Intelligent fire-fighting information management system based on AI voice interaction

Abstract

The invention relates to an intelligent fire-fighting information management system based on AI voice interaction, and relates to the technical field of intelligent fire fighting. The method comprises the steps of training a random forest model by collecting fire-fighting monitoring data, including fire-fighting information and corresponding fire-fighting problem occurrence conditions, in a historical period, and identifying key fire-fighting information causing a fire-fighting problem; according to the invention, the problem of low fire-fighting information management efficiency caused by difficult effective identification of main information causing fire-fighting problems in the prior art is solved. By constructing the Bayesian network model between the key fire-fighting information and the fire-fighting problem occurrence situation, the probability of fire-fighting problem occurrence is predicted according to the key fire-fighting information, the accuracy of fire-fighting information management is improved, and the fire-fighting problem is effectively prevented.

Description

Technical Field

[0001] The present invention belongs to the field of smart fire protection technology and relates to a smart fire protection information management system based on AI voice interaction. Background Art

[0002] In today's society, fire safety is of vital importance. With the continuous advancement of science and technology and the gradual improvement of people's safety awareness, daily fire monitoring work has received widespread attention. In the process of fire monitoring, a large amount of fire information is often generated, which covers various aspects, such as the status of fire protection facilities in the target building or area, the operation of electrical equipment, changes in smoke concentration, temperature fluctuations, etc. These rich fire information plays an irreplaceable role in assisting managers to conduct comprehensive and in-depth fire safety inspections.

[0003] However, there is a serious problem at present. Due to the extremely large amount and variety of fire information involved, it is difficult for existing technical means to effectively identify the main information that causes fire problems (fires). In the face of this massive amount of information, the limitations of technology are gradually highlighted. This not only makes the management of fire information inefficient, but managers need to spend a lot of time and energy to screen and analyze this information, but it is difficult to quickly and accurately extract key content, thus affecting the overall progress and effectiveness of fire management. Moreover, due to the inability to accurately identify key information, it is easy to overlook some potential major fire hazards. Therefore, how to improve the ability to identify fire information and accurately find out the main information that causes fire problems has become a key technical problem that needs to be solved in the current fire protection field. Summary of the invention

[0004] The present invention provides a smart fire information management system based on AI voice interaction, which aims to collect fire monitoring data from historical periods to establish an artificial intelligence model between fire information and fire problems, thereby identifying key fire information that can cause fire problems. The trained artificial intelligence model is used in real-time monitoring of fire information, and provides the dynamic situation of key fire information in real time through voice interaction, thereby improving the management efficiency of fire information.

[0005] The present invention provides an intelligent fire information management system based on AI voice interaction, comprising an information collection module, an information analysis module and a voice interaction module, wherein the information collection module, the information analysis module and the voice interaction module are communicatively connected, wherein: The information collection module is used to monitor and collect fire information of the target building in real time, and the fire information includes on-site image data, environmental data, fire equipment data and heat source monitoring data; The information analysis module is used to analyze the probability of a current fire problem occurring according to the fire information, and includes the following steps: S1. Collect fire monitoring data of historical periods, wherein the fire monitoring data includes fire information and corresponding fire problem occurrences; wherein the fire problem occurrences are divided into two situations, "occurrence" and "non-occurrence" according to whether the fire problem occurs; S2, using fire information as the explanatory variable and the occurrence of fire problems as the response variable, train a random forest model to identify the key fire information that causes fire problems; S3. Establish a Bayesian network model between key fire information and fire problem occurrence, and quantify the probability relationship between the two; S4, extract key fire information from real-time monitored fire information and input it into the Bayesian network model to predict the probability of the current fire problem; The voice interaction module notifies the administrator by voice when the probability of a fire problem is higher than a set threshold, and points out the key fire information causing the fire problem.

[0006] Furthermore, the on-site image data is monitored by installing high-definition cameras at key locations of the target building, which cover entrances, corridors and important equipment areas, and capture the on-site images in real time from different angles; The environmental data include temperature, humidity and smoke concentration; The heat source monitoring data is monitored using infrared thermal imaging technology to detect areas with abnormally high temperatures.

