Fire emergency disposal script generation method and system based on efficient multi-agent cooperation

Abstract

The application discloses a fire emergency disposal document generation method and system based on multi-agent efficient cooperation, and relates to the technical field of artificial intelligence and emergency management. The method comprises the following steps: a monitoring and evaluation agent collects fire monitoring data in real time, calls a pre-set knowledge base to store regional fire disposal plan files, divides the fire into risk levels, and obtains risk level information; the risk level information is transmitted to a resource planning agent; the resource planning agent performs task allocation and resource scheduling, transmits tasks to multiple resource management agents, and interacts with the multiple resource management agents; the resource scheduling result and emergency resource feedback information executed by the multiple resource management agents are transmitted to an emergency document generation agent to generate an emergency document; and an agent efficient cooperation mechanism based on information compression and implicit reasoning is adopted to optimize the reasoning and cooperation process of the agent. The application can improve the work efficiency of the multi-agent system.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and emergency management technology, and in particular to a method and system for generating fire emergency response documents based on efficient multi-agent collaboration. Background Technology

[0002] With the continuous improvement of urbanization, increased urban population density, diversified building forms, and rapid development of underground space, fire risk, as one of the most dangerous and destructive public safety incidents in cities, is becoming increasingly complex and diverse. Therefore, how to quickly and accurately formulate emergency response plans after a fire has become a key issue in urban safety management.

[0003] Currently, fire emergency response relies heavily on manual analysis at fire command centers. Commanders typically need to synthesize multi-source heterogeneous data from video, sensor networks, geographic information systems, and on-site reports, combined with historical cases and response experience, to formulate reasonable emergency plans. In this process, some emergency platforms have introduced intelligent decision-making support modules to provide suggestions on resource allocation, route selection, and personnel evacuation, thereby improving response efficiency to some extent. However, existing systems often only process single-dimensional information, and there are still significant shortcomings in the real-time integration and efficient utilization of multi-dimensional dynamic data.

[0004] Currently, the generation of emergency response documents for fires still relies heavily on manual processes, resulting in low efficiency and significant limitations when dealing with time-sensitive disasters like fires. These limitations primarily include: the rapid changes in fire situations leading to time lags in manual document processing; the complexity of information dissemination limiting the capacity of manual information processing to handle the information flow generated by fires, especially large-scale fires; and the inherent possibility of human error, making it impossible to guarantee the accuracy and correctness of judgments and decisions.

[0005] No automated generation methods were found for generating fire emergency response documents. For other emergency scenarios similar to fire emergencies, the following methods are currently available.

[0006] Model- and rule-based digital emergency response systems, such as those that combine disaster models, resource models, simulation models, and rule models to generate emergency plans and support 3D visualization and dynamic adjustment, are used to generate plans. Other systems rely on template matching and hotspot analysis, such as constructing emergency analysis tables and hotspot distribution maps, and then determining implementing agencies and plans based on keywords. Deep learning-based scenario-based generation methods use a large amount of historical and sensor data as training samples for the automatic generation of emergency plans for specific scenarios, such as tunnel fires. Systems focusing on personalized evacuation routes and distributed emergency responses combine location information and sensor data to generate individualized evacuation routes or dispatch instructions for different personnel. While these solutions each have advantages in automation, visualization, personalization, and learning capabilities, they are generally based on single models, centralized logic, or black-box deep learning, and their support for concurrent processing of multi-source heterogeneous data and complex collaborative decision-making remains insufficient.

[0007] In summary, current technologies for generating emergency response documents for fire incidents suffer from drawbacks such as low efficiency, poor information concurrency, and low decision-making accuracy. In emergency scenarios similar to fires, these technologies rely too heavily on centralized strategies, which are insufficient when faced with complex changes, heterogeneous data sources, or high-concurrency demands. Summary of the Invention

[0008] To address the technical problems of low efficiency, poor information concurrency, and low decision-making accuracy in existing technologies, this invention provides a method and system for generating fire emergency response documents based on efficient multi-agent collaboration. The technical solution is as follows:

[0009] On the one hand, a method for generating emergency response documents for fire incidents based on efficient multi-agent collaboration is provided. This method is implemented by a device for generating emergency response documents for fire incidents based on efficient multi-agent collaboration. The method includes:

[0010] S1. The monitoring and evaluation intelligent agent collects fire monitoring data in real time; the fire monitoring data includes: sensor detection data, geographical location information data, and video surveillance image data; based on the fire monitoring data, it calls the regional fire response plan file stored in the pre-set knowledge base; based on the risk classification criteria of the plan file, it classifies the fire risk level to obtain risk level information; and transmits the risk level information to the resource planning intelligent agent.

[0011] S2. The resource planning agent allocates tasks based on the received risk level information and transmits the tasks to multiple resource management agents, while interacting with multiple resource management agents. Among them, an efficient agent collaboration mechanism based on information compression and implicit reasoning is adopted to optimize the reasoning and collaboration processes of the monitoring and evaluation agent, the resource planning agent, multiple resource management agents, and the emergency document generation agent.

[0012] S3. Multiple resource management agents execute tasks to obtain emergency resource feedback; the emergency resource feedback is transmitted to the resource planning agent; the resource planning agent generates resource scheduling results based on the emergency resource feedback information and risk level information; the resource scheduling results and the emergency resource feedback information executed by the multiple resource management agents are transmitted to the emergency document generation agent.

[0013] S4. The emergency document generation agent integrates the resource scheduling results and emergency resource feedback information executed by multiple resource management agents to generate emergency documents that meet the format requirements of the regional emergency response plan.

