Fire-fighting monitoring alarm method and system based on Internet of Things sensor

Abstract

The invention relates to the technical field of fire-fighting alarm, in particular to a fire-fighting monitoring alarm method and system based on an Internet of Things sensor. The invention discloses a fire-fighting monitoring alarm system based on an Internet of Things sensor. The fire-fighting monitoring alarm system comprises a fire preliminary judgment module, a voting calculation module and a fire judgment module. According to the invention, voting is carried out on whether a fire occurs through all the fire alarm sensors around the target fire alarm sensor, information of all the fire alarm sensors around the target fire alarm sensor can be fused, and a fire false alarm condition caused by a sensor fault or a sensor failure is avoided. And the fire alarm system is directly transformed on the basis of a traditional fire alarm system, so that the cost is lower compared with the cost of reconstructing a set of intelligent fire-fighting system.

Description

Technical Field

[0001] The present invention relates to the technical field of fire alarm technology, and in particular to a fire monitoring alarm method and system based on an Internet of Things sensor. Background Art

[0002] Fire safety is related to social stability and the safety of life and property. With the popularization of digitalization, the construction of smart fire protection is also progressing steadily. However, the current construction of smart fire protection is mainly focused on the development of intelligent sensors, cloud platform development and application software development. There are few intelligent transformations of existing traditional fire alarm systems. In addition, due to the long service life of fire alarm sensors in existing traditional fire alarm systems, sensor failure or sensor failure may occur, resulting in false alarms or missed alarms of fires. Summary of the invention

[0003] The present invention uses all the fire alarm sensors around the target fire alarm sensor to vote on whether a fire has occurred. It can integrate the information of all the fire alarm sensors around the target fire alarm sensor, thereby avoiding false fire alarms caused by sensor malfunction or sensor failure, and more accurately judge the fire situation. It also directly transforms the traditional fire alarm system, which is less costly than rebuilding a set of smart fire protection systems.

[0004] The present invention provides a fire monitoring and alarm method based on an Internet of Things sensor, comprising: Step S1: for any fire alarm sensor, obtain sensor data value through the fire alarm sensor, and determine whether a fire situation occurs based on the sensor data value. If a fire situation occurs, record the fire alarm sensor as the target fire alarm sensor and proceed to step S2; if no fire situation occurs, continue to obtain sensor data value through the fire alarm sensor; Step S2: all fire alarm sensors within a preset distance from the target fire alarm sensor are obtained through the Internet of Things and recorded as voting nodes. The voting value corresponding to the voting node is weightedly calculated according to the sensor data change difference, consistency coefficient and change correlation coefficient corresponding to the voting node. The first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the weighted sensor data change difference, consistency coefficient and change correlation coefficient are optimized by the particle swarm optimization algorithm improved based on environmental perception. The voting operation corresponding to the voting node is completed. Each time the voting node performs a voting value calculation, the total number of votes corresponding to the voting node is increased by one. The voting value represents the confidence of the voting node in the occurrence of a fire. Step S3: perform weighted average calculation on the voting values ​​corresponding to all voting nodes to obtain the total voting value. The voting weight in the weighted average calculation process is determined by a weight adaptive adjustment model. The input data of the weight adaptive adjustment model are the historical accuracy rate corresponding to the voting node and the environmental data set in the current space. The historical accuracy rate is the ratio of the accurate number of times corresponding to the voting node to the total number of votes. The weight adaptive adjustment model is established based on the BP neural network. The improvement lies in setting a physical rule attention layer to compare the total voting value with the judgment threshold. If the total voting value is higher than the judgment threshold, a fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting value higher than the judgment threshold is added by one; if the total voting value is lower than the judgment threshold, no fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting value higher than the judgment threshold is added by one.

[0005] As a preferred aspect, the voting value corresponding to the voting node is weightedly calculated according to the sensor data change difference, consistency coefficient and change correlation coefficient corresponding to the voting node, specifically including the following steps: For any voting node, all sensor data values ​​corresponding to a preset time length are collected from the current time point forward, and the timestamp is used as the independent variable and the sensor data value is used as the dependent variable to fit the sensor data value change curve; cluster analysis is performed on the sensor data value change curves corresponding to all voting nodes, and the cluster center corresponding to the largest cluster is selected as the target sensor data value change curve, and in the cluster analysis process, the distance between any two sensor data value change curves is the correlation coefficient between the two sensor data value change curves; For any voting node, calculate the range of the corresponding sensor data value change curve and record it as the sensor data change difference F; calculate the correlation coefficient between the sensor data value change curve corresponding to the voting node and the target sensor data value change curve and record it as the consistency coefficient D; calculate the correlation coefficient between the sensor data value change curve corresponding to the voting node and the sensor data value change curve corresponding to the target fire alarm sensor and record it as the change correlation coefficient E; calculate the voting value Q=W1F+W2D+W3E corresponding to the voting node, where W1, W2 and W3 are the first weight coefficient, the second weight coefficient and the third weight coefficient respectively, and satisfy W1+W2+W3=1.