[0007] Furthermore, in step S2, the training of the random forest model comprises the following steps: S21. Data preparation: Standardize the collected fire monitoring data from historical periods, specifically extract key features from on-site image data through image recognition technology; perform numerical standardization on environmental data; encode and convert the equipment status in fire equipment data; and quantify the temperature anomaly areas in heat source monitoring data; S22, parameter setting: determining the key parameters of the random forest model, the key parameters including the number of decision trees, the maximum depth of each decision tree, the number of features when splitting a node, the minimum number of sample splits, and the minimum number of samples for a leaf node; S23, model training: randomly extract samples and features (i.e. key fire information) from the data set to build multiple decision trees; when building each decision tree, randomly select a portion of samples from the training data; then, randomly select a portion of features from all features, and select the best splitting method at each node based on these features to minimize the impurity of the node; repeat this process of building decision trees until the stopping condition is met, and build a random forest consisting of multiple decision trees; S24, Model evaluation: Use a portion of the data that was not involved in the training to evaluate the trained random forest model; S25. Determine key firefighting information: Based on the evaluation results of the random forest model containing different features, rank the importance of the features and determine the firefighting information ranked in the top 50% as key firefighting information.

[0008] Furthermore, in step S23, the decision tree is configured as a classification tree, and the classification tree is pruned by the Gini coefficient.

[0009] Furthermore, in step S24, the trained random forest model is evaluated using a portion of data that has not participated in the training, and the evaluation indicators include accuracy, precision, recall and F1 value.

[0010] Furthermore, in step S3, the Bayesian network model includes the following construction steps: S31, determine the model node: take the key fire information as the parent node, the fire problem occurrence situation as the child node, and construct the directed edge between the parent node and the child node; S32, estimating variable parameters: determining variable parameters according to data characteristics of key fire information and fire problem occurrence, wherein the variable parameters of key fire information are determined by dividing different intervals according to numerical values, and the variable parameters of fire problem occurrence are "occurrence" and "non-occurrence"; S33, obtaining a conditional probability table: determining a conditional probability table of the parent node to the child node variable parameter according to the direction of the directed edge; S34. Model evaluation: Use historical data that is not involved in model construction for testing, input the data into the Bayesian network model, and compare the differences between the model prediction results and the actual situation.

[0011] Furthermore, in step S32, the variable parameters of the key fire information are determined by dividing the values ​​into different intervals, and a clustering algorithm is specifically used to divide the key fire information into clusters of different intervals, and each cluster represents a variable parameter.

[0012] Furthermore, the information collection module has a fixed collection frequency, and does not collect information when it is detected that the fire protection information has not changed.

[0013] Furthermore, after the information analysis module, random forest model and Bayesian network model are constructed, they are stored in the system in an offline manner and are continuously optimized and updated according to actual fire monitoring conditions.

[0014] Furthermore, the voice interaction module can also visually display fire information according to the administrator's voice instructions.

[0015] Beneficial effects of the present invention: (1) By collecting fire monitoring data from historical periods, including fire information and corresponding fire problems, the random forest model is trained to identify the key fire information that causes fire problems; this solves the problem that existing technologies are difficult to effectively identify the main information that causes fire problems, resulting in inefficient fire information management.

[0016] (2) By constructing a Bayesian network model between key fire information and the occurrence of fire problems, the probability of fire problems occurring can be predicted based on key fire information, thereby improving the accuracy of fire information management and effectively preventing the occurrence of fire problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a structural diagram of an intelligent fire information management system based on AI voice interaction in the present invention.

[0019] Figure 2 The present invention is a flowchart of training a random forest model in one embodiment of the present invention.

[0020] Figure 3 The flowchart of constructing a Bayesian network model in one embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0022] See also Figure 1-Figure 3 The present application provides a smart fire information management system based on AI voice interaction, including an information collection module, an information analysis module and a voice interaction module, wherein the information collection module, the information analysis module and the voice interaction module are communicatively connected, wherein: The information collection module is used to monitor and collect fire information of the target building in real time, and the fire information includes on-site image data, environmental data, fire equipment data and heat source monitoring data; In this embodiment, the information collection module plays a vital role in the entire fire information management system. This module is mainly responsible for real-time monitoring and collecting fire information of the target building. Among them, the on-site image data is mainly monitored by installing high-definition cameras at key locations of the target building. These cameras can cover entrances and exits, corridors, important equipment areas, etc., and shoot the on-site scenes in real time from different angles. The camera is connected to the system network and transmits the collected image data to the information management center continuously so that managers can check the on-site situation at any time.