[0014] On the other hand, a system for generating fire emergency response documents based on efficient multi-agent collaborative action is provided. This system is applied to a method for generating fire emergency response documents based on efficient multi-agent collaborative action. The system includes:

[0015] A monitoring and assessment intelligent agent is used to collect fire monitoring data in real time. The fire monitoring data includes sensor detection data, geographic location information data, and video surveillance image data. Based on the fire monitoring data, the agent calls up the regional fire response plan files stored in a pre-set knowledge base. Based on the risk classification criteria of the plan files, the agent classifies the fire risk level to obtain risk level information. The agent then transmits the risk level information to the resource planning intelligent agent.

[0016] The resource planning agent is used to allocate tasks based on the received risk level information and transmit the tasks to multiple resource management agents, while interacting with multiple resource management agents; the resource planning agent generates resource scheduling results based on emergency resource feedback information and risk level information; and transmits the resource scheduling results and emergency resource feedback information executed by multiple resource management agents to the emergency document generation agent.

[0017] Multiple resource management agents are used to perform tasks and obtain emergency resource feedback; the emergency resource feedback is transmitted to the resource planning agent; wherein, an efficient agent collaboration mechanism based on information compression and implicit reasoning is adopted to optimize the reasoning and collaboration processes of the monitoring and evaluation agent, the resource planning agent, multiple resource management agents, and the emergency document generation agent;

[0018] An emergency document generation agent is used to integrate resource scheduling results and emergency resource feedback information executed by multiple resource management agents to generate emergency documents that meet the format requirements of regional emergency response plans.

[0019] On the other hand, a multi-agent efficient collaborative fire emergency response document generation device is provided. The multi-agent efficient collaborative fire emergency response document generation device includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the methods described above for multi-agent efficient collaborative fire emergency response document generation methods is implemented.

[0020] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement any of the above-described methods for generating fire emergency response documents based on efficient multi-agent collaboration.

[0021] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0022] In fire emergency response, the embodiments of the present invention adopt a multi-agent parallel collaborative mechanism, which eliminates the limitations of human manpower and can improve the real-time performance and accuracy of fire response.

[0023] The embodiments of the present invention can achieve efficient integration and optimized allocation of various types of emergency resources through the interaction between a resource planning intelligent agent and multiple resource management intelligent agents;

[0024] The emergency response document generation intelligent agent of this invention can automatically output response documents based on standard contingency plan formats, which can avoid the problem of long manual writing time, and at the same time ensure the standardization and completeness of the document content.

[0025] The efficient collaborative mechanism for intelligent agents based on information compression and implicit reasoning designed in this embodiment of the invention can accelerate the processing of fire information with long contexts, speed up the reasoning process of intelligent agents, and allow the thinking process of intelligent agents to occur in the hidden space rather than the linguistic space, thereby improving the working efficiency of the entire multi-agent system and the response speed to fires. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of a method for generating emergency response documents for fires based on efficient multi-agent collaboration, provided by an embodiment of the present invention.

[0028] Figure 2This is a detailed flowchart of a fire disaster emergency response plan generation system based on efficient collaborative analysis of a multi-agent system, provided by an embodiment of the present invention.

[0029] Figure 3 This is a detailed flowchart of the design of an efficient collaborative mechanism for intelligent agents based on information compression and implicit reasoning, provided by an embodiment of the present invention.

[0030] Figure 4 This is a block diagram of a fire emergency response document generation system based on efficient multi-agent collaborative action provided by an embodiment of the present invention;

[0031] Figure 5 This is a schematic diagram of a fire emergency response document generation device based on multi-agent efficient collaborative fire response provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0033] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0034] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0035] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0036] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0037] This invention provides a method for generating emergency response documents for fire incidents based on efficient multi-agent collaboration. This method can be implemented by a device for generating such documents, which can be a terminal or a server. Figure 1The flowchart shown is for a method to generate emergency response documents for fires based on efficient multi-agent collaboration. The processing flow of this method may include the following steps:

[0038] S1. The monitoring and evaluation agent collects fire monitoring data in real time. The fire monitoring data includes sensor detection data, geographic location information data, and video surveillance image data. Based on the fire monitoring data, the agent calls the regional fire response plan documents stored in the pre-set knowledge base. Based on the risk classification standards of the plan documents, the agent classifies the fire risk level and obtains the risk level information. The agent then transmits the risk level information to the resource planning agent.

[0039] The monitoring and assessment agent is used to collect fire monitoring data in real time, including but not limited to sensor detection data such as fire alarms and smoke detectors, geographic location information, and video surveillance images. When a suspected fire is detected, the monitoring and assessment agent retrieves the regional fire response plan documents stored in the knowledge base and determines the risk level of the fire based on the risk classification standards therein. The monitoring and assessment agent can dynamically predict the development trend of the fire and transmit the risk assessment results to the resource planning agent.

[0040] S2. The resource planning agent allocates tasks based on the received risk level information and transmits the tasks to multiple resource management agents, while interacting with multiple resource management agents. Among them, an efficient agent collaboration mechanism based on information compression and implicit reasoning is adopted to optimize the reasoning and collaboration processes of the monitoring and evaluation agent, the resource planning agent, multiple resource management agents, and the emergency document generation agent.

[0041] The resource planning agent is the core control unit of the multi-agent system. It receives risk level information from the monitoring and assessment agents and interacts with multiple resource management agents to allocate tasks and coordinate resources based on fire severity and response priorities. When resource conflicts or shortages occur, the resource planning agent uses optimization strategies to mediate and ensure the smooth execution of tasks and the continuity of the overall workflow.

[0042] S3. Multiple resource management agents execute tasks to obtain emergency resource feedback; transmit the emergency resource feedback to the resource planning agent; the resource planning agent generates resource scheduling results based on the emergency resource feedback information and risk level information; transmit the resource scheduling results and the emergency resource feedback information executed by multiple resource management agents to the emergency document generation agent.

