Geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph

Abstract

This invention relates to a decision-making method for geological disaster emergency intelligent agents based on dynamic knowledge graphs. It acquires monitoring data, satellite remote sensing data, and network data to form a standardized data pool; performs cross-modal alignment of text and images based on feature information to unify the semantic vector space; applies multi-granularity attention enhancement to feature information and deeply integrates it with the entity embedding vectors of the geological disaster knowledge graph; acquires the semantic vector of the disaster situation and resource status, compares it with the knowledge graph, and outputs a discretized task type package; generates multiple decision schemes, calculates the actual benefits and predictive value of each scheme, and outputs the emergency intelligent agent decision scheme. By employing cross-modal alignment to integrate all data with the knowledge graph, a combined and shared whole can be applied to subsequent intelligent decision-making, breaking down information barriers; and by calculating the actual benefits and predictive value of multiple emergency decision schemes, the optimal decision scheme is selected, improving rescue efficiency.

Description

Technical Field

[0001] This invention relates to the field of geological disaster decision-making technology, and in particular to a geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph. Background Technology

[0002] Geological disasters

[0003] Current emergency command systems for geological disasters often suffer from inaccurate comprehensive assessments due to the suddenness and complexity of disasters. This is because various types of data are stored in separate private networks with heterogeneous storage formats, making it impossible to combine multiple data sets.

[0004] In addition, the golden time for geological disaster rescue is short. The process of consulting with multiple experts is cumbersome and time-consuming, and it is easy to miss the best rescue time. Without expert consultation and guidance from professional rescue knowledge, rescue operations may be difficult or the rescue results may be poor.

[0005] The above problems urgently need to be addressed. Summary of the Invention

[0006] This invention discloses a decision-making method for geological disaster emergency intelligent agents based on dynamic knowledge graphs, aiming to solve the technical problems existing in the prior art.

[0007] The present invention adopts the following technical solution: On one hand, this invention provides a decision-making method for geological disaster emergency intelligent agents based on dynamic knowledge graphs, comprising: acquiring monitoring data, satellite remote sensing data, and network data to form a standardized data pool, wherein the monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images; performing cross-modal alignment on the text and images in the standardized data pool based on feature information to unify the semantic vector space and form an initial database for text and image matching; performing multi-granularity attention enhancement on the feature information in the initial database and deeply fusing it with the entity embedding vector of the geological disaster knowledge graph to form a geological disaster intelligence database; acquiring the semantic vector and resource status of the current disaster situation, comparing it with the knowledge graph in the geological disaster intelligence database, and outputting a discretized task type package; generating multiple decision schemes based on the task type package, calculating the actual benefit and predicted value of each decision scheme, and outputting an emergency intelligent agent decision scheme, wherein the multiple decision schemes are used to instruct the sorting and refinement of the task type package into scheduling instructions.

[0008] Optionally, cross-modal alignment is performed on the text and images in the standardized data pool based on feature information to unify the semantic vector space and form an initial database for text and image matching. This includes: transforming the text and images in the standardized data pool into feature representations; calculating the weights of each granularity unit through parallel word-level, phrase-level, and sentence-level attention networks in the feature representations, wherein the granularity unit is used to indicate the basic units at the word, phrase, or sentence level from which the text is segmented; matching the text and images in the standardized data pool based on the weights of the granularity units, while simultaneously performing weighted fusion to generate multi-granularity feature vectors that focus on feature information; and forming the initial database using all the feature information, the multi-granularity feature vectors, and the matched text and images.

[0009] Optionally, based on the weights of the granular units, text and images in the standardized data pool are matched and weighted fusion is performed to generate a multi-granularity feature vector focusing on feature information. This includes: calculating the matching degree between each text granular unit and image features using cosine similarity; matching text and images in the standardized data pool to form matching data; multiplying the weights of the granular units by the matching degree and performing normalization processing; and fusing the normalized text and images by weighted summation to form a multi-granularity feature vector focusing on core information.

[0010] Optionally, multi-granularity attention enhancement is applied to the feature information existing in the initial database, and it is deeply fused with the entity embedding vectors of the geological disaster knowledge graph to form a geological disaster intelligence database. This includes: performing similarity matching between the multi-granularity feature vectors in the initial database and the vector representations of entity nodes in the geological disaster knowledge graph to form entity links; based on the entity links, performing reasoning on the geological disaster knowledge graph using the breadth-first search principle to determine the direct and indirect relationships between adjacent entities; and based on the direct and indirect relationships between adjacent entities, selectively aggregating the associated entities and their embedding vectors to form knowledge enhancement vectors containing domain semantics, thereby obtaining the geological disaster intelligence database.

[0011] Optionally, the multi-granularity feature vectors in the initial database are matched with the vector representations of entity nodes in the geological hazard knowledge graph to form entity links, including: in, The feature vector corresponding to the matched entity link includes text features extracted by multi-granularity attention and domain semantic features of the knowledge graph; This is a multi-granularity feature vector in the initial database; This is element-wise multiplication; It represents the vector of the entity node in the geological disaster knowledge graph.

