Fire-fighting alarm receiving and answering system and method and storage medium

Abstract

The invention provides a fire-fighting alarm-receiving response system and method, and the method comprises the steps: responding to alarm-receiving communication, extracting alarm condition elements based on a preset inquiry rule, and generating a pre-alarm-receiving work order comprising the alarm condition elements; processing the alarm condition elements by using a preset judgment model; specifically, the method comprises the following steps: constructing a fire-fighting alarm condition dialogue model with Transform as a basic framework, taking alarm receiving questions and answers as a sequence generation problem, and reasoning a fire behavior which should be inquired in the next step according to the obtained fire condition in an autoregression mode until the model considers that enough information is obtained for fire grade judgment; the steps are executed for many times, understanding of the fire scene condition is deepened step by step, and real-time fire grade evaluation is generated on the basis. The defect that a response result deviates from the behavior of an actual operator to a certain extent due to system deviation existing in a strategy learning method adopting a Markov decision process and reinforcement learning in an existing alarm situation automatic alarm receiving response or type system is overcome.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a fire alarm answering system, method and storage medium. Background Art

[0002] In the field of disaster emergency management, especially in the fire emergency response scenario, real-time and accurate information acquisition and judgment are the keys to improving the emergency response efficiency. Traditional fire emergency responses usually rely on manual operations and judgments. However, with the development of artificial intelligence technology, automated alarm handling questions and answers and fire situation assessments have become possible. In this project, the core concept of the model is to obtain the real-time situation at the fire scene through automated interactive conversations, including but not limited to key information such as the burning area of the fire source, the number of trapped people, the smoke condition, and the available fire-fighting facilities at the scene. Through the interactive question-and-answer session, the model can gradually deepen its understanding of the fire scene situation and generate a real-time fire rating assessment based on this. Summary of the Invention

[0003] The present invention provides a fire alarm answering system, method, storage medium and management system, which solve the defects in the prior art that the processing results of the existing automatic alarm answering system and method deviate from the behavior of actual operators to a certain extent when using Markov decision process and reinforcement learning for policy learning methods, and at the same time, the model efficiency is low and it lacks interpretability.

[0004] The technical solution of the present invention is realized as follows: A fire alarm answering method includes the following steps:

[0005] In response to an alarm communication, extract alarm elements based on a preset inquiry rule, and generate a pre-alarm work order including the alarm elements;

[0006] Process the alarm elements using a preset determination model; specifically: construct a fire alarm dialogue model with Transformer as the basic architecture, regard the alarm answering as a sequence generation problem, and infer the next fire situation to be asked based on the obtained fire situation in an autoregressive manner until the model believes that enough information has been obtained for fire rating judgment; execute the above steps multiple times to gradually deepen the understanding of the fire scene situation and generate a real-time fire rating assessment based on this.

[0007] As a preferred technical solution, the model uses Transformer as the basic architecture, regards the alarm handling questions and answers as a sequence generation problem, and infers the next fire situation to be asked based on the obtained fire situation in an autoregressive manner until the model believes that enough information has been obtained for fire rating judgment.

[0008] As a preferred technical solution, the fire alarm dialogue model uses the explicit state of the fire as the model input to generate the implicit state through training, where the "END" symbol represents the end signal, indicating the end state generation and switching to fire classification reasoning; in the model training process, the prediction symbol [S] sequence is used to realize the autoregressive generation of the implicit state sequence, which can be regarded as learning multiple stages of state reasoning at one time; the level prediction symbol [D] is used to learn and generate the target level based on all fire status information, where the explicit state refers to the necessary inquiry and status response in the alarm receiving stage, and the invisible state refers to the optional inquiry and status response in the alarm receiving stage.

[0009] As a preferred technical solution, the model first obtains the explicit state of the fire, calculates the probability distribution of the implicit state by splicing the state prediction symbol [S], and inquires about the state with the highest probability in the implicit state. At this time, the user will reply True, False or uncertain to the question; for uncertain information, the above steps will be repeated to find the next state with the highest probability and ask the user; according to the user's True or False answer, the model will splice the newly obtained invisible state, and then use the state prediction symbol to predict the next state; when the end signal is inferred from the state symbol, or when the confidence of the predicted state probability is lower than the threshold, the model will stop asking and switch to level determination, and predict the target fire level by splicing the level prediction symbol [D].

[0010] As a preferred technical solution, the explicit status of the fire includes the number of trapped people and the burning area. One or more of the following explicit characteristics can be added as needed: the name of the caller, the caller's telephone number, the time of the call, the address of the alarm, the location of the caller, the burning floor, the number of trapped people, the number of injured people and the degree of injury, and the smoke conditions.