[0006] As a preferred aspect, the voting values ​​corresponding to all voting nodes are weighted averaged to obtain the total voting value. The voting weights in the weighted average calculation process are adaptively adjusted according to the historical accuracy rates corresponding to all voting nodes and the environmental data set in the current space, specifically including the following steps: The sensor data value corresponding to the target fire alarm sensor, the environmental data set in the current space and the historical accuracy rate corresponding to the voting node are added to the end of the sensor data value corresponding to each voting node at the current time point, and a voting weight analysis vector is constructed. The voting weight analysis vectors corresponding to all voting nodes are combined into a voting weight analysis data set. The voting weight analysis data set is processed through a weight adaptive adjustment model to output a voting weight set, which includes the voting weight corresponding to each voting node.

[0007] As a preferred aspect, it includes an input layer, a physical rule attention layer, a hidden layer and an output layer, wherein the input layer is used to receive a voting weight analysis data set, wherein the physical rule attention layer is used to establish a physical law association matrix based on the voting nodes, and perform a self-attention mechanism on the voting weight analysis data set through the physical law association matrix to construct a physical fusion voting weight analysis data set, the hidden layer is used to perform full connection processing on the physical fusion voting weight analysis data set, and the output layer is used to output the voting weight set.

[0008] As a preferred aspect, the training of the weight adaptive adjustment model specifically includes the following steps: Obtain weight adaptive adjustment training samples, which include voting weight analysis data sets; combine all weight adaptive adjustment training samples into a weight adaptive adjustment training set, train the weight adaptive adjustment model through the weight adaptive adjustment training set, and determine whether the number of training times reaches a preset number of times. If the number of training times reaches the preset number of times, output the trained weight adaptive adjustment model; otherwise, continue to train the weight adaptive adjustment model through the weight adaptive adjustment training set.

[0009] As a preferred aspect, the first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the weighted sensor data change difference, the consistency coefficient and the change correlation coefficient are optimized by an improved particle swarm optimization algorithm based on environmental perception, specifically comprising the following steps: Construct several weight coefficient simulation individuals, and divide all weight coefficient simulation individuals into environmentally sensitive population sets and environmentally stable population sets; Calculate the fitness corresponding to each weight coefficient simulation individual, specifically including the following steps: simulate individuals with weight coefficients in the environmentally sensitive population set, set the corresponding first weight coefficient, second weight coefficient and third weight coefficient through the weight coefficient simulation individual, and then traverse the environmentally sensitive fire analysis data set corresponding to the environmentally sensitive population set, the environmentally sensitive fire analysis data set includes a number of voting nodes under environmental sensitivity, and the fire conditions detected by the target fire alarm sensors corresponding to all voting nodes are true. For each fire analysis data set, calculate the voting value corresponding to each voting node. If the voting value is higher than the judgment threshold, it is regarded as a successful judgment. The ratio of the total number of successful judgments to the total number of voting nodes in the environmentally sensitive fire analysis data set is used as the number of voting nodes in the environmentally sensitive population set. The weight coefficient simulates the fitness of the individual; for the weight coefficient simulation individual in the environmentally stable population set, the first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the individual setting are simulated by the weight coefficient simulation individual, and then the environmentally stable fire analysis data set corresponding to the environmentally stable population set is traversed. The environmentally stable fire analysis data set includes several voting nodes under environmental stability. The fire conditions detected by the target fire alarm sensors corresponding to all voting nodes are true. For each fire analysis data set, the voting value corresponding to each voting node is calculated. If the voting value is higher than the judgment threshold, it is regarded as a successful judgment. The ratio of the total number of successful judgments to the total number of voting nodes in the environmentally stable fire analysis data set is used as the fitness of the individual simulated by the weight coefficient in the environmentally stable population set; The environmentally sensitive population set and the environmentally stable population set are iteratively updated by the particle swarm optimization algorithm, and the number of iterations is recorded. After every N iterations, the weight coefficient simulation individual with the highest fitness in the environmentally sensitive population set and the environmentally stable population set is selected as the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual, and then the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual are cross-fused to construct the third optimal weight coefficient simulation individual, and the third optimal weight coefficient simulation individual is used to replace the weight coefficient simulation individual with the lowest fitness in the environmentally sensitive population set and the environmentally stable population set; the specific method of cross-fusion is: the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual are summed and averaged and then normalized; When the number of iterations reaches the maximum number of iterations, the weight coefficient simulation individual with the highest fitness in the environmentally sensitive population set and the weight coefficient simulation individual with the highest fitness in the environmentally stable population set are cross-fused to obtain the optimal weight coefficient simulation individual, and the first weight coefficient, the second weight coefficient and the third weight coefficient are determined through the optimal weight coefficient simulation individual.

[0010] As a preferred aspect, the cluster analysis adopts k-means.