[0023] Environmental data monitoring relies on various high-precision sensors. Temperature sensors are distributed in different areas to sense changes in ambient temperature in real time; humidity sensors can accurately measure air humidity; and smoke sensors can sensitively detect smoke concentration in the air. These sensors transmit the monitored data to the system via wireless or wired communication, providing an important basis for judging fire risks.

[0024] The monitoring of fire equipment data is achieved by installing intelligent monitoring devices on the fire equipment. For example, a pressure sensor is installed on the fire extinguisher to monitor the internal pressure of the fire extinguisher in real time; a water pressure sensor is installed on the fire hydrant to keep track of the water supply pressure of the fire hydrant at any time; for the automatic sprinkler fire extinguishing system, the valve status, water flow rate and other parameters are monitored to ensure its normal operation. These monitoring devices transmit data to the system so that managers can understand the status of fire equipment in a timely manner.

[0025] Heat source monitoring data is mainly monitored using infrared thermal imaging technology. Installing infrared thermal imaging equipment in the target building can quickly detect areas with abnormally high temperatures, which are possible heat sources. These devices transmit thermal imaging data to the system. Once an abnormal heat source is found, the system can issue an early warning in time to remind relevant personnel to take measures. Through real-time monitoring and collection of these rich and diverse fire information, the information collection module provides a comprehensive and accurate data foundation for the entire smart fire information management system, enabling the system to analyze, warn and make decisions more efficiently, thereby providing strong protection for the fire safety of the target building.

[0026] Furthermore, the information collection module has a fixed collection frequency, and does not collect information when it is detected that the fire protection information has not changed.

[0027] In this embodiment, the setting that the information collection module has a fixed collection frequency is reasonable in many aspects. On the one hand, the fixed collection frequency can ensure that fire information can be obtained more regularly under normal circumstances, so as to grasp the dynamic changes of fire conditions in a timely manner. For example, collecting key fire information such as temperature, smoke concentration, and fire equipment status at a certain time interval can provide a stable data source for subsequent analysis and decision-making.

[0028] At the same time, when the fire information is monitored to be unchanged, the design of not collecting information can effectively save system resources and reduce the burden of data processing. In actual situations, if the fire information remains stable, continuous information collection will not only consume unnecessary energy and network bandwidth, but also increase the pressure of data storage and processing. Through this intelligent judgment mechanism, the operating efficiency of the system can be improved without affecting the effective monitoring of the fire situation.

[0029] In addition, this design also helps to reduce data redundancy. When the fire information does not change, it is meaningless to repeatedly collect the same data. Not collecting information can avoid storing a large amount of duplicate data, making data management more concise and efficient, and providing more refined and valuable information for subsequent data analysis and model building.

[0030] The information analysis module is used to analyze the probability of a current fire problem occurring according to the fire information, and includes the following steps: S1. Collect fire monitoring data of historical periods, wherein the fire monitoring data includes fire information and corresponding fire problem occurrences; wherein the fire problem occurrences are divided into two situations, "occurrence" and "non-occurrence" according to whether the fire problem occurs; In this embodiment, in the fire information management system, it is of great significance to collect fire monitoring data of historical periods. These fire monitoring data cover a wide range of content, including fire information such as on-site image data, environmental data, fire equipment data and heat source monitoring data, etc.; on the other hand, it also clearly records the occurrence of corresponding fire problems.

[0031] For the occurrence of fire problems, the data is strictly divided into two distinct situations: "occurrence" and "non-occurrence" according to the actual situation. Through this classification method, we can more clearly understand the probability and pattern of fire problems under different fire information conditions. For example, when environmental data such as temperature, humidity, and smoke concentration are collected within a certain period of time, if the corresponding fire problem occurrence is "occurrence", then the potential correlation between these environmental data and the occurrence of fire can be deeply analyzed. Similarly, for fire equipment data, if the fire equipment fails or is abnormal in certain cases, and the corresponding fire problem occurrence is "occurrence", the causal relationship between the status of the fire equipment and the fire can be further studied.