[0012] Optionally, multiple decision schemes are generated based on the task type package, the actual benefit and predicted value of each decision scheme are calculated, and an emergency agent decision scheme is output. This includes: inputting the task type package into a fully connected neural network to form simulated instructions; calculating the actual delayed reward after task completion under a random combination of the simulated instructions; applying a coefficient reduction to the actual delayed reward to form a rescue benefit score after task completion; constructing a loss function based on the rescue benefit score and the actual delayed reward, wherein the loss function indicates the loss relationship between the actual value and the predicted value; and outputting the emergency agent decision scheme with the minimum loss function as the optimization objective.

[0013] Optionally, calculating the actual delay reward after task completion under the simulated instruction random combination method includes: Where R represents the actual delayed reward; The delayed reward for the completed instruction; The delay reward for instructions completed within the time limit; A delayed reward for achieving road clearance upon completion of the instruction.

[0014] Optionally, based on the rescue benefit score and the actual delay reward, a loss function is constructed, including: Where L is the loss function; R is the true delayed reward; and γ is the discount factor. Score the expected benefits before the rescue begins; Rate the rescue benefits after the mission is completed.

[0015] According to another aspect of the present invention, a geological disaster emergency intelligent agent decision-making device based on a dynamic knowledge graph is also provided, comprising: an acquisition module for acquiring monitoring data, satellite remote sensing data, and network data to form a standardized data pool, wherein the monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images; a cross-modal alignment module for performing cross-modal alignment on the text and images in the standardized data pool based on feature information, unifying the semantic vector space, and forming an initial database for text and image matching; a fusion module for performing multi-granularity attention enhancement on the feature information in the initial database and deeply fusing it with the entity embedding vector of the geological disaster knowledge graph to form a geological disaster intelligence database; a task package generation module for acquiring the semantic vector and resource status of the current disaster situation, comparing it with the knowledge graph in the geological disaster intelligence database, and outputting a discretized task type package; and a decision-making module for generating multiple decision schemes based on the task type packages, calculating the actual benefit and predicted value of each decision scheme, and outputting an emergency intelligent agent decision scheme, wherein the multiple decision schemes are used to instruct the sorting and refinement of the task type packages into scheduling instructions.

[0016] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph as described in any one of the embodiments.

[0017] The technical solution adopted in this invention can achieve at least one of the following beneficial effects: This invention collects monitoring data, satellite remote sensing data, and network data, and integrates all data with a knowledge graph using a cross-modal alignment approach to form a cohesive and shared whole for subsequent intelligent decision-making, breaking down information barriers. By generating discretized task type packages and calculating the actual benefits and predicted values ​​of various emergency decision-making schemes, the optimal decision scheme is selected, effectively improving the efficiency of decision scheme generation and providing guidance strategies for rescue personnel, thereby enhancing rescue efficiency. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below, forming part of the present invention. The illustrative embodiments of the present invention and their descriptions explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 This is a flowchart illustrating the overall steps of a geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graphs in Embodiment 1 of the present invention. Figure 2This is a detailed flowchart of the decision-making method for geological disaster emergency response based on dynamic knowledge graph in Embodiment 1 of the present invention; Figure 3 This is a flowchart of cross-modal semantic alignment for a geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph in Embodiment 1 of the present invention; Figure 4 This is a flowchart of multi-granularity intelligence extraction from a knowledge graph in a geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graphs, as described in Embodiment 1 of the present invention. Figure 5 This is a schematic diagram of the structure of a geological disaster emergency intelligent decision-making device based on dynamic knowledge graph in Embodiment 2 of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. In the description of this invention, it should be noted that the term "or" is generally used to include the meaning of "and / or," unless otherwise expressly indicated.

[0020] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or a magnetic connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Furthermore, in the description of this application, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. In the description of this invention, "a plurality of" means at least two, such as two, three, or more, unless otherwise explicitly specified.

[0021] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] First, to facilitate understanding of the embodiments of the present invention, some terms or nouns involved in the present invention will be explained below: A dynamic knowledge graph is a knowledge graph that supports the continuous change, real-time updating, and incremental evolution of entities, relationships, and attributes over time. It can reflect the dynamism and timeliness of knowledge in the time dimension, rather than a fixed and static knowledge base.

[0023] To address the problems existing in related technologies, this application provides a decision-making method for geological disaster emergency intelligent agents based on dynamic knowledge graphs.