[0011] The present invention also provides a fire monitoring and alarm system based on an Internet of Things sensor, comprising: A fire preliminary judgment module is used for obtaining sensor data values ​​through the fire alarm sensor for any fire alarm sensor, and judging whether a fire situation has occurred based on the sensor data values. If a fire situation has occurred, the fire alarm sensor is recorded as a target fire alarm sensor; if no fire situation has occurred, the sensor data values ​​continue to be obtained through the fire alarm sensor; The voting calculation module is used to obtain all fire alarm sensors within a preset distance from the target fire alarm sensor through the Internet of Things, and record them as voting nodes. The voting value corresponding to the voting node is determined according to the consistency between the sensor data value corresponding to the voting node and the change of all sensor data values, and the voting operation corresponding to the voting node is completed. The number of votes corresponding to all voting nodes is increased by one, and the voting value represents the confidence of the voting node in the occurrence of the fire. The fire judgment module is used to perform weighted average calculation on the voting values ​​corresponding to all voting nodes to obtain the total voting value. The voting weights in the weighted average calculation process are adaptively adjusted according to the historical accuracy rates corresponding to all voting nodes and the environmental data set in the current space. The historical accuracy rate is the ratio of the accurate number of times corresponding to the voting node to the total number of votes. The total voting value is compared with the judgment threshold. If the total voting value is higher than the judgment threshold, a fire alarm signal is issued, and the accurate number of voting nodes with voting values ​​higher than the judgment threshold is incremented by one; if the total voting value is lower than the judgment threshold, no fire alarm signal is issued, and the accurate number of voting nodes with voting values ​​higher than the judgment threshold is incremented by one.

[0012] The present invention has the following advantages: The present invention uses all the fire alarm sensors around the target fire alarm sensor to vote on whether a fire has occurred. It can integrate the information of all the fire alarm sensors around the target fire alarm sensor, thereby avoiding false fire alarms caused by sensor malfunction or sensor failure, and more accurately judge the fire situation. It also directly transforms the traditional fire alarm system, which is less costly than rebuilding a set of smart fire protection systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a schematic diagram of the structure of a fire monitoring and alarm system based on Internet of Things sensors adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0015] Embodiment 1, a fire monitoring and alarm method based on an Internet of Things sensor, comprising: Step S1: for any fire alarm sensor, the fire alarm sensor generally includes a temperature sensor and a smoke sensor, a sensor data value is obtained through the fire alarm sensor, the sensor data value here can be a temperature data value or a smoke concentration data value, and whether a fire situation occurs is determined based on the sensor data value. The specific judgment method is determined by the specific form of the sensor data value. For example, a temperature data value higher than a temperature threshold is considered to be a fire situation, a smoke concentration data value higher than a smoke concentration threshold is considered to be a fire situation, or a temperature data value and a smoke concentration data value simultaneously higher than the corresponding temperature threshold and smoke concentration threshold are considered to be a fire situation (for a composite fire alarm). If a fire situation occurs, the fire alarm sensor is recorded as a target fire alarm sensor, and step S2 is entered; if a fire situation does not occur, the sensor data value is continued to be obtained through the fire alarm sensor; Step S2: Obtain all fire alarm sensors within a preset distance from the target fire alarm sensor through the Internet of Things, and record them as voting nodes. It should be noted that the preset distance here is determined by the developer, and it is ensured that all fire alarm sensors within the preset distance from the target fire alarm sensor are located in the same space. The voting value corresponding to the voting node is weighted according to the sensor data change difference, consistency coefficient and change correlation coefficient corresponding to the voting node. The first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the weighted sensor data change difference, consistency coefficient and change correlation coefficient are optimized by the particle swarm optimization algorithm improved based on environmental perception. The voting operation corresponding to the voting node is completed, and all votes are The number of votes corresponding to the node is incremented by one, and the voting value represents the confidence of the voting node in the occurrence of a fire. The sensor data value corresponding to the voting node here can be used to determine whether the temperature data values ​​obtained by the surrounding fire alarm sensors and the temperature sensor data values ​​obtained by the target fire alarm sensor satisfy the heat conduction model, or to determine whether the smoke concentration data values ​​obtained by the surrounding fire alarm sensors and the smoke concentration sensor data values ​​obtained by the target fire alarm sensor satisfy the gas diffusion model. The consistency of the changes in all sensor data values ​​is used to determine whether the sensor data values ​​obtained by the surrounding fire alarm sensors show the same changes. For example, after a fire occurs, the surrounding temperature data values ​​will show a similar increase. Step S3: The voting values ​​corresponding to all voting nodes are weighted averaged to obtain the total voting value. The voting weights in the weighted average calculation process are determined by a weight adaptive adjustment model. The input data of the weight adaptive adjustment model are the historical accuracy rate corresponding to the voting node and the environmental data set in the current space. The historical accuracy rate is the ratio of the accurate number of times corresponding to the voting node to the total number of votes. The weight adaptive adjustment model is established based on the BP neural network. The improvement lies in setting a physical rule attention layer. The environmental data set generally includes ambient temperature, wind speed and wind direction, etc. It should be noted that the environmental data set is obtained through the environmental monitoring component closest to the target fire alarm sensor. The total voting value is compared with the judgment threshold. If the total voting value is higher than the judgment threshold, it means that the fire situation determined by the target fire alarm sensor is true, a fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting values ​​higher than the judgment threshold is increased by one; if the total voting value is lower than the judgment threshold, it means that the fire situation determined by the target fire alarm sensor is not true, and the target fire alarm sensor may have a fault, so no fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting values ​​higher than the judgment threshold is increased by one; It should be noted that in order to carry out intelligent transformation of the traditional fire alarm system, the information obtained by all fire alarm sensors can be uploaded to the Internet of Things through the protocol transfer station to realize information interaction, and corresponding environmental monitoring components can be set up in different areas to obtain environmental data sets. In order to vote on the fire situation, under the voting operation jointly participated by multiple voting nodes, the information fusion between multiple fire alarm sensors can be realized, thereby avoiding false fire alarms caused by sensor failure or sensor failure; This application uses all the fire alarm sensors around the target fire alarm sensor to vote on whether a fire has occurred. It can integrate the information of all the fire alarm sensors around the target fire alarm sensor, thereby avoiding false fire alarms caused by sensor failure or sensor failure, and more accurately judge the fire situation. It also directly transforms the traditional fire alarm system, which is less costly than rebuilding a smart fire protection system.