[0043] Each of the multiple resource management agents is responsible for managing a specific type of emergency resource, including but not limited to fire trucks, ambulances, fire extinguishers, and medical supplies. Each resource management agent is responsible for maintaining the inventory or status information of its respective resource, such as location and quantity, and updating availability data in real time. After a task is assigned, the resource management agents interact with the resource planning agent, transmitting resource allocation information and execution feedback back to the system. Furthermore, the multiple resource management agents can collaborate horizontally, supporting cross-regional resource scheduling and sharing.

[0044] S4. The emergency document generation agent integrates the resource scheduling results and emergency resource feedback information executed by multiple resource management agents to generate emergency documents that meet the format requirements of the regional emergency response plan.

[0045] Among them, the emergency document generation intelligent agent is used to receive the scheduling results of the resource planning intelligent agent and the execution feedback of each resource management intelligent agent. It integrates the fire situation overview, risk level description, scheduling resource details and action plan, etc., and automatically generates emergency documents that meet the format requirements of the regional emergency response plan, so that relevant emergency departments can use and issue them quickly.

[0046] This invention, through the division of labor and collaboration among various intelligent agents, enables a closed-loop process for fire emergency response, including: fire awareness and risk assessment, resource planning and task coordination, resource allocation and status feedback, and emergency response document generation. This invention can rapidly complete risk assessment, resource allocation, and emergency response document generation when a fire occurs, providing scientific and efficient decision support for emergency management departments.

[0047] Optionally, the monitoring and evaluation agent, the resource planning agent, the multiple resource management agents, and the emergency document generation agent are all based on large language models for reasoning;

[0048] One feasible implementation method is, for example Figure 2 The diagram shows a detailed flowchart of a fire disaster emergency response plan generation system based on efficient collaborative analysis of a multi-agent system, according to an embodiment of the present invention. The system includes: a monitoring and assessment agent, a resource planning agent, multiple resource management agents, and an emergency document generation agent. The monitoring and assessment agent monitors fire information in real time, including information from video surveillance, alarms, and geographic information. Based on the monitored fire information, the monitoring agent calls a risk classification knowledge base to classify the fire level and obtain risk level information. The risk level is fed back to the resource planning agent in real time. The resource planning agent allocates tasks according to the risk level and transmits the task allocation to each resource management agent. Each agent is responsible for managing one type of emergency resource, such as... Figure 2Resource management agent 1 is responsible for fire trucks, resource management agent 2 is responsible for fire extinguishers, resource management agent 3 is responsible for ambulances, and resource management agent 4 is responsible for medical supplies. Multiple resource management agents and resource planning agents are dynamically scheduled and the scheduling results are obtained. The scheduling results and the execution feedback of each resource management agent are transmitted to the emergency document generation agent for information integration and output of the emergency document for the current fire situation.

[0049] In one feasible implementation, the fire emergency response document generation system based on multi-agent technology constructed in this embodiment of the invention consists of several agents, including a monitoring and evaluation agent, a resource planning agent, a resource management agent, and an emergency document generation agent. The collaborative feedback between agents is usually accompanied by a large amount of high-frequency information exchange. In this process, the working efficiency of the multi-agent technology is mainly affected by the following two points: the time consumed by the attention mechanism to process information with a long context sequence, and the time consumed by the agent reasoning process based on a large language model.

[0050] Optionally, the reasoning process based on a large language model includes:

[0051] The input message sequence is transformed into a vector through word embedding. The vector representation of the message sequence is obtained through positional encoding. The vector representation of the message sequence is then input into a multi-layer Transformer for stacked computation. The hidden state of the last layer is output. Softmax is performed on the hidden state of the last layer to obtain the conditional probability distribution of the next token. The reasoning process based on the large language model is represented by the following formulas (1)-(5):

[0052] (1)

[0053] (2)

[0054] (3)

[0055] (4)

[0056] (5)

[0057] in, Given the length of the input message sequence, ; This represents the input message sequence token; express Word embedding; This represents the location encoding vector; L is the number of layers in the Transformer model. This is the first parameter of the Transformer head of the model; This is the second parameter of the Transformer head of the model; This represents a mathematical function that normalizes a vector into a probability distribution for predicting the next token. Indicates the first Transformer layer; Indicates that it is located at the th The first Transformer layer in the layer The hidden state vector of each token; Indicates the position after encoding. Vector representation of each token; express 3D real vector; Indicates based on the previous The token prediction is as follows. The probability distribution of each token; Indicates the first The hidden state vector corresponding to the last token of the Transformer layer.

[0058] In one feasible implementation, the input message sequence in formula (1) is processed by word embedding in formula (2) and position encoding in formula (3) to obtain the vector representation of the sequence; the vector representation of the sequence is then computed by stacking multiple Transformers in formula (4) to finally obtain the hidden state of the last layer. The hidden state of the last layer is calculated using the language head and Softmax in formula (5) to obtain the conditional probability distribution of the next token. After obtaining the new token, the reasoning is performed again to continue generating the next token.

[0059] Among them, the efficient agent collaboration mechanism based on information compression and implicit reasoning includes: an efficient agent collaboration mechanism based on information compression and an efficient agent collaboration mechanism based on implicit reasoning.