[0024] Example 1 This embodiment provides a decision-making method for geological disaster emergency response based on dynamic knowledge graphs, such as... Figure 1 and Figure 2 As shown, Figure 1 This is a flowchart illustrating the overall steps of a geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graphs in Embodiment 1 of the present invention. Figure 2 This is a detailed flowchart of a decision-making method for geological disaster emergency response based on dynamic knowledge graphs in Embodiment 1 of the present invention. The method includes: Step S102: Acquire monitoring data, satellite remote sensing data, and network data to form a standardized data pool, wherein the monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images; Optionally, a static geological disaster knowledge graph can be formed by extracting basic instances from data sources such as historical cases, emergency plans, geographic information databases, and industry standards. Based on the access of multi-source dynamic data such as monitoring data, satellite remote sensing data, and network data (social media), dynamic entities are expanded. Combined with timed updates and event-triggered mechanisms, these dynamic entities are updated to the static geological disaster knowledge graph, achieving dynamic updates and ensuring coverage of the entire semantic chain of geological disasters from "induction to occurrence to derivation to treatment." A hybrid strategy of "pre-construction of basic instances + real-time dynamic data updates" is adopted to enable real-time iteration of the knowledge graph as the geological disaster situation changes.

[0025] Optionally, raw data from different sources and formats, such as monitoring data, satellite remote sensing data, drone patrol data, mass monitoring and prevention data, and public opinion texts or images, can be labeled with basic tags such as source and time to form a standardized data pool.

[0026] Specifically, the standardized data pool is as shown in the table below: Optionally, a dynamic geological disaster knowledge graph was constructed using a combination of pre-built basic instances and real-time dynamic data updates, enabling it to evolve in real time according to the disaster situation. Specifically, a cross-modal comparative learning model was used to fuse entity embeddings from the geological disaster knowledge graph. Entity vectors from the geological disaster knowledge graph were used as semantic anchors to guide heterogeneous data such as satellite imagery, sensor data, and text reports to be mapped into a single geological disaster knowledge graph during training, enhancing the unified semantic space of the geological disaster knowledge graph. This effectively solved the problem of data fusion difficulties caused by technical terminology barriers between different modalities, achieving precise alignment at the professional semantic level. This significantly improved the accuracy of cross-modal retrieval and matching (approximately 30%), laying a reliable foundation for subsequent intelligence fusion.

[0027] Step S104 involves performing cross-modal alignment of text and images within the standardized data pool based on feature information, unifying the semantic vector space, and forming an initial database for text-image matching; such as Figure 3 As shown, Figure 3 This is a flowchart of cross-modal semantic alignment for a geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph in Embodiment 1 of the present invention.

[0028] Optionally, a deep neural network model based on cross-modal contrastive learning can be used as a cross-modal semantic alignment model. The deep neural network model is pre-trained on a large number of disaster-related "text-image" datasets. The goal is to learn a unified semantic mapping space, so that the vector representations of semantically similar texts and images in this space are very close, while those that are semantically unrelated are far apart.

[0029] Optionally, cross-modal semantic alignment requires collecting massive amounts of text-image comparison data in the field of geological disasters (i.e., a standardized data pool). By using a pre-trained Transformer model and a convolutional neural network to extract features from the text and images respectively, text semantic vectors and image semantic vectors are formed. The loss function is used to perform comparative learning training on the "text-image" pair, mapping semantically related text and images to similar positions in a unified semantic space.

[0030] In some preferred embodiments, cross-modal alignment of text and images within a standardized data pool is performed based on feature information to unify the semantic vector space and form an initial database for text and image matching. This includes: transforming text and images within the standardized data pool into feature representations; calculating the weights of each granularity unit through parallel word-level, phrase-level, and sentence-level attention networks in the feature representations, where the granularity unit indicates the basic unit at the word, phrase, or sentence level from which the text is segmented; matching text and images within the standardized data pool based on the weights of the granularity units, while simultaneously performing weighted fusion to generate multi-granularity feature vectors that focus on feature information; and forming the initial database using all feature information, the multi-granularity feature vectors, and the matched text and images.

[0031] Optionally, by performing knowledge-enhanced deep semantic understanding on text and images in a standardized data pool, and then performing knowledge-guided multi-granularity attention intelligence extraction on the enhanced semantic understanding, intelligent transformation from raw information to structured intelligence can be achieved.

[0032] Optionally, key intelligence elements need to be identified first based on a multi-granularity attention mechanism, converting the extracted text and image data into semantic feature representations. The feature set is then fed into a pre-trained language model (such as BERT) for decomposition, obtaining feature representations of words, phrases, and sentences. Parallel word-level, phrase-level, and sentence-level attention networks are used to calculate the saliency weights of each granularity unit, and based on this, the original feature vectors (image and text data) are weighted and fused to generate multi-granularity feature representations focusing on key information. This mechanism can effectively identify and reinforce core content in text, such as highlighting keywords like "cracks," "displacement," and "rainfall"; capturing important phrases describing trends, such as "continuous expansion" and "overall sliding"; and comprehensively understanding the risk level conveyed by the semantics of the entire sentence.