[0016] The voting value corresponding to the voting node is determined according to the consistency between the sensor data value corresponding to the voting node and the change of all sensor data values, which specifically includes the following steps: For any voting node, all sensor data values ​​corresponding to a preset time length are collected from the current time point forward, and the timestamp is used as the independent variable and the sensor data value is used as the dependent variable to fit the sensor data value change curve; cluster analysis is performed on the sensor data value change curves corresponding to all voting nodes, and the cluster center corresponding to the largest cluster cluster is selected as the target sensor data value change curve. The cluster analysis generally adopts k-means, and in the cluster analysis process, the distance between any two sensor data value change curves is the correlation coefficient between the two sensor data value change curves. The correlation coefficient between the two sensor data value change curves is generally calculated in the following way: determine the inner product of the two sensor data value change curves on the definition domain, calculate the norms corresponding to the sensor data value change curves respectively, and then divide the calculated inner product by the product of the norms corresponding to the two sensor data value change curves to obtain the correlation coefficient between the two sensor data value change curves; the correlation coefficient here can reflect the consistency of the change situation represented by the two sensor data value change curves; For any voting node, calculate the range of the corresponding sensor data value change curve and record it as the sensor data change difference F. The sensor data change difference F can reflect the changes of the voting node in the past. Calculate the correlation coefficient between the sensor data value change curve corresponding to the voting node and the target sensor data value change curve and record it as the consistency coefficient D. The consistency coefficient D can reflect whether the sensor data value change of the voting node is consistent with the sensor data value change of all voting nodes, and can determine whether there is a problem with the voting node itself; Calculate the sensor data value change curve corresponding to the voting node and the sensor data value change curve corresponding to the target fire alarm sensor The correlation coefficient between the data value change curves is recorded as the change correlation coefficient E. The change correlation coefficient E can reflect whether the change of the sensor data value of the voting node is consistent with the change of the sensor data value corresponding to the target fire alarm sensor. The higher the change correlation coefficient E, the higher the trust of the voting node in the target fire alarm sensor. The voting value Q=W1F+W2D+W3E corresponding to the voting node is calculated, where W1, W2 and W3 are the first weight coefficient, the second weight coefficient and the third weight coefficient respectively, and W1+W2+W3=1 is satisfied. It should be noted that the first weight coefficient, the second weight coefficient and the third weight coefficient here are determined by the developer; It should be noted that the voting value here can reflect the degree of trust of the voting node in the target fire alarm sensor, and it can be expected that both the target fire alarm sensor and the voting node may have problems, so it is necessary to determine whether the fire alarm of the target fire alarm sensor is credible through a voting operation involving multiple voting nodes; The voting values ​​corresponding to all voting nodes are weighted averaged to obtain the total voting value. The voting weights in the weighted average calculation process are adaptively adjusted according to the historical accuracy rates corresponding to all voting nodes and the environmental data set in the current space. The specific steps include the following: At the end of the sensor data value corresponding to the current time point of each voting node, the sensor data value corresponding to the target fire alarm sensor, the environmental data set in the current space, and the historical accuracy rate corresponding to the voting node are added to construct a voting weight analysis vector, and the voting weight analysis vectors corresponding to all voting nodes are combined into a voting weight analysis data set, and the voting weight analysis data set is processed by the weight adaptive adjustment model to output a voting weight set, which includes the voting weight corresponding to each voting node; the weight adaptive adjustment model can be a BP neural network, including an input layer, a physical rule attention layer, a hidden layer, and an output layer, wherein the input layer is used to receive the voting weight analysis data set, wherein the physical rule attention layer is used to establish a physical law association matrix based on the voting node, and the voting weight analysis data set is executed The self-attention mechanism is performed on the voting weight analysis data set through the physical law association matrix to construct a physical fusion voting weight analysis data set, the hidden layer is used to perform full connection processing on the physical fusion voting weight analysis data set, and the output layer is used to output the voting weight set; It should be noted that, in this embodiment, for the smoke alarm, the physical law association matrix is ​​constructed in the following way: the physical law association matrix is ​​a matrix of H×H, H is the total number of voting nodes in the voting weight analysis data set, and the smoke diffusion association value R(i, j)=exp(-δL stored in the i-th row and j-th column of the physical law association matrix 2 (i, j)), where δ is the smoke diffusion attenuation coefficient, which is related to the air diffusion characteristics, and L(i, j) is the distance between the i-th voting node and the j-th voting node; In the process of processing the voting weight analysis data set through the weight adaptive adjustment model, the weight adaptive adjustment model will analyze whether the sensor data values ​​of different voting nodes are reasonable according to the current environmental conditions and historical accuracy, and weight the voting values ​​output by the corresponding voting nodes according to the rationality of the sensor data values. If the sensor data value corresponding to the sensor corresponding to the voting node itself is unreasonable, its trust in the voting process will be reduced; The training of the weight adaptive adjustment model includes the following steps: Obtain weight adaptive adjustment training samples, which include voting weight analysis data sets. The voting weight analysis data sets here come from a simulated fire experiment actually performed, that is, in a space, set up multiple fire alarm sensors by yourself, and these fire alarm sensors are good and bad. Then randomly set open flames (under the condition of setting safety protection), test the alarm conditions of the fire alarm sensors, and take the fire alarm sensor closest to the open flame as the target fire alarm sensor to obtain the corresponding voting weight analysis data set; all weight adaptive adjustment training samples form a weight adaptive adjustment training set, and train the weight adaptive adjustment model through the weight adaptive adjustment training set. During the training, the training of the weight adaptive adjustment model is carried out simultaneously with the simulated fire experiment, and the alarm accuracy of the fire situation is used as the goal to determine whether the number of training times reaches the preset number of times. The preset number of times is determined by the developer. If the number of training times reaches the preset number of times, the trained weight adaptive adjustment model is output; otherwise, the weight adaptive adjustment model continues to be trained through the weight adaptive adjustment training set.