[0060] In one feasible implementation, during the process of an agent processing information from other agents, the time complexity of the attention mechanism is on the quadratic level of the message sequence length. When the message sequence length is long, the computational cost of the attention mechanism cannot be ignored. Furthermore, fire emergency response information is usually numerous and complex; when the sequence length exceeds the model's maximum context window length, it exceeds the upper limit of the agent's processing capacity, resulting in both long processing times and the potential loss of important fire information due to the agent truncating the message sequence. During agent inference, in addition to the necessary Transformer computation, the large model vocabulary and the computational process of normalizing the Softmax vector exponent in the language model head are time-consuming. To ensure efficient and timely fire emergency warning work and improve the overall system response speed, it is necessary to optimize the agent's inference and collaboration processes. Therefore, this invention designs an efficient agent collaboration mechanism based on information compression to alleviate the attention computation bottleneck caused by long message sequence lengths, while simultaneously improving the agent's processing speed of fire information; it also designs an efficient agent collaboration mechanism based on implicit reasoning to improve the speed of the agent's judgment and reasoning process regarding fire information; and finally, it combines these two mechanisms to construct an efficient agent collaboration mechanism based on information compression and implicit reasoning.

[0061] Optionally, the implementation process of the efficient agent collaboration mechanism based on information compression includes:

[0062] In formula (1), the message sequence length is t. Considering that when the sequence length is long, the agent model takes a long time to process once. Therefore, the message sequence is divided into M sub-message sequences. One processing is divided into M processing sessions, and each processing session processes one sub-message sequence, thereby achieving information compression and alleviating the quadratic computational cost of the attention mechanism.

[0063] (1) The input message sequence is segmented to obtain multiple sub-message sequences; the message sequence segmentation process is represented by the following formula (6):

[0064] (6)

[0065] in, Indicates the first A sequence of sub-messages, , ;

[0066] To ensure the transmission of contextual information during multiple processing steps, an implicit summary vector is introduced for each processing step. Referring to efficient agent collaboration methods based on implicit reasoning, the hidden state of the last layer is used as the implicit summary vector in each processing step, containing rich contextual information. After a single processing step, the implicit summary vector is passed (unlike message passing between agents, this is information passing between steps when a single agent processes a long information sequence) to the next processing step as a prefix for the sub-message sequence, thus achieving complete transmission of contextual information. For message compression and cost reduction, only the hidden states of the last few tokens are transmitted. Furthermore, to avoid the contextual information carried by directly transmitting the hidden states of the last few tokens in a sub-sequence being too one-sided, token positions for the summary vector are reserved in advance at the end of the sub-sequence in each processing step.

[0067] (2) The first sub-message sequence is converted into a vector by word embedding and processed by position encoding and multi-layer Transformer to obtain the hidden state of the last layer; the hidden state of the last layer is passed as the implicit digest vector; wherein, in the process of processing the end of each sub-sequence, the token position of the digest vector needs to be reserved in advance, which is expressed by the following formulas (7)-(9):

[0068] (7)

[0069] (8)

[0070] (9)

[0071] in, It only serves as a placeholder; Indicates processing the first When the sub-message sequence is compared with the previous one A new sub-message sequence is constructed by concatenating the implicit summary vectors obtained from processing the sub-message sequences. This indicates the token corresponding to the reserved digest vector. Hidden state vectors of the Transformer layer; express Word embedding vectors; Indicates processing the first The implicit summary vector obtained from the sub-message sequence;

[0072] (3) The hidden state is used as a prefix for the next sub-message sequence and then processed for further information compression, ultimately compressing the initial message sequence of length t into a sequence of length t. The implicit summary vector sequence is represented by the following formula (10):

[0073] (10)

[0074] in, Represents an implicit summary vector sequence.

[0075] Among them, the implicit summary vector sequence The implicit summary vector sequence retains complete message context information, greatly reducing the processing time of long sequence information. At the same time, the shorter implicit summary vector sequence can also accelerate the subsequent reasoning process.

[0076] Optionally, the efficient collaborative mechanism of intelligent agents based on information compression is trained by optimizing the first loss function; wherein the first loss function is expressed by the following formula (11):

[0077] (11)

[0078] Where t represents the length of the input message sequence; M represents the total number of segments in the input message sequence; This represents the value of the first loss function; Indicates the first One token; Indicates processing the first The implicit summary vector obtained from the sub-message sequence; Indicates the first The left boundary position of each sub-message sequence; Indicates the first The right boundary position of each sub-message sequence.

[0079] In one feasible implementation, during the processing of formula (4), the hidden state of the last layer... Carrying rich contextual information, at the same time The dimension is the same as the word embedding dimension. Considering that the vocabulary and Softmax calculations in formula (5) are time-consuming, this embodiment of the invention will... Substitute word embeddings are used for the next step of reasoning, thus saving time spent generating tokens. Furthermore, since the agent's reasoning process mainly occurs in the hidden space and does not need to be expressed in natural language, the information passed by the agent to other agents is more concise and the expression is more efficient.

[0080] In one feasible implementation, the standard reasoning process is a recursive iteration of formula (1)-formula (5), and the reasoning process for the k-th token is represented by the following formula (12):

[0081] (12)

[0082] in, Indicates the predicted first One token; This represents a mathematical function that normalizes a vector into a probability distribution for predicting the next token. These represent the parameters of the Transformer header of the model; This indicates that the entire Transformer is related to the previous... The stacking calculation process of word embeddings for each token.

[0083] Optionally, the implementation process of the efficient agent cooperation mechanism based on implicit reasoning includes:

[0084] The hidden state of the last layer is embedded in the word for recursive iterative implicit reasoning. The hidden state of the last layer when the k-th token is reasoned is replaced by the k-th token is represented by the following formula (13):

[0085] (13)

[0086] in, Indicates replacing the first The hidden state vector for reasoning with each token; Indicates the reasoning of the first When the token is the first The hidden state vector of the last Transformer layer corresponding to each token; express The corresponding word embedding vector; This represents the entire Transformer stacking computation process;

[0087] In this process, the hidden state of the last layer is used to replace the word embedding. Except for the first step of reasoning, the reasoning process is a recursive iteration of formula (3) - formula (4).