[0032] By comprehensively collecting and classifying fire monitoring data from historical periods, valuable experience references can be provided for the fire information management system. The system can use these data for in-depth data analysis and pattern recognition, thereby better predicting possible fire problems in the future, taking effective preventive measures in advance, and improving the level and efficiency of fire safety management.

[0033] S2, using fire information as the explanatory variable and the occurrence of fire problems as the response variable, train a random forest model to identify the key fire information that causes fire problems; In fire information management, using fire information as the explanatory variable and the occurrence of fire problems as the response variable, training a random forest model is an effective way to identify the key fire information that causes fire problems.

[0034] First, the collected fire monitoring data from the historical period is sorted and preprocessed. The on-site image data, environmental data, fire equipment data, and heat source monitoring data in the fire information are digitized and standardized so that the model can better understand and process these data. At the same time, the occurrence of fire problems is clearly marked as a binary variable of "occurred" or "did not occur".

[0035] The preprocessed data is then used to train a random forest model. Random forest is an ensemble learning method that improves the accuracy and stability of predictions by building multiple decision trees and combining their predictions. During the training process, the random forest model automatically learns the complex relationship between fire information and the occurrence of fire problems.

[0036] As the model is trained, it will gradually identify which firefighting information has a key impact on the occurrence of firefighting problems. For example, it may be found that certain temperature ranges, excessive smoke concentrations, and faulty conditions of firefighting equipment are closely related to the occurrence of firefighting problems. By analyzing the importance of each explanatory variable in the random forest model, the key firefighting information that causes firefighting problems can be determined.

[0037] By using the trained random forest model, in actual fire monitoring, we can pay more attention to key fire information, take preventive measures in time, and improve the efficiency and accuracy of fire safety management. At the same time, as new data continues to accumulate, the model can be updated and optimized regularly to ensure that it can always accurately identify key fire information and provide continuous protection for fire safety.

[0038] Furthermore, in step S2, the training of the random forest model comprises the following steps: S21. Data preparation: First, comprehensively organize the fire monitoring data collected in historical periods. Standardize fire information, including on-site image data, environmental data, fire equipment data, and heat source monitoring data. For on-site image data, key features can be extracted through image recognition technology, such as the shape and color of smoke and other characteristic values; the temperature, humidity, smoke concentration, etc. in the environmental data are numerically standardized; the equipment status in the fire equipment data is coded and converted; the temperature abnormality areas in the heat source monitoring data are quantified (such as area, volume). At the same time, clearly mark the occurrence of fire problems as "occurred" or "did not occur" to ensure the accuracy and consistency of the data and provide a high-quality data foundation for subsequent model training.

[0039] S22. Parameter setting: Determine the key parameters of the random forest model. For example, the number of decision trees generally needs to be reasonably set according to the data scale and computing resources, usually between dozens and hundreds of trees. Set the maximum depth of each decision tree to control the complexity of the tree and avoid overfitting. Determine the number of features to consider when splitting a node, usually around the square root of the total number of features. In addition, other parameters such as the minimum number of sample splits, the minimum number of leaf node samples, etc. can also be set to optimize the performance of the model.

[0040] S23. Model training: Use the prepared data to train the random forest model. Randomly extract samples and features from the data set to build multiple decision trees. When building each decision tree, randomly select a portion of samples from the training data (with replacement sampling), called the bootstrap sample set. Then, randomly select a portion of features from all features, and select the best splitting method at each node based on these features to minimize the impurity of the node. The measure of impurity can be Gini impurity or information entropy, etc. Repeat this process until the stopping condition is met, such as the maximum depth of the tree reaches a preset value or the number of samples in the node is less than a certain threshold. In this way, a random forest consisting of multiple decision trees is constructed.

[0041] Furthermore, in step S23, the decision tree is configured as a classification tree, and the classification tree is pruned by the Gini coefficient.