[0017] The first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the weighted sensor data change difference, the consistency coefficient and the change correlation coefficient are optimized by the particle swarm optimization algorithm improved based on environmental perception, specifically including the following steps: Construct a number of weight coefficient simulation individuals, and divide all weight coefficient simulation individuals into an environmentally sensitive population set and an environmentally stable population set; the weight coefficient simulation individuals are constructed as follows: three randomly initialized values ​​correspond to the first weight coefficient, the second weight coefficient and the third weight coefficient, respectively, and then the three randomly initialized values ​​are normalized to obtain the weight coefficient simulation individuals; Calculate the fitness corresponding to each weight coefficient simulation individual, specifically including the following steps: simulate individuals with weight coefficients in the environmentally sensitive population set, set the corresponding first weight coefficient, second weight coefficient and third weight coefficient through the weight coefficient simulation individual, and then traverse the environmentally sensitive fire analysis data set corresponding to the environmentally sensitive population set, the environmentally sensitive fire analysis data set includes a number of voting nodes under environmental sensitivity, and the fire conditions detected by the target fire alarm sensors corresponding to all voting nodes are true. For each fire analysis data set, calculate the voting value corresponding to each voting node. If the voting value is higher than the judgment threshold, it is regarded as a successful judgment. The ratio of the total number of successful judgments to the total number of voting nodes in the environmentally sensitive fire analysis data set is used as the number of voting nodes in the environmentally sensitive population set. The weight coefficient simulates the fitness of the individual; for the weight coefficient simulation individual in the environmentally stable population set, the first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the individual setting are simulated by the weight coefficient simulation individual, and then the environmentally stable fire analysis data set corresponding to the environmentally stable population set is traversed. The environmentally stable fire analysis data set includes several voting nodes under environmental stability. The fire conditions detected by the target fire alarm sensors corresponding to all voting nodes are true. For each fire analysis data set, the voting value corresponding to each voting node is calculated. If the voting value is higher than the judgment threshold, it is regarded as a successful judgment. The ratio of the total number of successful judgments to the total number of voting nodes in the environmentally stable fire analysis data set is used as the fitness of the individual simulated by the weight coefficient in the environmentally stable population set; It should be noted that both the environmentally sensitive fire analysis dataset and the environmentally stable fire analysis dataset are historical real data collected artificially, and environmental sensitivity refers to the situation where the change amplitude is large after analyzing the environment (temperature change, wind speed change, and wind direction change). Here, the judgment of the change amplitude is compared with the corresponding threshold, and the environmental stability means that the change amplitude is small; The environmentally sensitive population set and the environmentally stable population set are iteratively updated by the particle swarm optimization algorithm, and the number of iterations is recorded. After every N iterations, the weight coefficient simulation individual with the highest fitness in the environmentally sensitive population set and the environmentally stable population set is selected as the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual, and then the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual are cross-fused to construct the third optimal weight coefficient simulation individual, and the third optimal weight coefficient simulation individual is used to replace the weight coefficient simulation individual with the lowest fitness in the environmentally sensitive population set and the environmentally stable population set; the specific method of cross-fusion is: the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual are summed and averaged and then normalized; The particle swarm iterative optimization here adopts the existing technology. The difference is that in the process of calculating the voting value, environmental sensitivity and environmental stability will correspond to different situations. When the environment is sensitive, the difference in sensor data changes plays an important role. When the environment is stable, the importance of consistency coefficient and change correlation coefficient is highlighted. Therefore, when simulating the calculation of the first weight coefficient, the second weight coefficient and the third weight coefficient, two parallel environmentally sensitive population sets and environmentally stable population sets are set for particle swarm optimization, and in the iterative process, cross-level particle experience exchanges (cross fusion) are set up to make the obtained first weight coefficient, second weight coefficient and third weight coefficient more consistent with the actual fire scene; When the number of iterations reaches the maximum number of iterations, the weight coefficient simulation individual with the highest fitness in the environmentally sensitive population set and the weight coefficient simulation individual with the highest fitness in the environmentally stable population set are cross-fused to obtain the optimal weight coefficient simulation individual, and the first weight coefficient, the second weight coefficient and the third weight coefficient are determined through the optimal weight coefficient simulation individual.