[0033] In some preferred embodiments, based on the weights of granular units, text and images within a standardized data pool are matched and weighted fusion is performed to generate a multi-granularity feature vector focusing on feature information. This includes: calculating the matching degree between each text granular unit and image features using cosine similarity; matching text and images within the standardized data pool to form matching data; multiplying the weights of granular units by the matching degree and performing normalization processing; and fusing the normalized text and images by weighted summation to form a multi-granularity feature vector focusing on core information.

[0034] Optionally, cosine similarity can be used to quantify the "association between text units and images" (for example, if the image is an "accident scene," then "personal injury" has a higher matching degree). The matching degree is the "external association weight," and the attention weight is the "internal importance weight." Combining the two is more accurate. First, the attention weight is multiplied by the matching degree, and then normalized to ensure that units with high importance and high matching degree receive higher weights. Finally, through weighted summation, the scattered word, phrase, and sentence-level features are fused into a multi-granularity feature vector that focuses on core information.

[0035] Step S106 involves performing multi-granularity attention enhancement on the feature information in the initial database and deeply fusing it with the entity embedding vectors of the geological disaster knowledge graph to form a geological disaster intelligence database; such as Figure 4 As shown, Figure 4 This is a flowchart of multi-granularity intelligence extraction from a knowledge graph in a geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graphs, as described in Embodiment 1 of the present invention.

[0036] Optionally, by establishing a knowledge-enhanced multi-granularity attention neural network, attention weights are calculated at multiple granularities (words, phrases, sentences) to focus on key information, and innovatively, entity embedding vectors obtained from dynamic knowledge graphs are used. By performing element-level deep fusion (⊙ operation) with text features, deep neural network models can accurately identify and deeply understand the semantics and relationships of professional elements such as "landslides," "cracks," and "disaster-bearing bodies," much like experts. This deep fusion not only significantly improves the accuracy of geological disaster entity identification and key attribute extraction, but also naturally combines with pre-set relationships in the map to output structured intelligence that reflects the potential correlations of disasters, providing more complete and semantically rich intelligence support for decision-making.

[0037] In some preferred embodiments, multi-granularity attention enhancement is applied to the feature information existing in the initial database, and then deeply fused with the entity embedding vectors of the geological disaster knowledge graph to form a geological disaster intelligence database. This includes: performing similarity matching between the multi-granularity feature vectors in the initial database and the vector representations of entity nodes in the geological disaster knowledge graph to form entity links; based on the entity links, performing reasoning on the geological disaster knowledge graph using the breadth-first search principle to determine the direct and indirect relationships between adjacent entities; and based on the direct and indirect relationships between adjacent entities, selectively aggregating the associated entities and their embedding vectors to form knowledge enhancement vectors containing domain semantics, thereby obtaining the geological disaster intelligence database.

[0038] Optionally, the text and image semantic vectors obtained from the cross-modal semantic alignment process are first matched with the vector representations of entity nodes in the geological hazard knowledge graph for similarity. The Approximate Nearest Neighbor (HNSW) algorithm is used to calculate the similarity between the input text or image vectors and the entity vectors in the geological hazard knowledge graph, achieving accurate entity linking. Secondly, based on entity linking, breadth-first search is used to perform reasoning on the knowledge graph, discovering direct and indirect relationships between entities and inferring potential hazard chain risks. For example, starting from the entity "mountain crack," the relationship "may trigger" leads to "landslide," and the relationship "may block" leads to secondary hazard scenarios such as "mudslide" or "landslide dammed lake." Finally, the associated entities and their embedded vectors are selectively aggregated to form knowledge-enhanced vectors containing domain semantics. Deep interaction with the geological disaster knowledge graph enables the deep integration of multimodal perception data and knowledge in the field of geological disasters, providing semantic support for subsequent intelligence extraction.

[0039] In some preferred embodiments, multi-granularity feature vectors in the initial database are matched with the vector representations of entity nodes in the geological hazard knowledge graph to form entity links, including: in, The feature vector corresponding to the matched entity link includes text features extracted by multi-granularity attention and domain semantic features of the knowledge graph; This is a multi-granularity feature vector in the initial database; This is element-wise multiplication; It represents the vector of the entity node in the geological disaster knowledge graph.

[0040] Optionally, based on dynamic geological disaster knowledge graphs, semantic enhancement of disaster information can be applied to pay attention to the features of multi-source disaster information. Entity embedding vectors of geological disaster knowledge graph Deep integration. Through this process, key identified content (such as "cracks") is precisely linked to corresponding professional entities in the knowledge graph (such as "landslide precursors"), and associated risk concepts (such as "landslide risk") are activated, thereby injecting domain knowledge at the feature level. This effectively achieves knowledge-enhanced representation of disaster information, enabling machines to possess cognitive-level deep semantic parsing capabilities.

[0041] The final feature vector after fusion contains both text features extracted by multi-granularity attention and domain semantic features from the knowledge graph; This involves element-wise multiplication, multiplying textual features with domain semantic features one by one, thus "binding" key information in the text to domain knowledge. For each entity node in the geological disaster knowledge graph, there is a vector. These are multi-granularity feature vectors in the initial database.