[0042] In this embodiment, setting the decision tree as a classification tree plays an important role. The classification tree can classify and predict the occurrence of fire problems based on key fire information, such as dividing fire problems into two categories of "occurrence" and "non-occurrence" according to temperature, smoke concentration, fire equipment status, etc. The Gini coefficient is used to measure the impurity of the classification tree node, that is, the probability of two sample categories being inconsistent when randomly selected from a node. The smaller the Gini coefficient, the purer the sample category. It can be used to select the best split feature, evaluate the splitting effect of different features to select the feature with the smallest Gini coefficient for splitting. At the same time, pruning the classification tree by the Gini coefficient can prevent overfitting, and check the nodes from the bottom of the tree step by step upward. If changing the node to a leaf node can reduce the overall Gini coefficient, the pruning operation is performed. The random forest model constructed in this way is more accurate, concise and has strong generalization ability, providing strong support for fire management to identify key fire information.

[0043] S24. Model evaluation: Use a portion of the data that was not involved in the training to evaluate the trained random forest model. Some common evaluation indicators can be calculated, such as accuracy, precision, recall, F1 value, etc. The performance of the model can be judged by comparing the difference between the model's prediction results and the actual situation in the two fire problems of "occurrence" and "non-occurrence". If the evaluation results are not ideal, the parameters can be adjusted and retrained until the model reaches a satisfactory performance level.

[0044] S25. Determine key firefighting information: Based on the evaluation results of the random forest model containing different features, rank the importance of the features and determine the firefighting information ranked in the top 50% as key firefighting information.

[0045] S3. Establish a Bayesian network model between key fire information and fire problem occurrence, and quantify the probability relationship between the two; In this embodiment, the Bayesian network model constructed between key fire information and the occurrence of fire problems is an important innovative tool in the field of fire management: Structurally, a Bayesian network is a directed acyclic graph consisting of a series of nodes and directed edges. Among them, the key fire information nodes represent the factors that have a critical impact on fire problems carefully identified by the random forest model. These factors cover many aspects, such as specific environmental parameters such as temperature, humidity, smoke concentration, etc.; the status of fire equipment, including whether the pressure of the fire extinguisher is normal, whether the fire hydrant can supply water normally, whether the automatic sprinkler system is in good working condition, etc. The fire problem occurrence node clearly indicates whether the fire problem has occurred, and there are only two states: "occurred" and "not occurred". Directed edges play a key connecting role in the model, which represents the causal relationship or influence relationship between nodes. The determination of this relationship is based on in-depth expertise in the field of fire protection and careful analysis of a large amount of historical data. For example, if a specific fire equipment failure tends to increase the probability of fire problems, then a directed edge will be drawn from the fire equipment status node to the fire problem occurrence node.

[0046] In terms of parameter learning, the parameters of the Bayesian network are mainly the conditional probability distribution of each node. Through comprehensive statistical analysis of rich historical fire monitoring data, these parameters can be estimated more accurately. For example, for a key fire information node representing temperature, the probability of fire problems occurring in different temperature ranges can be calculated by counting the ratio of the number of fire problems occurring in different temperature ranges in the historical data to the total number of data in the temperature range.

[0047] In actual fire management applications, this Bayesian network model plays a huge role. It can quickly calculate the probability of fire problems based on the key fire information monitored in real time. This provides extremely important decision-making support for fire management personnel. If the calculated probability is high, managers can immediately take emergency measures, such as strengthening patrols, launching emergency plans in advance, and notifying relevant personnel to prepare for evacuation, thereby effectively reducing the risk of fire. At the same time, the model can also analyze the interaction between different key fire information. When multiple key fire information are abnormal at the same time, the model can comprehensively consider these factors and give a more accurate prediction of the probability of fire problems through complex probability calculations.