[0088] In the first step of reasoning, the word embedding of the input sequence is still used, which is represented by the following formula (14):

[0089] (14)

[0090] Where t represents the length of the input message sequence.

[0091] In one feasible implementation, the above process allows the agent to reason in the hidden space without the need for redundant steps of converting into explicit natural language. This can speed up the agent's processing of various fire information while reducing the amount of low-density reasoning information in the information transmission between agents.

[0092] Among them, after a preset number of implicit inferences, the standard inference process is restored, which is represented by the following formula (15):

[0093] (15)

[0094] in, The first inference One token; These represent the parameters of the Transformer header of the model; This represents a mathematical function that normalizes a vector into a probability distribution for predicting the next token.

[0095] The final output of the agent's reasoning result is represented as ,in This represents the reasoning result of the intelligent agent; This indicates that for a length of The input message sequence The first output token predicted after implicit inference; This indicates that for a length of The input message sequence The second output token predicted after implicit inference; This indicates that for a length of The input message sequence The last output token predicted after the implicit inference step.

[0096] The variable T can be dynamically adjusted based on the needs of accuracy and timeliness. A larger T results in more reasoning steps for the agent and a more accurate understanding of fire-related information, while a smaller T results in faster reasoning and a quicker response to the fire. Finally, the agent's processing results are fed back to other agents as information.

[0097] In one feasible implementation, the agent-efficient collaborative mechanism based on implicit reasoning gradually increases the number of implicit reasoning steps during the training process, gradually acquiring the ability of multi-step implicit reasoning from simple to complex.

[0098] Optionally, the efficient agent collaboration mechanism based on implicit reasoning is trained by optimizing the second loss function; wherein the second loss function is expressed by the following formula (16):

[0099] (16)

[0100] in, This represents the value of the second loss function; Indicates the number of implicit inferences; This represents the first step in the entire reasoning process. Step; t represents the length of the input message sequence; This represents the last token in the current prediction input; Indicates the first The hidden state vector corresponding to the last token during implicit inference; Indicates the first The target token for step-by-step reasoning.

[0101] Among them, such as Figure 3 The diagram shows a detailed flowchart of an efficient agent collaboration mechanism based on information compression and implicit reasoning, provided by an embodiment of the present invention. In one feasible implementation, the process of implementing the efficient agent collaboration mechanism based on information compression and implicit reasoning includes: after receiving a message sequence, the agent compresses the information using the efficient agent collaboration mechanism based on information compression, compressing the message sequence of initial length t into a sequence of length t. Implicit summary vector sequence ; Implicit summary vector sequence The prompt words of the agent are concatenated, and implicit reasoning is performed through an efficient collaborative mechanism of agents based on implicit reasoning. Formula (13) can be further expressed as the following formula (17):

[0102] (17)

[0103] in This represents the sequence of prompts given by the agent. After a preset number of implicit inferences, the standard inference process is restored, and equation (15) can be further expressed as equation (18):

[0104] (18)

[0105] One feasible implementation involves designing an efficient agent collaboration mechanism based on information compression and implicit reasoning. Information compression saves time spent on message understanding while enabling agents to process long sequences of fire information. This allows for the condensation of lengthy contextual information into shorter implicit summary vectors, further accelerating the agent's reasoning process. Implicit reasoning, building upon message compression, operates within the hidden space, avoiding the step of converting to explicit natural language, thus improving both reasoning quality and speed. This efficient agent collaboration mechanism enables efficient processing of fire emergency response information and agent collaboration, improving the overall system response speed.

[0106] In fire emergency response, the embodiments of the present invention adopt a multi-agent parallel collaborative mechanism, which eliminates the limitations of human manpower and can improve the real-time performance and accuracy of fire response.

[0107] The embodiments of the present invention can achieve efficient integration and optimized allocation of various types of emergency resources through the interaction between a resource planning intelligent agent and multiple resource management intelligent agents;

[0108] The emergency response document generation intelligent agent of this invention can automatically output response documents based on standard contingency plan formats, which can avoid the problem of long manual writing time, and at the same time ensure the standardization and completeness of the document content.

[0109] The efficient collaborative mechanism for intelligent agents based on information compression and implicit reasoning designed in this embodiment of the invention can accelerate the processing of fire information with long contexts, speed up the reasoning process of intelligent agents, and allow the thinking process of intelligent agents to occur in the hidden space rather than the linguistic space, thereby improving the working efficiency of the entire multi-agent system and the response speed to fires.

[0110] Figure 4 This is a block diagram of a system for generating emergency response documents for fires based on efficient multi-agent collaboration, provided by an embodiment of the present invention. This system is used for generating emergency response documents for fires based on efficient multi-agent collaboration. (Refer to...) Figure 4 The system includes a monitoring and evaluation agent 410, a resource planning agent 420, multiple resource management agents 430, and an emergency document generation agent 440. Among them:

[0111] The monitoring and evaluation intelligent agent 410 is used to collect fire monitoring data in real time. The fire monitoring data includes: sensor detection data, geographical location information data, and video surveillance image data. Based on the fire monitoring data, it calls the regional fire response plan file stored in the pre-set knowledge base; based on the risk classification criteria of the plan file, it classifies the fire risk level to obtain risk level information; and transmits the risk level information to the resource planning intelligent agent.

[0112] Resource planning agent 420 is used to allocate tasks based on received risk level information and transmit tasks to multiple resource management agents, while interacting with multiple resource management agents; the resource planning agent generates resource scheduling results based on emergency resource feedback information and risk level information; and transmits the resource scheduling results and emergency resource feedback information executed by multiple resource management agents to the emergency document generation agent.