[0011] As a preferred technical solution, the hidden states include open flames, smoke, white smoke, black smoke, explosion hazards, surrounding combustibles, spread, high-rise buildings, collapse, whether fire-fighting facilities are complete and whether water sources are sufficient. The hidden states are adjusted according to the climate zone, season, time period, building density, and safety alert level of the area.

[0012] As a preferred technical solution, the alarm classification results include repeated alarms, non-duplicate alarms, ordinary alarms and important alarms; when the alarm classification result is determined to be a repeated alarm, a secondary confirmation of the repeated alarm is performed, and when a positive response to the repeated alarm is obtained, the current alarm communication is terminated.

[0013] As a preferred technical solution, transfer questions and answers are carried out based on the interactive voice response component to obtain the question and answer results. When the question and answer results meet the preset conditions for the person in the fire pre-alarm voice communication, the alarm elements are extracted based on the preset inquiry rules; when the question and answer results do not meet the preset conditions, manual response is initiated.

[0014] A fire alarm answering system includes the following modules:

[0015] Alarm receiving center module: used to receive alarm information and receive early warning danger information, extract alarm elements based on preset inquiry rules, and generate a pre-alarm work order including the alarm elements;

[0016] Fire alarm dialogue model module: A fire alarm dialogue model with Transformer as the basic architecture is preset; it is used to process the alarm elements received by the alarm receiving center module, regard the alarm question and answer as a sequence generation problem, and infer the next fire situation to be asked in an autoregressive manner according to the obtained fire situation until the model believes that enough information has been obtained for fire level judgment; execute the above steps multiple times to gradually deepen the understanding of the fire scene situation, and generate a real-time fire level assessment on this basis;

[0017] Danger information display module: used to obtain and display the danger scene information in real time by using GIS map management technology according to the information of the danger warning module of the alarm receiving center module and the fire alarm dialogue model module;

[0018] Danger warning module: used to collect the danger factors in the urban monitoring area in real time, then identify and predict the danger in the monitoring area, and then judge the early warning danger information and analyze the probability of danger occurrence.

[0019] A non-transitory storage medium, when the instructions in the storage medium are executed by a fire alarm answering system, enables the device to execute the above-mentioned fire alarm answering method.

[0020] Compared with the prior art, the present solution has the following beneficial effects:

[0021] Manifest and latent traits are introduced for police situation elements, and then the fire alarm response is regarded as a multi-step reasoning problem. Considering that the wiring process can be naturally regarded as the generation of a sequence, in this work, automatic wiring is reformulated as a sequence generation task. Different from previous reinforcement learning methods, the multi-step query process is explicitly modeled as a task of generating a fire situation sequence. This can improve the efficiency and interpretability of the multi-step reasoning process. The problem of how to capture the uncertainty in the reasoning process is solved, and the accurate level is judged with very little labeled data and as few conversation turns as possible. At the same time, the system bias existing in the existing police situation automatic alarm response or type system using the Markov decision process and reinforcement learning for policy learning methods is solved, which leads to the defect that the response result deviates from the behavior of the actual operator to a certain extent.

[0022] Different from the requirement of only accuracy rate in traditional multi-turn Q&A, the goal of fire alarm handling Q&A is to determine the main situation of the fire and the fire level in as few inquiry turns as possible, so as to accurately dispatch fire forces and resources. Brief Description of the Drawings

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0024] Figure 1 It is the model architecture diagram of the fire alarm dialogue model based on the Transformer architecture of the present invention;

[0025] Figure 2 It is the schematic diagram of the reasoning method of the fire alarm dialogue model based on the Transformer architecture of the present invention;

[0026] Figure 3 It is the schematic diagram of a fire alarm response system of the present invention. Detailed Embodiments

[0027] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0028] A fire alarm response method of the present application responds to an alarm communication, extracts alarm elements based on a preset interrogation rule, and generates a pre-alarm work order including the alarm elements; processes the alarm elements using a preset determination model; specifically: constructs a fire alarm dialogue model with Transformer as the basic architecture, regards the alarm answering as a sequence generation problem, and infers the next fire situation to be asked according to the obtained fire situation in an autoregressive manner until the model believes that enough information has been obtained to judge the fire level; executes the above steps multiple times to gradually deepen the understanding of the fire scene situation, and generates a real-time fire level assessment based on this.