[0042] By fusing and structuring multi-source heterogeneous information to assess its credibility, standardized geological disaster intelligence that can be used for emergency decision-making is generated. By dynamically assigning credibility weights to information from different sources, a weighted fusion can be performed to obtain the final confidence level. Let the confidence level of text extraction be... The confidence level of the image evidence is The confidence level of knowledge graph reasoning is The fusion confidence level is: Where α, β, and γ are weighting coefficients, and α + β + γ = 1.

[0043] Based on this, a structured integration of multi-element intelligence is performed, which semantically aligns and logically integrates the key entities and attributes extracted from the text, the visual representation evidence provided by the images, and the disaster chain risks and related paths inferred from the knowledge graph. Finally, a standardized geological disaster intelligence record with complete content, clear semantics, and unified format is generated, providing reliable information support for subsequent emergency command and decision-making.

[0044] Step S108: Obtain the semantic vector and resource status of the current disaster situation, compare them with the knowledge graph in the geological disaster intelligence database, and output a discretized task type package; Optionally, the current disaster status can be input into the cross-modal aligned semantic model to dynamically update the geological disaster knowledge graph and generate structured disaster intelligence information. This information, along with the real-time resource pool (available material resources), is then sent to the decision-making model (hierarchical reinforcement learning referee network) of the geological disaster emergency intelligent agent to quickly generate an executable instruction list (Cmd-List) and realize the mapping from disaster status to tasks.

[0045] Optionally, a Transformer decoder structure is adopted to perform macro-strategic analysis based on the input disaster semantic vector and resource status, and output discretized task type packages, including four core tasks: evacuation, fire fighting, material delivery, and medical rescue.

[0046] Step S110: Generate multiple decision schemes based on task type packages, calculate the actual benefits and predicted value of each decision scheme, and output the emergency intelligent agent decision scheme. The multiple decision schemes are used to indicate the sorting of task type packages and refine them into scheduling instructions.

[0047] Optionally, based on the same model, the task type package (macro task) output by the upper layer can be further refined into specific, executable resource scheduling instructions, including units such as shift (personnel), vehicle (vehicle), rack (aircraft), and ton (materials), and finally an executable decision-making scheme can be completed.

[0048] Specifically, a hierarchical reinforcement learning referee network architecture was constructed. Upper-layer strategies handle macro-level task planning, while lower-layer strategies manage micro-level resource allocation. An independent referee model is introduced for value assessment. A three-indicator delayed reward function for geological disaster emergency response effectiveness is embedded in the independent referee model. This function quantifies and weights three core business indicators: task achievement rate, timeliness achievement rate, and road restoration rate, transforming the quality of decisions into calculable benefit scores. By minimizing the difference between predicted value and actual benefits, the model drives the continuous self-optimization of the instruction generation strategy, achieving second-level mapping from disaster situation to instructions and ensuring the scientific nature and long-term effectiveness of the rescue plan.

[0049] In some preferred embodiments, multiple decision schemes are generated based on task type packages, the actual benefit and predicted value of each decision scheme are calculated, and the emergency agent decision scheme is output. This includes: inputting the task type package into a fully connected neural network to form simulated instructions; calculating the actual delayed reward after task completion under a random combination of simulated instructions; reducing the actual delayed reward by a coefficient to form a rescue benefit score after task completion; constructing a loss function based on the rescue benefit score and the actual delayed reward, wherein the loss function is used to indicate the loss relationship between the actual value and the predicted value; and outputting the emergency agent decision scheme with the minimum loss function as the optimization objective.

[0050] Optionally, the referee model is a two-layer fully connected neural network, with the current state s (structured disaster intelligence information) as input and the state value function as output. Its loss function is defined as: Where L is the loss value of the loss function, and R is the actual delayed reward / profit score after the task is completed, which is composed of instruction completion degree, timeliness achievement degree, and road recovery degree. This represents the expected benefit score before the rescue begins, which is a score calculated based on the expected completion of the instructions, the expected completion time, and the expected degree of road restoration. This represents the assessment score of the remaining rescue benefits after the rescue operation concludes. To avoid optimistic estimates by command personnel, an experience-based discount factor of 0.9 is applied. γ is the discount factor, typically set to 0.95, used to discount future expected benefits. Unlike the experience-based discount, this discount factor is a mathematical discount used for training the gradient.

[0051] Optionally, R is composed of the instruction completion degree, the timeliness achievement degree, and the road restoration degree, and the specific calculation is as follows: ,, ,, are weight coefficients, . is the weight of the task completion degree, which directly reflects "whether the command is fulfilled" and has the highest weight; is the weight of the timeliness achievement degree, which reflects the value of the "golden time" and has the second highest weight; is the weight of the road restoration degree, which is the benchmark for the smoothness of the lifeline and has a slightly lower weight. Generally, takes 0.45, takes 0.40, takes 0.15.