[0048] The model has many significant advantages and characteristics. First, it can clearly quantify the probabilistic relationship between key fire information and the occurrence of fire problems, making fire management more scientific and accurate. No longer relying solely on empirical judgment, there is a specific numerical basis to guide decision-making. Secondly, Bayesian networks can handle uncertainty well. In the field of fire protection, many situations are uncertain, and this model uses probability distribution to represent the possibility of various situations, which is more in line with actual conditions. Furthermore, the model establishment process fully combines the professional knowledge in the field of fire protection and a large amount of historical data, which makes the model highly reliable and practical. Finally, with the continuous accumulation of new fire monitoring data, the model can regularly update parameters to always maintain accuracy and effectiveness and adapt to the ever-changing fire environment.

[0049] Furthermore, in step S3, the Bayesian network model includes the following construction steps: S31, determine the model node: take the key fire information as the parent node, the fire problem occurrence situation as the child node, and construct the directed edge between the parent node and the child node; S32, estimating variable parameters: determining variable parameters according to data characteristics of key fire information and fire problem occurrence, wherein the variable parameters of key fire information are determined by dividing different intervals according to numerical values, and the variable parameters of fire problem occurrence are "occurrence" and "non-occurrence"; In this embodiment, in the process of estimating the variable parameters, it is necessary to fully consider the data characteristics of the key fire information and the occurrence of fire problems. For the key fire information, since its variable parameters are determined by dividing different intervals according to the numerical value, the key fire information must first be analyzed in detail. For example, for the key fire information of temperature, different temperature intervals can be divided according to the actual situation and historical data. The temperature can be divided into low temperature zone, medium temperature zone and high temperature zone, etc. Each temperature interval corresponds to a different degree of fire risk. Similarly, similar interval division can be performed for key fire information such as smoke concentration and humidity. In this way, the degree of influence of key fire information in different numerical ranges on the occurrence of fire problems can be more accurately reflected. For the occurrence of fire problems, its variable parameters are relatively simple and clear, with only two states of "occurrence" and "non-occurrence". These two states directly reflect the final result of the fire protection system. When determining the variable parameters of the occurrence of fire problems, the probability of "occurrence" and "non-occurrence" of fire problems under different key fire information conditions can be calculated through statistical analysis of historical data.

[0050] Furthermore, in step S32, the variable parameters of the key fire information are determined by dividing the values ​​into different intervals, and a clustering algorithm is specifically used to divide the key fire information into clusters of different intervals, and each cluster represents a variable parameter.

[0051] In this embodiment, the benefits of dividing the key fire information by clustering algorithm are: first, it can help us better understand the distribution characteristics of the key fire information and discover the potential patterns and laws in the data. Second, the clusters after division can be used as variable parameters to more accurately reflect the relationship between the key fire information and the occurrence of fire problems. Finally, this method has certain flexibility and adaptability, and the clustering parameters and methods can be adjusted according to the actual situation to obtain better results.

[0052] S33, obtaining a conditional probability table: determining a conditional probability table of the parent node to the child node variable parameter according to the direction of the directed edge; In this embodiment, for each interval state of the key fire information (parent node) and the "occurrence" and "non-occurrence" states of the fire problem occurrence (child node), the following formula can be used to calculate the conditional probability: Conditional probability P(child node state | parent node state) = the number of times the child node is in a specific state when the parent node is in that state ÷ the total number of times the parent node is in that specific state.

[0053] For example, for the key fire information of temperature, it is assumed that it is divided into three intervals: low temperature, medium temperature and high temperature. In the statistical historical data, when the temperature is in the low temperature interval, the number of times the fire problem occurs is A, and the total number of times the temperature is in the low temperature interval is B. Then the conditional probability of the fire problem occurring when the temperature is in the low temperature interval is A / B; similarly, the conditional probability of the fire problem not occurring when the temperature is in the low temperature interval can be calculated, as well as the various conditional probabilities of the fire problem occurring and not occurring in the medium temperature and high temperature intervals.

[0054] A similar method is used to calculate other key fire information, and finally a complete conditional probability table of parent node to child node variable parameters is obtained.

[0055] S34. Model evaluation: Use historical data that is not involved in model construction for testing, input the data into the Bayesian network model, and compare the differences between the model prediction results and the actual situation.

[0056] S4, extract key fire information from real-time monitored fire information and input it into the Bayesian network model to predict the probability of the current fire problem; Furthermore, after the information analysis module, random forest model and Bayesian network model are constructed, they are stored in the system in an offline manner and are continuously optimized and updated according to actual fire monitoring conditions.