[0029] The model uses Transformer as the basic architecture, regards the alarm answering as a sequence generation problem, and infers the next fire situation to be asked according to the obtained fire situation in an autoregressive manner until the model believes that enough information has been obtained to judge the fire level. Referring to the fire emergency rescue guide, the classification of fire levels mainly depends on the burning area and the number of trapped people. However, in actual situations, some other factors such as the smoke condition and the type of the building on fire will also affect the actual level determination and the number of dispatched vehicles, which may lead to an upgrade or additional dispatch. Therefore, in this model, the burning area and the number of trapped people are regarded as explicit states, which need to be asked and obtained in the first two rounds of the dialogue. Other fire situations are regarded as implicit states, which need to be predicted round by round according to the previous dialogue information. Among them, the explicit state refers to the necessary interrogation and state reply in the alarm receiving stage, and the implicit state refers to the optional interrogation and state reply in the alarm receiving stage.

[0030] The architecture of the model is as Figure 1 shown. The explicit state of the fire is used as the model input to train and generate the implicit state, where the "END" symbol represents the end signal, indicating the end of the state generation and transferring to the fire classification inference. During the training process, the autoregressive generation of the implicit state sequence is realized by using the predicted symbol [S] sequence, which can be regarded as inferring the states of multiple stages in one learning; using the level prediction symbol [D], the target level is learned and generated based on all fire state information.

[0031] The working principle of the model in the inference stage is as Figure 2As shown in the figure. The model first obtains the explicit state of the fire (number of trapped people, burning area), calculates the probability distribution of the hidden state by splicing the state prediction symbol [S], and queries the state with the highest probability in the hidden state. At this time, the user will reply True, False, or uncertain to the question. For uncertain information, the above steps will be repeated to find the next state with the highest probability and query the user. According to the user's True or False answer, the model splices the newly obtained hidden state and then uses the state prediction symbol to predict the next state. When the end signal is inferred from the state symbol or when the confidence level of the predicted state probability is lower than the threshold, the model will stop querying and transfer to the level determination, and predict the target fire level by splicing the level prediction symbol [D].

[0032] Among them, the explicit state of the fire includes the number of trapped people and the burning area. One or more of the following explicit traits can be added as needed: name of the alarm caller, phone number of the alarm caller, alarm time, police situation address, location of the alarm caller, burning floor, number of trapped people, number of injured people and degree of injury, and smoke condition. The hidden state includes situations such as open fire, smoke, white smoke, black smoke, explosive danger, surrounding combustibles, spread, high-rise building, collapse, integrity of fire-fighting facilities, and adequacy of water sources. The hidden state is adjusted according to the climate zone, season, time period, building density at the location, and security alert level of the area.

[0033] Among them, the police situation classification results include repeated police situations, non-repeated police situations, ordinary police situations, and important police situations; when the police situation classification result is determined to be a repeated police situation, a secondary confirmation of the repeated police situation is carried out. When an affirmative response to the repeated police situation is obtained, the current police call communication is ended. Based on the interactive voice response component, transfer questions and answers are carried out to obtain the question and answer results. When the question and answer results of the person in the fire pre-alarm voice communication meet the preset conditions, the police situation elements are extracted based on the preset inquiry rules; when the question and answer results do not meet the preset conditions, manual response is started.

[0034] Meanwhile, a fire alarm answering system for executing the fire alarm answering method is provided, including the following modules:

[0035] Alarm receiving center module: used to receive alarm information and early warning danger information, extract police situation elements based on preset inquiry rules, and generate a pre-alarm work order including the police situation elements;

[0036] Fire Alarm Conversation Model Module: It is pre - installed with a fire alarm conversation model based on the Transformer architecture; it is used to process the alarm elements received by the Alarm Receiving Center Module, regard the alarm Q&A as a sequence generation problem, and infer the next fire situation to be asked in an autoregressive manner according to the obtained fire situation until the model believes that enough information has been obtained for fire level judgment; execute the above steps multiple times, gradually deepen the understanding of the fire scene situation, and generate a real - time fire level assessment on this basis.

[0037] Danger Information Display Module: It is used to obtain and display the danger scene information in real - time by using GIS map management technology according to the information of the danger warning module of the Alarm Receiving Center Module and the Fire Alarm Conversation Model Module.

[0038] Danger Warning Module: It is used to collect the danger factors in the urban monitoring area in real - time, then identify and predict the danger in the monitoring area, and then judge the warning danger information and analyze the probability of danger occurrence.