[0052] Optionally, make an empirical reduction according to a coefficient of 0.9, and the calculation is as follows: In addition, R + γV(s') is "actual harvest + discounted future value", which is equivalent to discounting the remaining potential after rescue back to the current, avoiding only focusing on the present and ignoring the follow-up.

[0053] Optionally, after calculating based on the above loss function, compare the relationship between "actual harvest + discounted future value" (R + γV(s')) and the predicted value V(s) to evaluate the pros and cons of the instruction plan: If R + γV(s') > V(s) (the actual score is higher), it means that the current instruction plan has higher benefits than expected (higher instruction completion degree, better timeliness achievement, faster road restoration), and it can be continuously optimized in the current direction; if R + γV(s') < V(s) (the actual score is lower), it means that the plan benefits are insufficient, and resource allocation or task priorities need to be adjusted; the model realizes self-optimization by minimizing the loss L and gradually approaches the optimal instruction generation strategy.

[0054] The typical performance grading is as follows: Excellent model: L < 100 points 2 ; Usable model: L ≈ 100 - 400 points 2 ; L > 400 points 2 It is basically considered that the prediction is out of control, and retraining or checking for data anomalies is required.

[0055] Optionally, the Cmd-List generated by the agent can be directly integrated with existing command and dispatch platforms, or offline path distribution can be completed through an edge caching model to ensure that instructions are reachable and executable in complex field environments. Finally, the field execution results can be fed back to the system, forming a learning closed loop for the geological disaster emergency agent. This data will be used to drive the optional self-learning and evolution of the agent's internal models (such as dynamic knowledge graphs and referee networks), enabling the agent to continuously optimize its perception and decision-making capabilities based on the actual effects of historical decisions.

[0056] In some preferred embodiments, the calculation of the actual delayed reward after task completion under a random combination of simulated instructions includes: Where R represents the actual delayed reward; The delayed reward for the completed instruction; The delay reward for instructions completed within the time limit; A delayed reward for achieving road clearance upon completion of the instruction.

[0057] In some preferred embodiments, a loss function is constructed based on the rescue benefit score and the actual delayed reward, including: Where L is the loss function; R is the true delayed reward; and γ is the discount factor. Score the expected benefits before the rescue begins; Rate the rescue benefits after the mission is completed.

[0058] Through steps S102 to S110, the system first fuses multi-source information such as satellite imagery and UAV video. Based on a dynamic geological disaster knowledge graph, it rapidly identifies disaster situations using an emergency multimodal large-scale model and outputs structured disaster intelligence information. Next, in the decision engine, this information, along with real-time resource data, is fed into a hierarchical reinforcement learning adjudicator network. A delayed reward function is constructed using three indicators: task achievement, timeliness, and road restoration, automatically planning the optimal rescue task list. Then, the rescue task list can be directly connected to the existing command platform via a scheduling interface or distributed offline via an edge caching model. This effectively achieves precise alignment at the professional semantic level, accurate cross-modal matching, and second-level mapping from disaster situation to instructions, ensuring the scientific nature and long-term effectiveness of the rescue plan.

[0059] Example 2 This embodiment also provides a geological disaster emergency intelligent agent decision-making device based on a dynamic knowledge graph. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0060] According to an embodiment of the present invention, an apparatus embodiment for implementing the above method is also provided. Figure 5 This is a schematic diagram of the structure of a geological disaster emergency intelligent decision-making device based on dynamic knowledge graph in Embodiment 2 of the present invention, as shown below. Figure 5 As shown, the above-mentioned device includes: an acquisition module 201, a cross-modal alignment module 202, a fusion module 203, a task package generation module 204, and a decision module 205, wherein: The acquisition module 201 acquires monitoring data, satellite remote sensing data, and network data to form a standardized data pool. The monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images. The cross-modal alignment module 202 performs cross-modal alignment of text and images in the standardized data pool based on feature information, unifies the semantic vector space, and forms an initial database for mutual matching of text and images; The fusion module 203 performs multi-granularity attention enhancement on the feature information in the initial database and deeply fuses it with the entity embedding vector of the geological disaster knowledge graph to form a geological disaster intelligence database. The task package generation module 204 obtains the semantic vector and resource status of the current disaster situation, compares them with the knowledge graph in the geological disaster intelligence database, and outputs a discretized task type package. The decision module 205 generates multiple decision schemes based on task type packages, calculates the actual benefits and predicted value of each decision scheme, and outputs the emergency intelligent agent decision scheme. Among them, multiple decision schemes are used to indicate the sorting of task type packages and refine them into scheduling instructions.

[0061] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0062] It should be noted that the acquisition module 201, cross-modal alignment module 202, fusion module 203, task package generation module 204, and decision module 205 mentioned above correspond to steps S102 to S110 in the embodiments. The instances and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run on a computer terminal.