[0057] In this embodiment, offline storage ensures the stability and security of the model. The model is stored inside the system, free from external network interference and potential security risks, and can be called quickly and reliably when analysis is needed. Even in the case of unstable or interrupted network connection, the system can still rely on the locally stored model to analyze fire information and ensure the continuity of fire management work.

[0058] Secondly, continuous optimization and updating based on actual fire monitoring conditions is the key to maintaining the accuracy and effectiveness of the model. Over time, the fire environment may change, new fire problems may emerge, or the characteristics of key fire information may change. By continuously collecting actual fire monitoring data and using this data to optimize and update the model, the model can be better adapted to the changing fire situation.

[0059] The voice interaction module notifies the administrator by voice when the probability of a fire problem is higher than a set threshold, and points out the key fire information causing the fire problem.

[0060] In this embodiment, the voice interaction module plays a vital role in the fire management system. When the Bayesian network model predicts that the probability of a fire problem is higher than the set threshold, the module will start quickly and notify the administrator by voice. First, the voice interaction module will convey emergency information to the administrator in a clear and unambiguous voice broadcast. For example, "Dear administrator, please note! The current system detects that the probability of a fire problem is higher than the set threshold." Such a voice notification can alert the administrator at the first time and ensure that they can take timely action.

[0061] At the same time, the module will further point out the key fire information that causes fire problems. If the risk is increased due to excessive temperature, the voice interaction module will broadcast "The key fire information is that the temperature exceeds the safe range, please check the relevant areas in time." If the risk is caused by fire equipment failure, it will clearly point out "Xuan fire equipment has failed, which may cause fire problems. Please deal with it as soon as possible." In this way, administrators can quickly understand the key to the problem and take targeted measures to investigate and deal with it.

[0062] The timeliness and accuracy of the voice interaction module can greatly improve the efficiency of fire management, help administrators make correct decisions quickly in emergency situations, reduce the probability of fire accidents, and protect people’s lives and property.

[0063] Furthermore, the voice interaction module can also visually display fire information according to the administrator's voice instructions.

[0064] In this embodiment, the role of the voice interaction module in fire management is further expanded, and fire information can be visually displayed according to the administrator's voice instructions, bringing great convenience to fire management.

[0065] First, when the administrator issues a voice command, the voice interaction module uses advanced voice recognition technology to accurately understand the command content. For example, if the administrator says "show the temperature information of the current area", the module can quickly recognize and execute the corresponding operation.

[0066] There are many ways to visualize fire information. Temperature information can be displayed in the form of a chart, with the horizontal axis being the time axis and the vertical axis being the temperature value, clearly showing the trend of temperature changes over time. Different colors can represent different temperature ranges, allowing administrators to see at a glance whether the current temperature is within a safe range.

[0067] Dynamic graphics can be used to display smoke concentration information. For example, different shades of colors can be used to represent different smoke concentration levels. When the smoke concentration increases, the color gradually deepens, and the specific concentration value can be displayed in combination with numbers.

[0068] Fire equipment status information can be displayed in list or icon form. If the equipment is operating normally, the icon is displayed in green; if the equipment fails or is abnormal, the icon turns red and flashes to attract the attention of the administrator. Click the icon to view detailed information about the equipment, such as equipment type, location, fault type, etc.

[0069] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. An intelligent fire information management system based on AI voice interaction, characterized by: It includes an information collection module, an information analysis module and a voice interaction module, which are communicatively connected to each other, wherein: The information collection module is used to monitor and collect fire information of the target building in real time, and the fire information includes on-site image data, environmental data, fire equipment data and heat source monitoring data; The information analysis module is used to analyze the probability of a current fire problem occurring according to the fire information, and includes the following steps: S1. Collect fire monitoring data of historical periods, wherein the fire monitoring data includes fire information and corresponding fire problem occurrences; wherein the fire problem occurrences are divided into two situations, "occurrence" and "non-occurrence" according to whether the fire problem occurs; S2, using fire information as the explanatory variable and the occurrence of fire problems as the response variable, train a random forest model to identify the key fire information that causes fire problems; S3. Establish a Bayesian network model between key fire information and fire problem occurrence, and quantify the probability relationship between the two; S4, extract key fire information from real-time monitored fire information and input it into the Bayesian network model to predict the probability of the current fire problem; The voice interaction module notifies the administrator by voice when the probability of a fire problem is higher than a set threshold, and points out the key fire information causing the fire problem.