[0018] Example 2, a fire monitoring and alarm system based on IoT sensors, see Figure 1 ,include: The fire preliminary judgment module is used for any fire alarm sensor, where the fire alarm sensor generally includes a temperature sensor and a smoke sensor, and obtains a sensor data value through the fire alarm sensor, where the sensor data value can be a temperature data value or a smoke concentration data value, and judges whether a fire situation occurs based on the sensor data value. The specific judgment method is determined by the specific form of the sensor data value. For example, a temperature data value higher than a temperature threshold is considered to be a fire situation, a smoke concentration data value higher than a smoke concentration threshold is considered to be a fire situation, or a temperature data value and a smoke concentration data value simultaneously higher than the corresponding temperature threshold and smoke concentration threshold are considered to be a fire situation (for a composite fire alarm). If a fire situation occurs, the fire alarm sensor is recorded as a target fire alarm sensor; if no fire situation occurs, continue to obtain sensor data values ​​through the fire alarm sensor; The voting calculation module is used to obtain all fire alarm sensors within a preset distance from the target fire alarm sensor through the Internet of Things, and record them as voting nodes. It should be noted that the preset distance here is determined by the developer, and it is ensured that all fire alarm sensors within the preset distance from the target fire alarm sensor are located in the same space. The voting value corresponding to the voting node is determined according to the consistency of the sensor data value corresponding to the voting node and the change of all sensor data values, and the voting operation corresponding to the voting node is completed. The number of votes corresponding to all voting nodes is increased by one. The voting value represents the confidence of the voting node in the occurrence of a fire. The sensor data value corresponding to the voting node here can be used to determine whether the temperature data value obtained by the surrounding fire alarm sensors and the temperature sensor data value obtained by the target fire alarm sensor meet the heat conduction model, or determine whether the smoke concentration data value obtained by the surrounding fire alarm sensors and the smoke concentration sensor data value obtained by the target fire alarm sensor meet the gas diffusion model. The consistency of the change of all sensor data values ​​is used to determine whether the sensor data values ​​obtained by the surrounding fire alarm sensors show the same change. For example, after a fire occurs, the surrounding temperature data values ​​will show a similar increase. The fire judgment module is used to perform weighted average calculation on the voting values ​​corresponding to all voting nodes to obtain the total voting value. The voting weights in the weighted average calculation process are adaptively adjusted through the historical accuracy rates corresponding to all voting nodes and the environmental data set in the current space. The historical accuracy rate is the ratio of the accurate number of times corresponding to the voting node to the total number of votes. The environmental data set generally includes ambient temperature, wind speed and wind direction, etc. It should be noted that the environmental data set is obtained through the environmental monitoring component closest to the target fire alarm sensor, and the total voting value is compared with the judgment threshold. If the total voting value is higher than the judgment threshold, it means that the fire situation determined by the target fire alarm sensor is true, a fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting values ​​higher than the judgment threshold is increased by one; if the total voting value is lower than the judgment threshold, it means that the fire situation determined by the target fire alarm sensor is not true, the target fire alarm sensor may have a fault, no fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting values ​​higher than the judgment threshold is increased by one.