[0063] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0064] The aforementioned geological disaster emergency intelligent agent decision-making device based on dynamic knowledge graph may further include a processor and a memory. The aforementioned acquisition module 201, cross-modal alignment module 202, fusion module 203, task package generation module 204, and decision module 205 are all stored in the memory as program modules, and the processor executes the aforementioned program modules stored in the memory to realize the corresponding functions.

[0065] The processor contains a core that retrieves the corresponding program modules from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0066] According to an embodiment of this application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein, when the program runs, it controls the device where the non-volatile storage medium is located to execute any of the above-mentioned geological disaster emergency intelligent agent decision-making methods based on dynamic knowledge graphs.

[0067] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals, and the non-volatile storage medium includes stored programs.

[0068] Optionally, during program execution, the device containing the non-volatile storage medium performs the following functions: acquire monitoring data, satellite remote sensing data, and network data to form a standardized data pool, wherein the monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images; perform cross-modal alignment of text and images in the standardized data pool based on feature information, unify the semantic vector space, and form an initial database for text and image matching; perform multi-granularity attention enhancement on the feature information in the initial database and deeply fuse it with the entity embedding vector of the geological disaster knowledge graph to form a geological disaster intelligence database; acquire the semantic vector and resource status of the current disaster situation, compare it with the knowledge graph in the geological disaster intelligence database, and output a discretized task type package; generate multiple decision schemes based on the task type package, calculate the actual benefit and predicted value of each decision scheme, and output an emergency intelligent agent decision scheme, wherein the multiple decision schemes are used to instruct the sorting of task type packages and refine them into scheduling instructions.

[0069] According to an embodiment of this application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-described geological disaster emergency intelligent agent decision-making methods based on dynamic knowledge graphs.

[0070] According to an embodiment of this application, an embodiment of a computer program product is also provided. Optionally, in this embodiment, the computer program product includes a computer program that, when executed by a processor, implements the steps of any of the above-mentioned geological disaster emergency intelligent agent decision-making methods based on dynamic knowledge graphs.

[0071] Optionally, when the aforementioned computer program product is executed on a data processing device, it is suitable to execute an initialization program with the following steps: acquiring monitoring data, satellite remote sensing data, and network data to form a standardized data pool, wherein the monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images; performing cross-modal alignment on the text and images in the standardized data pool based on feature information, unifying the semantic vector space, and forming an initial database for text and image matching; performing multi-granularity attention enhancement on the feature information existing in the initial database, and deeply fusing it with the entity embedding vector of the geological disaster knowledge graph to form a geological disaster intelligence database; acquiring the semantic vector and resource status of the current disaster situation, comparing it with the knowledge graph in the geological disaster intelligence database, and outputting a discretized task type package; generating multiple decision schemes based on the task type package, calculating the actual benefit and predicted value of each decision scheme, and outputting an emergency intelligent agent decision scheme, wherein the multiple decision schemes are used to instruct the sorting and refinement of the task type package into scheduling instructions.

[0072] This invention provides an electronic device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring monitoring data, satellite remote sensing data, and network data to form a standardized data pool, wherein the monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images; performing cross-modal alignment on the text and images in the standardized data pool based on feature information, unifying the semantic vector space, and forming an initial database for text-image matching; performing multi-granularity attention enhancement on the feature information in the initial database and deeply fusing it with the entity embedding vectors of a geological disaster knowledge graph to form a geological disaster intelligence database; acquiring the semantic vector and resource status of the current disaster situation, comparing it with the knowledge graph in the geological disaster intelligence database, and outputting a discretized task type package; generating multiple decision schemes based on the task type package, calculating the actual benefit and predicted value of each decision scheme, and outputting an emergency intelligent agent decision scheme, wherein the multiple decision schemes are used to instruct the sorting and refinement of the task type package into scheduling instructions.

[0073] The order of the above embodiments of the present invention is merely for description and does not represent the superiority or inferiority of the embodiments.

[0074] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of modules described above can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between modules, and may be electrical or other forms.

[0076] The modules described above as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0077] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0078] If the aforementioned integrated modules are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0079] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A decision-making method for geological disaster emergency response based on dynamic knowledge graphs, characterized in that, include: Acquire monitoring data, satellite remote sensing data, and network data to form a standardized data pool, wherein the monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images; The text and images in the standardized data pool are aligned across modalities based on feature information to unify the semantic vector space and form an initial database for text and image matching. Multi-granularity attention enhancement is applied to the feature information existing in the initial database, and it is deeply fused with the entity embedding vector of the geological disaster knowledge graph to form a geological disaster intelligence database. Obtain the semantic vector and resource status of the current disaster situation, compare them with the knowledge graph in the geological disaster intelligence database, and output a discretized task type package; Based on the task type package, multiple decision schemes are generated, the actual benefits and predicted value of each decision scheme are calculated, and the emergency intelligent agent decision scheme is output. The multiple decision schemes are used to indicate the sorting and refinement of the task type package into scheduling instructions.