[0113] Multiple resource management agents 430 are used to perform tasks to obtain emergency resource feedback; and transmit the emergency resource feedback to the resource planning agent; wherein, an efficient agent collaboration mechanism based on information compression and implicit reasoning is adopted to optimize the reasoning and collaboration processes of the monitoring and evaluation agent, the resource planning agent, multiple resource management agents, and the emergency document generation agent;

[0114] Emergency document generation agent 440 is used to integrate resource scheduling results and emergency resource feedback information executed by multiple resource management agents to generate emergency documents that meet the format requirements of the regional emergency response plan.

[0115] Optionally, the monitoring and evaluation intelligent agent, the resource planning intelligent agent, the multiple resource management intelligent agents, and the emergency document generation intelligent agent are all based on reasoning using a large language model;

[0116] Among them, the efficient agent collaboration mechanism based on information compression and implicit reasoning includes: an efficient agent collaboration mechanism based on information compression and an efficient agent collaboration mechanism based on implicit reasoning.

[0117] Optionally, the reasoning process based on the large language model includes:

[0118] The input message sequence is transformed into a vector through word embedding. The vector representation of the message sequence is obtained through positional encoding. The vector representation of the message sequence is then input into a multi-layer Transformer for stacked computation. The hidden state of the last layer is output. Softmax is performed on the hidden state of the last layer to obtain the conditional probability distribution of the next token. The reasoning process based on the large language model is represented by the following formulas (1)-(5):

[0119] (1)

[0120] (2)

[0121] (3)

[0122] (4)

[0123] (5)

[0124] in, Given the length of the input message sequence, ; This represents the input message sequence token; express Word embedding; This represents the location encoding vector; L is the number of layers in the Transformer model. This is the first parameter of the Transformer head of the model; This is the second parameter of the Transformer head of the model; This represents a mathematical function that normalizes a vector into a probability distribution for predicting the next token. Indicates the first Transformer layer; Indicates that it is located at the th The first Transformer layer in the layer The hidden state vector of each token; Indicates the position after encoding. Vector representation of each token; express 3D real vector; Indicates based on the previous The token prediction is as follows. The probability distribution of each token; Indicates the first The hidden state vector corresponding to the last token of the Transformer layer.

[0125] Optionally, the implementation process of the information compression-based efficient agent collaboration mechanism includes:

[0126] (1) The input message sequence is segmented to obtain multiple sub-message sequences; the message sequence segmentation process is represented by the following formula (6):

[0127] (6)

[0128] in, Indicates the first A sequence of sub-messages, , ;

[0129] (2) The first sub-message sequence is converted into a vector by word embedding and processed by position encoding and multi-layer Transformer to obtain the hidden state of the last layer; the hidden state of the last layer is passed as the implicit digest vector; wherein, in the process of processing the end of each sub-sequence, the token position of the digest vector needs to be reserved in advance, which is expressed by the following formulas (7)-(9):

[0130] (7)

[0131] (8)

[0132] (9)

[0133] in, It only serves as a placeholder; Indicates processing the first When the sub-message sequence is compared with the previous one A new sub-message sequence is constructed by concatenating the implicit summary vectors obtained from processing the sub-message sequences. This indicates the token corresponding to the reserved digest vector. Hidden state vectors of the Transformer layer; express Word embedding vectors; Indicates processing the first The implicit summary vector obtained from the sub-message sequence;

[0134] (3) The hidden state is used as a prefix for the next sub-message sequence and then processed for further information compression, ultimately compressing the initial message sequence of length t into a sequence of length t. The implicit summary vector sequence is represented by the following formula (10):

[0135] (10)

[0136] in, Represents an implicit summary vector sequence.

[0137] Optionally, the efficient collaborative mechanism of intelligent agents based on information compression is trained by optimizing a first loss function; wherein the first loss function is expressed by the following formula (11):

[0138] (11)

[0139] Where t represents the length of the input message sequence; M represents the total number of segments in the input message sequence; This represents the value of the first loss function; Indicates the first One token; Indicates processing the first The implicit summary vector obtained from the sub-message sequence; Indicates the first The left boundary position of each sub-message sequence; Indicates the first The right boundary position of each sub-message sequence.

[0140] Optionally, the implementation process of the efficient agent collaboration mechanism based on implicit reasoning includes:

[0141] The hidden state of the last layer is embedded in the word for recursive iterative implicit reasoning. The hidden state of the last layer when the k-th token is reasoned is replaced by the k-th token is represented by the following formula (12):

[0142] (12)

[0143] in, Indicates replacing the first The hidden state vector for reasoning with each token; Indicates the reasoning of the first When the token is the first The hidden state vector of the last Transformer layer corresponding to each token; express The corresponding word embedding vector; This represents the entire Transformer stacking computation process;

[0144] The process of reverting to standard reasoning after a predetermined number of implicit inferences is represented by the following formula (13):

[0145] (13)

[0146] in, The first inference One token; These represent the parameters of the Transformer header of the model; This represents a mathematical function that normalizes a vector into a probability distribution for predicting the next token.

[0147] The final output of the agent's reasoning result is represented as ,in This represents the reasoning result of the intelligent agent; This indicates that for a length of The input message sequence The first output token predicted after implicit inference; This indicates that for a length of The input message sequence The second output token predicted after implicit inference; This indicates that for a length of The input message sequence The last output token predicted after implicit inference.

[0148] Optionally, the agent-efficient collaborative mechanism based on implicit reasoning is trained by optimizing a second loss function; wherein the second loss function is expressed by the following formula (14):

[0149] (14)

[0150] in, This represents the value of the second loss function; Indicates the number of implicit inferences; This represents the first step in the entire reasoning process. Step; t represents the length of the input message sequence; This represents the last token in the current prediction input; Indicates the first The hidden state vector corresponding to the last token during implicit inference; Indicates the first The target token for step-by-step reasoning.