[0039] Working Principle: Due to the existence of hidden states, this task can be regarded as a multi - step reasoning problem. Its challenge lies in how to capture the uncertainty in the reasoning process and judge the accurate level with little labeled data and as few dialogue turns as possible. Most previous methods usually handle this problem as a continuous decision - making process (Markov decision process) and use reinforcement learning for policy learning. However, there is a deviation in how reinforcement learning queries the fire situation and makes this level division only under the reward of the final accuracy, which also deviates from the actual behavior of operators to a certain extent. Considering that the connection process can be naturally regarded as a sequence generation, in this work, automatic connection is reformulated as a sequence generation task. Different from the previous reinforcement learning methods, the multi - step query process is explicitly modeled as a task of generating a sequence containing fire situations. This can improve the efficiency and interpretability of the multi - step reasoning process.

[0040] For the above - mentioned model training and test data:

[0041] Due to the particularity of the fire alarm handling problem, it is difficult to obtain real dialogue records. Therefore, all the training and test data in this project are from artificially constructed conversations. The dialogue data are divided into 10 fire scenarios, including markets, vehicles, high - rise buildings, underground buildings, etc. In the specific construction process, the rules of fire alarm handling Q&A in the Fire Emergency Rescue Manual are referred to, such as key questions, level and decision descriptions, dispatching tips, etc.

[0042] Model Input Example

[0043] Specific data examples are as follows:

[0044] {

[0045] "type": "Market fire",

[0046] "explicit_inform_slots": {

[0047] "Number of trapped people = 0": true,

[0048] "Combustion area = 0 - 200": true

[0049] },

[0050] "implicit_inform_slots": {

[0051] "Open fire": true,

[0052] "Smoke": true,

[0053] "White smoke": false,

[0054] "Black smoke": true,

[0055] "Explosive hazard": true,

[0056] "Surrounding combustibles": false,

[0057] "Spread": false,

[0058] "High-rise building": true,

[0059] "Collapse": false,

[0060] "Fire-fighting facilities intact": false,

[0061] "Adequate water source": true

[0062] },

[0063] "fire_tag": 1

[0064] }

[0065] The above status description is the questioning elements of the fire alarm receiver, and true / false is the feedback from the alarm sender. For example, in the above case, "Number of trapped people 0": true means that the alarm receiver asks "Are there any trapped people?", and the alarm sender answers "No"; the alarm receiver continues to ask "Is the combustion area more than 200 square meters?", and the alarm sender answers "No"; the alarm receiver continues to ask "Can open fire be seen?", and the alarm sender answers "Open fire can be seen"... Finally, the alarm receiver ends the questioning and judges that the fire level = 1 ("fire_tag": 1).

[0066] In a specific application scenario, the model will only receive the fire scenario type (type) and explicit status (explicit_inform_slots), and complete the inference of the implicit status step by step based on the content of the above two fields.

[0067] The performance index chart is as follows:

[0068] Dacc (Accuracy of Grade Judgment) FRec (Recall Rate of Hidden Fire) ATurn (Average Number of Dialogue Turns) 0.785 0.811 10.2

[0069] Expected application scenarios

[0070] In the fire emergency response scenario, timely, accurate, and efficient information collection and processing are the keys to ensuring personnel safety and reducing property losses. This model aims to provide precise auxiliary decision-making support for fire alarm operators through an intelligent question-and-answer system, thereby optimizing resource allocation and the emergency response process. Specific application scenarios include the following aspects:

[0071] First of all, the model can be applied to the rapid assessment in the initial stage of a fire. By automatically collecting key on-site information, etc., the model can generate a detailed report on the fire situation in real time and provide an initial fire level prediction for the alarm receivers, which helps to quickly understand the severity of the fire and provides data support for subsequent rescue operations.

[0072] Secondly, the application of the model can optimize the allocation of fire rescue resources. By analyzing the collected data, the model can provide suggestions for the alarm center on how to optimize rescue resources such as personnel, vehicles, and equipment, etc., to ensure the most effective use of rescue resources, shorten the response time, and improve the rescue efficiency.

[0073] In addition, the long-term application of the model is expected to accumulate a large amount of valuable fire emergency response data, providing important references for future fire early warning and risk assessment. Through deep learning of historical cases, the model can continuously optimize its prediction accuracy and response efficiency, providing strong technical support for fire emergency management.

[0074] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A fire alarm response method, characterized in that: The following steps are involved: In response to an alarm communication, extracting alarm elements based on preset query rules, and generating a pre-alarm work order including the alarm elements; The alarm elements are processed using a preset judgment model. Specifically, a fire alarm dialogue model with Transformer as the basic architecture is constructed, and the alarm question and answer is regarded as a sequence generation problem. The fire situation that should be asked next is inferred in an autoregressive manner based on the fire situation that has been obtained, until the model believes that enough information has been obtained to judge the fire level. The above steps are performed multiple times to gradually deepen the understanding of the fire scene situation, and on this basis, a real-time fire level assessment is generated.