[0019] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A fire monitoring and alarm method based on Internet of Things sensors, characterized in that: include: Step S1: Recording the fire alarm sensor that detects the fire situation as the target fire alarm sensor; Step S2: all fire alarm sensors within a preset distance from the target fire alarm sensor are obtained through the Internet of Things and recorded as voting nodes. The voting value corresponding to the voting node is weighted and calculated according to the sensor data change difference, consistency coefficient and change correlation coefficient corresponding to the voting node. The first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the weighted sensor data change difference, consistency coefficient and change correlation coefficient are optimized by the particle swarm optimization algorithm improved based on environmental perception. Step S3: Perform weighted average calculation on the voting values ​​corresponding to all voting nodes to obtain the total voting value. The voting weight in the weighted average calculation process is determined by a weight adaptive adjustment model. The input data of the weight adaptive adjustment model are the historical accuracy rate corresponding to the voting node and the environmental data set in the current space. The historical accuracy rate is the ratio of the accurate number of times corresponding to the voting node to the total number of votes. The weight adaptive adjustment model is established based on the BP neural network. The improvement lies in setting up a physical rule attention layer to judge the fire situation based on the total voting value.

2. A fire monitoring and alarm method based on Internet of Things sensors according to claim 1, characterized in that: The voting value corresponding to the voting node is calculated by weighting according to the sensor data change difference, consistency coefficient and change correlation coefficient corresponding to the voting node, specifically including the following steps: For any voting node, all sensor data values ​​corresponding to a preset time length are collected from the current time point forward, and the timestamp is used as the independent variable and the sensor data value is used as the dependent variable to fit the sensor data value change curve; cluster analysis is performed on the sensor data value change curves corresponding to all voting nodes, and the cluster center corresponding to the largest cluster is selected as the target sensor data value change curve, and in the cluster analysis process, the distance between any two sensor data value change curves is the correlation coefficient between the two sensor data value change curves; For any voting node, calculate the range of the corresponding sensor data value change curve and record it as the sensor data change difference F; calculate the correlation coefficient between the sensor data value change curve corresponding to the voting node and the target sensor data value change curve and record it as the consistency coefficient D; Calculate the correlation coefficient between the sensor data value change curve corresponding to the voting node and the sensor data value change curve corresponding to the target fire alarm sensor, and record it as the change correlation coefficient E; calculate the voting value Q=W1F+W2D+W3E corresponding to the voting node, where W1, W2 and W3 are the first weight coefficient, the second weight coefficient and the third weight coefficient respectively, and satisfy W1+W2+W3=1.

3. A fire monitoring and alarm method based on Internet of Things sensors according to claim 2, characterized in that: The voting values ​​corresponding to all voting nodes are weighted averaged to obtain the total voting value. The voting weights in the weighted average calculation process are adaptively adjusted according to the historical accuracy rates corresponding to all voting nodes and the environmental data set in the current space. The specific steps include the following: The sensor data value corresponding to the target fire alarm sensor, the environmental data set in the current space and the historical accuracy rate corresponding to the voting node are added to the end of the sensor data value corresponding to each voting node at the current time point, and a voting weight analysis vector is constructed. The voting weight analysis vectors corresponding to all voting nodes are combined into a voting weight analysis data set. The voting weight analysis data set is processed through a weight adaptive adjustment model to output a voting weight set, which includes the voting weight corresponding to each voting node.

4. A fire monitoring and alarm method based on Internet of Things sensors according to claim 3, characterized in that: The weight adaptive adjustment model is a BP neural network, including an input layer, a physical rule attention layer, a hidden layer and an output layer. The input layer is used to receive the voting weight analysis data set. The physical rule attention layer is used to establish a physical law association matrix based on the voting nodes, and perform a self-attention mechanism on the voting weight analysis data set through the physical law association matrix to construct a physical fusion voting weight analysis data set. The hidden layer is used to perform full connection processing on the physical fusion voting weight analysis data set. The output layer is used to output the voting weight set.

5. A fire monitoring and alarm method based on Internet of Things sensors according to claim 4, characterized in that: The training of the weight adaptive adjustment model includes the following steps: Obtain weight adaptive adjustment training samples, which include voting weight analysis data sets; combine all weight adaptive adjustment training samples into a weight adaptive adjustment training set, train the weight adaptive adjustment model through the weight adaptive adjustment training set, and determine whether the number of training times reaches a preset number of times. If the number of training times reaches the preset number of times, output the trained weight adaptive adjustment model; otherwise, continue to train the weight adaptive adjustment model through the weight adaptive adjustment training set.

6. A fire monitoring and alarm method based on Internet of Things sensors according to claim 5, characterized in that: The fire situation is judged based on the total vote value, specifically including the following steps: if the total vote value is higher than the judgment threshold, a fire alarm signal is issued, which is regarded as a successful fire alarm, and the exact number of voting nodes whose voting values ​​are higher than the judgment threshold is increased by one; if the total vote value is lower than the judgment threshold, no fire alarm signal is issued, which is regarded as a failed fire alarm, and the exact number of voting nodes whose voting values ​​are higher than the judgment threshold is increased by one, and each time the voting node performs a voting value calculation, the total number of votes corresponding to the voting node is increased by one.