2. The geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph as described in claim 1, characterized in that, The text and images within the standardized data pool are cross-modal aligned based on feature information to unify the semantic vector space, forming an initial database for text-image matching, including: Transform text and images within a standardized data pool into feature representations; The weights of each granular unit are calculated by using parallel word-level, phrase-level, and sentence-level attention networks in the feature representation, wherein the granular unit is used to indicate the basic units at the word, phrase, or sentence level from which the text is segmented. Based on the weights of the granularity units, text and images in the standardized data pool are matched and weighted fusion is performed to generate multi-granularity feature vectors that focus on feature information. The initial database is formed by utilizing all feature information, multi-granularity feature vectors, and matched text and images.

3. The geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph according to claim 2, characterized in that, Based on the weights of the granularity units, text and images within the standardized data pool are matched and weighted fusion is performed to generate multi-granularity feature vectors that focus on key feature information, including: The matching degree between each text granularity unit and image features is calculated by cosine similarity. The text and images in the standardized data pool are matched to form matching data. The weight of the granularity unit is multiplied by the matching degree, and then normalized. By weighted summation of normalized text and images, a multi-granularity feature vector focusing on core information is fused.

4. The geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph according to claim 1, characterized in that, Multi-granularity attention enhancement is applied to the feature information existing in the initial database, and deep fusion is performed with the entity embedding vectors of the geological disaster knowledge graph to form a geological disaster intelligence database, including: The multi-granularity feature vectors in the initial database are matched with the vector representations of entity nodes in the geological disaster knowledge graph to form entity links; Based on entity links, the breadth-first search principle is used to reason on the geological disaster knowledge graph to determine the direct and indirect relationships between adjacent entities; Based on the direct and indirect relationships between adjacent entities, the associated entities and their embedding vectors are selectively aggregated to form knowledge-enhanced vectors containing domain semantics, thus obtaining a geological disaster intelligence database.

5. The geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph according to claim 4, characterized in that, The multi-granularity feature vectors in the initial database are matched with the vector representations of entity nodes in the geological disaster knowledge graph to form entity links, including: in, The feature vector corresponding to the matched entity link includes text features extracted by multi-granularity attention and domain semantic features of the knowledge graph; This is a multi-granularity feature vector in the initial database; This is element-wise multiplication; This is a vector of entity nodes in the geological disaster knowledge graph.

6. The geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph according to claim 1, characterized in that, Based on the task type package, multiple decision schemes are generated, the actual benefit and predicted value of each decision scheme are calculated, and the emergency agent decision scheme is output, including: The task type package is input into a fully connected neural network to form simulated instructions, and the real delayed reward after the task is completed under the random combination of the simulated instructions is calculated. The actual delayed reward is reduced by a coefficient to form a rescue benefit score after the task is completed; Based on the rescue benefit score and the actual delayed reward, a loss function is constructed, wherein the loss function is used to indicate the loss relationship between the actual value and the predicted value; With the minimum loss function as the optimization objective, the output of the minimum loss function is the corresponding emergency agent decision-making scheme.

7. The geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph as described in claim 6, characterized in that, Calculating the actual delayed reward after task completion under the simulated instruction random combination method includes: Where R represents the actual delayed reward; The delayed reward for the completed instruction; The delay reward for instructions completed within the time limit; A delayed reward for achieving road clearance upon completion of the instruction.

8. The geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph according to claim 7, characterized in that, Based on the rescue benefit score and the actual delay reward, a loss function is constructed, including: Where L is the loss function; R is the true delayed reward; and γ is the discount factor. Score the expected benefits before the rescue begins; Rate the rescue benefits after the mission is completed.

9. A geological disaster emergency intelligent agent decision-making device based on dynamic knowledge graph, characterized in that, include: The acquisition module acquires monitoring data, satellite remote sensing data, and network data to form a standardized data pool. The monitoring data is text, the satellite remote sensing data is images, and the network data includes both text and images. The cross-modal alignment module performs cross-modal alignment of text and images in the standardized data pool based on feature information, unifies the semantic vector space, and forms an initial database for mutual matching of text and images; The fusion module performs multi-granularity attention enhancement on the feature information in the initial database and deeply fuses it with the entity embedding vector of the geological disaster knowledge graph to form a geological disaster intelligence database. The task package generation module obtains the semantic vector and resource status of the current disaster situation, compares them with the knowledge graph in the geological disaster intelligence database, and outputs a discretized task type package. The decision-making module generates multiple decision schemes based on the task type package, calculates the actual benefit and predicted value of each decision scheme, and outputs the emergency intelligent agent decision scheme. The multiple decision schemes are used to indicate the sorting and refinement of the task type package into scheduling instructions.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the geological disaster emergency intelligent agent decision-making method based on dynamic knowledge graph as described in any one of claims 1 to 8.

Images