[0151] In fire emergency response, the embodiments of the present invention adopt a multi-agent parallel collaborative mechanism, which eliminates the limitations of human manpower and can improve the real-time performance and accuracy of fire response.

[0152] The embodiments of the present invention can achieve efficient integration and optimized allocation of various types of emergency resources through the interaction between a resource planning intelligent agent and multiple resource management intelligent agents;

[0153] The emergency response document generation intelligent agent of this invention can automatically output response documents based on standard contingency plan formats, which can avoid the problem of long manual writing time, and at the same time ensure the standardization and completeness of the document content.

[0154] The efficient collaborative mechanism for intelligent agents based on information compression and implicit reasoning designed in this embodiment of the invention can accelerate the processing of fire information with long contexts, speed up the reasoning process of intelligent agents, and allow the thinking process of intelligent agents to occur in the hidden space rather than the linguistic space, thereby improving the working efficiency of the entire multi-agent system and the response speed to fires.

[0155] Figure 5 This is a schematic diagram of a multi-agent efficient collaborative fire emergency response document generation device provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the multi-agent-based efficient collaborative fire emergency response document generation device may include the above-mentioned... Figure 4 The illustrated system is a multi-agent collaborative fire emergency response document generation system. Optionally, the multi-agent collaborative fire emergency response document generation device 510 may include a first processor 2001.

[0156] Optionally, the multi-agent efficient collaborative fire emergency response document generation device 510 may also include a memory 2002 and a transceiver 2003.

[0157] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0158] The following is combined with Figure 5 A detailed introduction to each component of the 510 multi-agent efficient collaborative fire emergency response document generation device is provided:

[0159] The first processor 2001 is the control center of the multi-agent efficient collaborative fire emergency response document generation device 510. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0160] Optionally, the first processor 2001 can execute various functions of the multi-agent efficient collaborative fire emergency response document generation device 510 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0161] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 5 CPU0 and CPU1 are shown in the diagram.

[0162] In a specific implementation, as one example, the multi-agent efficient collaborative fire emergency response document generation device 510 may also include multiple processors, such as... Figure 5 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0163] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0164] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently, and may be connected via the interface circuit of the multi-agent efficient collaborative fire emergency response document generation device 510. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0165] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0166] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 5 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0167] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently, and it can be connected to the interface circuit of the multi-agent efficient collaborative fire emergency response document generation device 510. Figure 5 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0168] It should be noted that, Figure 5 The structure of the multi-agent efficient collaborative fire emergency response document generation device 510 shown in the figure does not constitute a limitation on the router. The actual multi-agent efficient collaborative fire emergency response document generation device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0169] Furthermore, the technical effects of the multi-agent efficient collaborative fire emergency response document generation device 510 can be referred to the technical effects of the multi-agent efficient collaborative fire emergency response document generation method described in the above method embodiments, and will not be repeated here.

[0170] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or it may be any conventional processor, etc.

[0171] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0172] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0173] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0174] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0175] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0176] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0177] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, systems, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0178] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, 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; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0179] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0180] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0181] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a 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 described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0182] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for generating emergency response documents for fires based on efficient multi-agent collaboration, characterized in that, The method is implemented by a monitoring and evaluation agent, a resource planning agent, multiple resource management agents, and an emergency response document generation agent of a multi-agent fire emergency response document generation system; the method includes: S1. The monitoring and evaluation intelligent agent collects fire monitoring data in real time; the fire monitoring data includes: sensor detection data, geographical location information data, and video surveillance image data; based on the fire monitoring data, it calls the regional fire response plan file stored in the pre-set knowledge base; based on the risk classification criteria of the plan file, it classifies the fire risk level to obtain risk level information; and transmits the risk level information to the resource planning intelligent agent. S2. The resource planning agent allocates tasks based on the received risk level information and transmits the tasks to multiple resource management agents, while interacting with multiple resource management agents. Among them, an efficient agent collaboration mechanism based on information compression and implicit reasoning is adopted to optimize the reasoning and collaboration processes of the monitoring and evaluation agent, the resource planning agent, multiple resource management agents, and the emergency document generation agent. Among them, the monitoring and evaluation intelligent agent, the resource planning intelligent agent, the multiple resource management intelligent agents, and the emergency document generation intelligent agent are all based on large language models for reasoning; Among them, the efficient agent collaboration mechanism based on information compression and implicit reasoning includes: an efficient agent collaboration mechanism based on information compression and an efficient agent collaboration mechanism based on implicit reasoning. The implementation process of the information compression-based efficient agent collaboration mechanism includes: The input message sequence is segmented to obtain multiple sub-message sequences; the message sequence segmentation process is represented by the following formula (1): （1） in, Indicates the first A sequence of sub-messages, , ; Indicates the first The left boundary position of each sub-message sequence; Indicates the first The right boundary position of each sub-message sequence; M represents the total number of segments into which the input message sequence is divided. The first sub-message sequence is transformed into a vector through word embedding and processed by positional encoding and multi-layer Transformer to obtain the hidden state of the last layer; the hidden state of the last layer is passed as an implicit digest vector; wherein, in the process of processing the end of each sub-sequence, the token position of the digest vector needs to be reserved in advance, which is expressed by the following formulas (2)-(4): （2） （3） （4） in, It only serves as a placeholder; Indicates processing the first When the sub-message sequence is compared with the previous one A new sub-message sequence is constructed by concatenating the implicit summary vectors obtained from processing the sub-message sequences. This indicates the token corresponding to the reserved digest vector. Hidden state vectors of the Transformer layer; express Word embedding vectors; Indicates processing the first The implicit summary vector obtained from the sub-message sequence; Will The hidden state is used as a prefix for the next sub-message sequence and then processed for further information compression, ultimately compressing the initial message sequence of length t into a sequence of length t. The implicit summary vector sequence is represented by the following formula (5): （5） in, Represents an implicit summary vector sequence; The implementation process of the efficient agent collaboration mechanism based on implicit reasoning includes: The hidden state of the last layer is embedded in the word for recursive iterative implicit reasoning. The hidden state of the last layer when the k-th token is reasoned is replaced by the k-th token is represented by the following formula (6): （6） in, Indicates replacing the first The hidden state vector for reasoning with each token; Indicates the reasoning of the first When the token is the first The hidden state vector of the last Transformer layer corresponding to each token; express The corresponding word embedding vector; This represents the entire Transformer stacking computation process; Among them, after a preset number of implicit inferences, the process returns to the standard inference process, which is represented by the following formula (7): （7） in, Indicates the generated first One token; These represent the parameters of the Transformer header of the model; This represents a mathematical function that normalizes a vector into a probability distribution for predicting the next token. The final output of the agent's reasoning result is represented as ,in This represents the reasoning result of the intelligent agent; This indicates that for a length of The input message sequence The first output token predicted after implicit inference; This indicates that for a length of The input message sequence The second output token predicted after implicit inference; This indicates that for a length of The input message sequence The last output token predicted after implicit inference; S3. Multiple resource management agents execute tasks to obtain emergency resource feedback information; transmit the emergency resource feedback information to the resource planning agent; the resource planning agent generates resource scheduling results based on the emergency resource feedback information and risk level information; transmit the resource scheduling results and the emergency resource feedback information executed by multiple resource management agents to the emergency document generation agent. S4. The emergency document generation agent integrates the resource scheduling results and emergency resource feedback information executed by multiple resource management agents to generate emergency documents that meet the format requirements of the regional emergency response plan.