7. A fire monitoring and alarm method based on Internet of Things sensors according to claim 6, characterized in that: The first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the weighted sensor data change difference, the consistency coefficient and the change correlation coefficient are optimized by the particle swarm optimization algorithm improved based on environmental perception, which specifically includes the following steps: Construct several weight coefficient simulation individuals, and divide all weight coefficient simulation individuals into environmentally sensitive population sets and environmentally stable population sets; Calculate the fitness corresponding to each weight coefficient simulation individual, specifically including the following steps: simulate individuals with weight coefficients in the environmentally sensitive population set, set the corresponding first weight coefficient, second weight coefficient and third weight coefficient through the weight coefficient simulation individual, and then traverse the environmentally sensitive fire analysis data set corresponding to the environmentally sensitive population set, the environmentally sensitive fire analysis data set includes a number of voting nodes under environmental sensitivity, and the fire conditions detected by the target fire alarm sensors corresponding to all voting nodes are true. For each fire analysis data set, calculate the voting value corresponding to each voting node. If the voting value is higher than the judgment threshold, it is regarded as a successful judgment. The ratio of the total number of successful judgments to the total number of voting nodes in the environmentally sensitive fire analysis data set is used as the number of voting nodes in the environmentally sensitive population set. The weight coefficient simulates the fitness of the individual; for the weight coefficient simulation individual in the environmentally stable population set, the first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the individual setting are simulated by the weight coefficient simulation individual, and then the environmentally stable fire analysis data set corresponding to the environmentally stable population set is traversed. The environmentally stable fire analysis data set includes several voting nodes under environmental stability. The fire conditions detected by the target fire alarm sensors corresponding to all voting nodes are true. For each fire analysis data set, the voting value corresponding to each voting node is calculated. If the voting value is higher than the judgment threshold, it is regarded as a successful judgment. The ratio of the total number of successful judgments to the total number of voting nodes in the environmentally stable fire analysis data set is used as the fitness of the individual simulated by the weight coefficient in the environmentally stable population set; The environmentally sensitive population set and the environmentally stable population set are iteratively updated by the particle swarm optimization algorithm, and the number of iterations is recorded. After every N iterations, the weight coefficient simulation individual with the highest fitness in the environmentally sensitive population set and the environmentally stable population set is selected as the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual, and then the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual are cross-fused to construct the third optimal weight coefficient simulation individual, and the third optimal weight coefficient simulation individual is used to replace the weight coefficient simulation individual with the lowest fitness in the environmentally sensitive population set and the environmentally stable population set; the specific method of cross-fusion is: the first optimal weight coefficient simulation individual and the second local optimal weight coefficient simulation individual are summed and averaged and then normalized; When the number of iterations reaches the maximum number of iterations, the weight coefficient simulation individual with the highest fitness in the environmentally sensitive population set and the weight coefficient simulation individual with the highest fitness in the environmentally stable population set are cross-fused to obtain the optimal weight coefficient simulation individual, and the first weight coefficient, the second weight coefficient and the third weight coefficient are determined through the optimal weight coefficient simulation individual.

8. A fire monitoring and alarm method based on Internet of Things sensors according to claim 7, characterized in that: Cluster analysis was performed using k-means.

9. A fire monitoring and alarm system based on Internet of Things sensors, characterized in that: The system applies a fire monitoring and alarm method based on an Internet of Things sensor as described in any one of claims 1 to 8, including: A fire preliminary judgment module is used for obtaining sensor data values ​​through the fire alarm sensor for any fire alarm sensor, and judging whether a fire situation has occurred based on the sensor data values. If a fire situation has occurred, the fire alarm sensor is recorded as a target fire alarm sensor; if no fire situation has occurred, the sensor data values ​​continue to be obtained through the fire alarm sensor; A voting calculation module is used to obtain all fire alarm sensors within a preset distance from a target fire alarm sensor through the Internet of Things, and record them as voting nodes. The voting value corresponding to the voting node is weightedly calculated according to the sensor data change difference, consistency coefficient and change correlation coefficient corresponding to the voting node. The first weight coefficient, the second weight coefficient and the third weight coefficient corresponding to the weighted sensor data change difference, consistency coefficient and change correlation coefficient are optimized by the particle swarm optimization algorithm improved based on environmental perception to complete the voting operation corresponding to the voting node. Each time the voting node performs a voting value calculation, the total number of votes corresponding to the voting node is increased by one. The voting value represents the confidence of the voting node in the occurrence of a fire. The fire judgment module is used to perform weighted average calculation on the voting values ​​corresponding to all voting nodes to obtain the total voting value. The voting weight in the weighted average calculation process is determined by a weight adaptive adjustment model. The input data of the weight adaptive adjustment model are the historical accuracy rate corresponding to the voting node and the environmental data set in the current space. The historical accuracy rate is the ratio of the accurate number of times corresponding to the voting node to the total number of votes. The weight adaptive adjustment model is established based on the BP neural network. The improvement lies in setting up a physical rule attention layer to compare the total voting value with the judgment threshold. If the total voting value is higher than the judgment threshold, a fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting value higher than the judgment threshold is added by one; if the total voting value is lower than the judgment threshold, no fire alarm signal is issued, and the accurate number of voting nodes corresponding to the voting value higher than the judgment threshold is added by one.

Images