster event characteristics of each historical environmental disaster event are used as the decision attribute. The triggering rules of each disposal method in the corresponding disposal plan are analyzed to obtain the disposal plan triggering rules; Based on the environmental disaster scenario pattern triggering rules and disposal plan triggering rules, a disaster environment knowledge graph is constructed by combining the multimodal event characteristics of various historical environmental disaster events.

4. The method for assisting environmental emergency decision-making based on artificial intelligence according to claim 1 is characterized in that: The environmental monitoring sensor network is built in the target monitoring area to sense the regional environmental conditions and obtain environmental condition perception information, and environmental emergency event prediction is performed in combination with the disaster environment knowledge graph to obtain environmental emergency event prediction information, specifically including: An environmental monitoring sensor network is set in the target monitoring area, and the environmental condition of the target area is sensed based on the set environmental monitoring sensor network to obtain environmental condition perception information; The environmental condition perception information is preprocessed using a hierarchical cleaning strategy, and outliers are identified by randomly dividing the feature space using an isolation forest algorithm, and outliers whose sampling period is greater than a preset period threshold are defined as outliers; The identified outliers are removed, and the spatial autocorrelation is analyzed by using the spatiotemporal Kriging interpolation algorithm through the semivariogram function. The optimal estimate of the missing points is dynamically estimated by combining the time series sliding window to generate pre-processed environmental condition perception information. Based on the preprocessed environmental condition perception information, the real-time environmental condition perception features of the target area are extracted, and the attention mechanism is introduced to map the real-time environmental condition perception features into query vectors, key matrices and value matrices for attention score calculation. The calculated attention analysis is used to perform feature fusion and construct an attention fusion feature map. Obtain a disaster environment knowledge graph, import the attention fusion feature graph into the disaster environment knowledge graph to predict environmental emergency events, and calculate and obtain a similarity value between the attention fusion feature graph and the environmental disaster scene pattern triggering rule in the disaster environment knowledge graph; Based on the calculated similarity value, the environmental disaster scenario pattern trigger rules that are greater than the preset similarity threshold are selected and sorted based on the similarity. The environmental emergency event prediction information is generated according to the sorting results based on the environmental disaster scenario pattern with the highest similarity.

5. The method for assisting environmental emergency decision-making based on artificial intelligence according to claim 1 is characterized in that: The event situation prediction model is constructed based on the generative adversarial network, and the situation change of the current environmental emergency event is predicted by combining the environmental condition perception information and the environmental emergency event prediction information to obtain the event situation prediction information, which specifically includes: Obtain a disaster environment knowledge graph, extract the diffusion evolution characteristics and early warning characteristics corresponding to each historical environmental disaster event based on the disaster environment knowledge graph, perform time series processing on the extracted diffusion evolution characteristics to obtain a diffusion evolution characteristic sequence, and use the early warning characteristics to mark early warning characteristic nodes in the diffusion evolution characteristic sequence; A Markov algorithm is introduced to construct a state space according to the diffusion evolution characteristic sequence, and the state space is used to calculate the state transition probability of the environmental disaster state at different times in each historical environmental disaster event and to construct a state transition matrix; An event situation prediction model is constructed based on a generative adversarial network, a training data set is constructed using the diffusion evolution feature sequence to train the event situation prediction model, constraints of the generator are constructed by maximum likelihood estimation according to the constructed state transfer matrix, model parameters are retained after iterative training, and an event situation prediction model that meets expectations is output; Obtain environmental emergency event prediction information and environmental condition perception information, input them into the trained event situation prediction model for analysis, and retrieve the state transfer matrix matching the current environmental state in the preset state space based on the input environmental emergency event prediction information; Generate an environmental condition perception sequence according to the environmental condition perception information and input it into a generator for sequence generation, and use the retrieved state transition matrix as a generation constraint to obtain a generation sequence; The generated sequence is input into the discriminator for judgment. If the discriminator does not accept the current generated sequence, the next generated sequence is obtained for iterative judgment to obtain the final generated sequence, and the event situation prediction information of the current environmental conditions is constructed based on the final generated sequence.

6. The method for assisting environmental emergency decision-making based on artificial intelligence according to claim 1 is characterized in that: The method of using the disaster environment knowledge graph to obtain candidate disposal plans for real-time environmental emergency events in the target area and introducing a multi-objective grey wolf optimization algorithm to optimize the disposal plans and generate the optimal disposal plan to assist in environmental emergency decision-making specifically includes: Acquire event situation prediction information and environmental condition perception information, extract change trend characteristics and warning characteristics of real-time environmental emergency events in the target area based on the event situation prediction information, and use the extracted warning characteristics to define the severity of real-time environmental emergency events in the target area; Obtain a disaster environment knowledge graph, construct a feature portrait of the real-time environmental emergency event in the target area through the event situation prediction information and environmental condition perception information, and import it into the disaster environment knowledge graph to analyze the disposal plan; The candidate disposal plans are selected by calculating the Euclidean distance between the disposal plan triggering rules corresponding to each disposal plan stored in the disaster environment knowledge graph and the characteristic portrait of the real-time environmental emergency event in the target area, and a number of candidate disposal plans are obtained; A multi-objective gray wolf optimization algorithm is introduced to optimize the disposal scheme, a search space is generated based on the obtained candidate disposal schemes, and the initial gray wolf population is obtained by random generation through the search space; The objective function is preset and the constraints are set. The objective function value of each individual in the initial gray wolf population is calculated through the objective function. All individuals in the population are divided into different frontier levels through non-dominated sorting based on the calculated objective function values. Congestion calculation is performed on each frontier level to obtain the crowding distance, the individual with the largest crowding degree in each level is selected as the leader wolf, the remaining individuals are used as follower wolves, and the direction vector of the leader wolf in the decision space is calculated to update its position; The optimal solution set is outputted through repeated iterative optimization until the stopping condition is met. The optimal disposal plan for real-time environmental emergency events in the target area is generated according to the optimal solution set to assist in environmental emergency decision-making.

7. An artificial intelligence-based environmental emergency decision-making support system, characterized in that: The system includes: a memory and a processor, wherein the memory contains an artificial intelligence-based environmental emergency decision-making assistance method program, and when the artificial intelligence-based environmental emergency decision-making assistance method program is executed by the processor, the following steps are implemented: Obtain several historical environmental disaster event examples, analyze the warning characteristics and diffusion evolution characteristics of each historical environmental disaster event, and construct the multimodal event characteristics of each historical environmental disaster event; Based on the multimodal event characteristics of each historical environmental disaster event, the event triggering rules of each historical environmental disaster event and the corresponding disposal plan triggering rules are analyzed, and a disaster environment knowledge graph is constructed; An environmental monitoring sensor network is built in the target monitoring area to sense the regional environmental conditions and obtain environmental condition perception information, and environmental emergency events are predicted in combination with the disaster environment knowledge graph to obtain environmental emergency event prediction information; An event situation prediction model is constructed based on a generative adversarial network, and the situation changes of the current environmental emergency events are predicted by combining environmental condition perception information and environmental emergency event prediction information to obtain event situation prediction information; The disaster environment knowledge graph is used to obtain candidate disposal plans for real-time environmental emergency events in the target area, and a multi-objective grey wolf optimization algorithm is introduced to optimize the disposal plans to generate the optimal disposal plan for environmental emergency decision-making assistance.

Images