2. The method for generating emergency response documents for fires based on efficient multi-agent collaboration as described in claim 1, characterized in that, The reasoning process based on the large language model includes: The input message sequence is transformed into a vector through word embedding. The vector representation of the message sequence is obtained through positional encoding. The vector representation of the message sequence is then input into a multi-layer Transformer for stacked computation. The hidden state of the last layer is output. Softmax is performed on the hidden state of the last layer to obtain the conditional probability distribution of the next token. The reasoning process based on the large language model is represented by the following formulas (8)-(12): （8） （9） （10） （11） （12） in, Given the length of the input message sequence, ; This represents the input message sequence token; express Word embedding; This represents the location encoding vector; L is the number of layers in the Transformer model. This is the first parameter of the Transformer head of the model; This is the second parameter of the Transformer head of the model; This represents a mathematical function that normalizes a vector into a probability distribution for predicting the next token. Indicates the first Transformer layer; Indicates that it is located at the th The first Transformer layer in the layer The hidden state vector of each token; Indicates the position after encoding. Vector representation of each token; express 3D real vector; Indicates based on the previous The token prediction is as follows. The probability distribution of each token; Indicates the first The hidden state vector corresponding to the last token of the Transformer layer.

3. The method for generating emergency response documents for fires based on efficient multi-agent collaboration as described in claim 1, characterized in that, The efficient collaborative mechanism of intelligent agents based on information compression is trained by optimizing the first loss function; wherein the first loss function is expressed by the following formula (13): （13） Where t represents the length of the input message sequence; M represents the total number of segments in the input message sequence; This represents the value of the first loss function; Indicates the generated first One token; Indicates processing the first The implicit summary vector obtained from the sub-message sequence; Indicates the first The left boundary position of each sub-message sequence; Indicates the first The right boundary position of each sub-message sequence.

4. The method for generating emergency response documents for fires based on efficient multi-agent collaboration as described in claim 1, characterized in that, The efficient agent collaboration mechanism based on implicit reasoning is trained by optimizing the second loss function; wherein the second loss function is expressed by the following formula (14): （14） in, This represents the value of the second loss function; Indicates the number of implicit inferences; This indicates the generation of the k-th token; t represents the length of the input message sequence. This represents the last token in the current prediction input; Indicates the first The hidden state vector corresponding to the last token during implicit inference; Indicates the generated first A token.

5. A system for generating emergency response documents for fire incidents based on efficient multi-agent collaboration, wherein the system is used to implement the method for generating emergency response documents for fire incidents based on efficient multi-agent collaboration as described in any one of claims 1-4, characterized in that, The system includes: A monitoring and assessment intelligent agent is used to collect fire monitoring data in real time. The fire monitoring data includes sensor detection data, geographic location information data, and video surveillance image data. Based on the fire monitoring data, the agent calls up the regional fire response plan files stored in a pre-set knowledge base. Based on the risk classification criteria of the plan files, the agent classifies the fire risk level to obtain risk level information. The agent then transmits the risk level information to the resource planning intelligent agent. The resource planning agent is used to allocate tasks based on the received risk level information and transmit the tasks to multiple resource management agents, while interacting with multiple resource management agents; the resource planning agent generates resource scheduling results based on emergency resource feedback information and risk level information; and transmits the resource scheduling results and emergency resource feedback information executed by multiple resource management agents to the emergency document generation agent. Multiple resource management agents are used to perform tasks and obtain emergency resource feedback information; the emergency resource feedback information is transmitted to the resource planning agent; among them, an efficient agent collaboration mechanism based on information compression and implicit reasoning is adopted to optimize the reasoning and collaboration processes of the monitoring and evaluation agent, the resource planning agent, multiple resource management agents, and the emergency document generation agent; An emergency document generation agent is used to integrate resource scheduling results and emergency resource feedback information executed by multiple resource management agents to generate emergency documents that meet the format requirements of regional emergency response plans.

6. A device for generating emergency response documents for fires based on efficient multi-agent collaboration, characterized in that, The multi-agent efficient collaborative fire emergency response document generation device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 4.

Images