2. A fire alarm response method as claimed in claim 1, characterized in that: The model uses Transformer as the basic architecture, regards the alarm question-answering problem as a sequence generation problem, and uses autoregression to infer the fire situation that should be asked next based on the fire situation that has been obtained, until the model believes that enough information has been obtained to judge the fire level.

3. A fire alarm response method as claimed in claim 1, characterized in that: The fire alarm dialogue model uses the explicit state of the fire as the model input to generate the implicit state through training, wherein the "END" symbol represents the end signal, indicating the end state generation and switching to fire classification reasoning; during the model training process, the prediction symbol [S] sequence is used to realize the autoregressive generation of the implicit state sequence, which can be regarded as learning multiple stages of state reasoning at one time; the level prediction symbol [D] is used to learn and generate the target level based on all fire status information, wherein the explicit state refers to the necessary inquiry and status response at the alarm receiving stage, and the invisible state refers to the optional inquiry and status response at the alarm receiving stage.

4. A fire alarm response method as claimed in claim 1, characterized in that: The model first obtains the explicit state of the fire, calculates the probability distribution of the implicit state by splicing the state prediction symbol [S], and inquires about the state with the highest probability in the implicit state. At this time, the user will reply True, False or uncertain to the question; for uncertain information, the above steps will be repeated to find the next state with the highest probability and ask the user; according to the user's True or False answer, the model will splice the newly obtained invisible state, and then use the state prediction symbol to predict the next state; when the end signal is inferred from the state symbol, or when the confidence of the predicted state probability is lower than the threshold, the model will stop asking and switch to level judgment, and predict the target fire level by splicing the level prediction symbol [D].

5. A fire alarm response method as claimed in claim 3, characterized in that: The explicit status of a fire includes the number of trapped persons and the burning area. One or more of the following explicit characteristics may be added as needed: the name of the caller, the caller's telephone number, the time of the call, the address of the call, the location of the caller, the burning floor, the number of trapped persons, the number of injured persons and the degree of injury, and the smoke conditions.

6. A fire alarm response method as claimed in claim 3, characterized in that: The latent states include open flames, smoke, white smoke, black smoke, explosion hazards, surrounding combustibles, spread, high-rise buildings, collapse, whether fire-fighting facilities are complete and whether water sources are sufficient. The latent states are adjusted according to the climate zone, season, time period, building density, and safety alert level of the area.

7. A fire alarm response method as claimed in claim 1, characterized in that: The alarm classification results include repeated alarms, non-repeated alarms, ordinary alarms and important alarms; when the alarm classification result is determined to be a repeated alarm, a secondary confirmation of the repeated alarm is performed, and when an affirmative response to the repeated alarm is obtained, the current alarm communication is terminated.

8. A fire alarm response method as claimed in claim 1, characterized in that: Based on the interactive voice response component, the question and answer are transferred to obtain the question and answer results. When the question and answer results meet the preset conditions, the person who receives the fire pre-alarm voice communication extracts the alarm elements based on the preset inquiry rules; when the question and answer results do not meet the preset conditions, the manual response is initiated.

9. A fire alarm response system, characterized in that: Includes the following modules: Alarm center module: used to receive alarm information and early warning dangerous situation information, extract alarm elements based on preset query rules, and generate pre-alarm work orders including the alarm elements; Fire alarm dialogue model module: a fire alarm dialogue model with Transformer as the basic architecture is pre-installed; it is used to process the alarm elements received by the alarm center module, regard the alarm questions and answers as sequence generation problems, and infer the fire situation that should be asked next based on the fire situation obtained in an autoregressive manner, until the model believes that enough information has been obtained to judge the fire level; perform the above steps multiple times to gradually deepen the understanding of the fire scene situation, and generate a real-time fire level assessment on this basis; Danger information display module: used to obtain and display the danger scene information in real time using GIS map management technology based on the information of the danger warning module of the alarm center module and the fire alarm dialogue model module; Danger warning module: used to collect danger factors in the urban monitoring area in real time, and then identify and predict the danger in the monitoring area, and then judge the warning danger information and analyze the probability of danger occurring.

10. A non-temporary storage medium, characterized in that: When the instructions in the storage medium are executed by a fire alarm response system as described in claim 9, the device is enabled to execute any one of the fire alarm response methods in claims 1 to 8 above.

Images