2. According to claim 1, a smart fire information management system based on AI voice interaction is characterized in that: The on-site image data is monitored by installing high-definition cameras at key locations of the target building, which cover entrances, corridors, and important equipment areas, and capture the on-site images in real time from different angles; The environmental data include temperature, humidity and smoke concentration; The heat source monitoring data is monitored using infrared thermal imaging technology to detect areas with abnormally high temperatures.

3. According to claim 1, a smart fire information management system based on AI voice interaction is characterized in that: In step S2, the training of the random forest model includes the following steps: S21. Data preparation: Standardize the collected fire monitoring data from historical periods, specifically extract key features from on-site image data through image recognition technology; perform numerical standardization on environmental data; encode and convert the equipment status in fire equipment data; and quantify the temperature anomaly areas in heat source monitoring data; S22, parameter setting: determining the key parameters of the random forest model, the key parameters including the number of decision trees, the maximum depth of each decision tree, the number of features when splitting a node, the minimum number of sample splits, and the minimum number of samples for a leaf node; S23, model training: randomly extract samples and features from the data set to build multiple decision trees; when building each decision tree, randomly select a part of the samples from the training data; then, randomly select a part of the features from all the features, and select the best splitting method at each node based on these features to minimize the impurity of the node; repeat this process of building decision trees until the stopping condition is met, and build a random forest composed of multiple decision trees; S24, Model evaluation: Use a portion of the data that was not involved in the training to evaluate the trained random forest model; S25. Determine key firefighting information: Based on the evaluation results of the random forest model containing different features, rank the importance of the features and determine the firefighting information ranked in the top 50% as key firefighting information.

4. According to claim 3, the intelligent fire information management system based on AI voice interaction is characterized in that: In step S23, the decision tree is configured as a classification tree, and the classification tree is pruned using the Gini coefficient.

5. According to claim 3, the intelligent fire information management system based on AI voice interaction is characterized in that: In step S24, the trained random forest model is evaluated using a portion of data that has not participated in the training, and the evaluation indicators include accuracy, precision, recall and F1 value.

6. According to claim 1, the intelligent fire information management system based on AI voice interaction is characterized in that: In step S3, the Bayesian network model includes the following construction steps: S31, determine the model node: take the key fire information as the parent node, the fire problem occurrence situation as the child node, and construct the directed edge between the parent node and the child node; S32, estimating variable parameters: determining variable parameters according to data characteristics of key fire information and fire problem occurrence, wherein the variable parameters of key fire information are determined by dividing different intervals according to numerical values, and the variable parameters of fire problem occurrence are "occurrence" and "non-occurrence"; S33, obtaining a conditional probability table: determining a conditional probability table of the parent node to the child node variable parameter according to the direction of the directed edge; S34. Model evaluation: Use historical data that is not involved in model construction for testing, input the data into the Bayesian network model, and compare the differences between the model prediction results and the actual situation.

7. According to claim 1, a smart fire information management system based on AI voice interaction is characterized in that: In step S32, the variable parameters of the key fire information are determined by dividing the values ​​into different intervals. Specifically, a clustering algorithm is used to divide the key fire information into clusters of different intervals, and each cluster represents a variable parameter.

8. According to claim 1, a smart fire information management system based on AI voice interaction is characterized in that: The information collection module has a fixed collection frequency, and does not collect information when it is detected that the fire protection information has not changed.

9. According to claim 1, a smart fire information management system based on AI voice interaction is characterized in that: After being constructed, the information analysis module, random forest model and Bayesian network model are stored in the system in an offline manner and are continuously optimized and updated according to actual fire monitoring conditions.

10. The intelligent fire information management system based on AI voice interaction according to claim 1 is characterized in that: The voice interaction module can also visually display fire information according to the administrator's voice instructions.

Images