Laboratory fire detection method based on multi-node sensor data fusion

Abstract

The invention discloses a laboratory fire detection method based on multi-node sensor data fusion, and the method comprises the following steps: building a laboratory multi-node fire monitoring network through employing a LoRa communication technology, and increasing the time slot interval between nodes, so as to improve a time division multiplexing and incremental transmission communication mode; introducing an extreme learning machine to perform feature layer data fusion on data including temperature, smoke concentration and CO concentration acquired by each node in the laboratory, thereby outputting the probability of occurrence of each fire; optimizing the extreme learning machine by using a sparrow search algorithm; and by means of a fuzzy reasoning method, in combination with the occurrence probability of each fire and the duration of the fire, performing feature fusion in a decision-making level, and determining the level of a fire alarm. By fusing multi-sensor data, the perceptual ability of the system to fire events is improved, and by combining a fuzzy reasoning technology in a decision layer, the flexibility and practicability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of laboratory safety monitoring and early warning, and in particular to a laboratory fire detection method based on multi-node sensor data fusion. Background Art

[0002] Fire is a common laboratory accident that has a serious impact on public safety, with a large scope of harm and economic losses. Therefore, fire detection and real-time warning have attracted much attention. The intelligent fire monitoring system is currently a widely used method. However, in actual applications, the number of consumables required for the monitoring system is large, maintenance is difficult, and the monitoring lines are prone to aging and damage, which affects the real-time and reliability of fire monitoring. The key to fire detection and early warning is to perceive the occurrence of fire, issue early warnings in time, and take the best treatment plan to minimize the losses caused by fire. Therefore, improving fire detection and early warning capabilities is crucial to ensuring public safety. Innovating technical means and management methods to improve the efficiency of fire detection and early warning systems are of great significance to reducing fire risks and protecting the safety of people and property.

[0003] Extreme Learning Machine (ELM) is an emerging machine learning technology with fast learning speed and good generalization ability. Unlike traditional neural network methods, ELM only needs to randomly initialize the connection weights from the input layer to the hidden layer during the network training phase, and then directly determine the connection weights from the output layer to the hidden layer through analytical solutions, which greatly reduces the training time and complexity. ELM has efficient computing speed and less parameter adjustment, and is suitable for large-scale data processing and real-time application scenarios. However, since the output layer of the extreme learning machine is obtained through random initialization, the model may converge slowly during the training process, requiring more training time and data to achieve ideal performance. Summary of the invention

[0004] The purpose of the present invention is to provide a laboratory fire detection method of multi-node sensor data fusion.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions: A laboratory fire detection method based on multi-node sensor data fusion includes the following steps: A multi-node fire monitoring network in the laboratory was built by using LoRa communication technology, and the time slot interval between nodes was increased to improve the time division multiplexing and incremental transmission communication mode; The extreme learning machine is introduced to perform feature layer data fusion on the data collected by each node in the laboratory, including temperature, smoke concentration and CO concentration, so as to output the probability of each fire occurrence. Use the sparrow search algorithm to optimize the extreme learning machine; With the help of fuzzy reasoning method, combined with the probability of each fire occurrence and the fire duration, feature fusion is performed at the decision-making level to determine the level of fire alarm.

[0006] As a specific implementation method, the process of building a laboratory multi-node fire monitoring network using LoRa communication technology is as follows: the STM32 series microprocessor is used as the node control core, the LoRa-SX1278 chip is used as the communication module, and the LoRa-SX1278 chip is responsible for building each monitoring node of the overall laboratory environment and realizing the aggregation of data from each monitoring node.

[0007] As a specific implementation, the networking mode of the laboratory multi-node fire monitoring network is to form a star network with the aggregation node and the monitoring node. The improved time division multiplexing (TDM) and incremental transmission communication mode is different from the traditional TDM. In each cycle of the signal, the time slot interval between nodes is increased to protect the data security between each information source and solve the data collision risk caused by communication delay.

[0008] As a specific implementation method, the data collected by each node in the laboratory needs to be preprocessed before feature fusion. The preprocessing includes normalizing the sensor data. The normalization formula is as follows: , in c Indicates the number of parameters collected by the sensor, including CO concentration, smoke concentration, and temperature. A c Represents input c The normalized value of the parameter, X c Determined by experimental data c The input value of the parameter, X min and X max are the minimum and maximum input values ​​of the selected parameters respectively.

[0009] As a specific implementation method, the network output model of the extreme learning machine is: T=[ t 1 t 2 … t m ], , Where: T is the output data set, t m represents the network output value of the mth training sample, l jrepresents the output vector of the 1st to mth output neurons corresponding to the hidden layer neuron j, t mj Indicates the output value of the mth output neuron corresponding to the hidden layer neuron j, b im represents the connection weight of hidden layer neuron j to output neuron m, g(•) is the activation function of hidden layer neuron, x j is the input sample of hidden layer neuron j, w i is the input sample x corresponding to the hidden layer neuron j j The weight of i is the threshold of hidden layer neuron j.

[0010] As a specific implementation method, the sparrow search algorithm first needs to initialize the sparrow population and fitness. The population uses a random position generation method to initialize the individual positions of the sparrow population. The mathematical expression is as follows: , In the formula, s is the number of sparrows, d is the variable dimension to be optimized, x sd Indicates s A sparrow for the d The value of the parameter, Among them, in the sparrow search algorithm, the finder will give priority to obtaining food during the foraging process, and send out a signal after obtaining the food to provide food directions for other individuals. The position update formula of the finder is as follows: , Where: t is the number of current iterations; X t,qr and X (t+1),qr Respectively represent t In the t and t+1 iterations, the first q The position of the sparrow in dimension r; t max is the maximum number of iterations in this calculation; R2 is the warning value, R2∈[0,1]; F is the safety value, F∈[0.5,1]; Q is a random number, which obeys the normal distribution; a Represents a random number, a (0,1); L is a 1× d A matrix where all elements have the value 1. when R 2 <F This indicates that there are no predators at this time, and the discoverer further expands the search; R 2 ≥F, indicating that when a predator is found, all sparrows need to fly to other safe areas, where the position of the follower changes with the fitness, the formula is as follows: , Where: X (t+1),p is the optimal position of the finder after the t+1th iteration; X t,w is the global worst position of the t-th iteration; X t,qr and X (t+1),qr Respectively represent t In the t and t+1 iterations, the first q The position of the sparrow in dimension r; A is a 1×d matrix, each element is randomly selected as 1 or -1, A + = A T ( AA T ) -1 ; T represents the transpose of the matrix; L represents a 1× d Matrix of s 1 is the number of followers, not the population size; when q>s 1 / 2, indicating that the fitness is low q A follower did not get food and had to fly to other places to find food; The alerter is the sparrow that is partially responsible for alerting during the foraging process. That is, when danger approaches, no matter the sparrow is a discoverer or a tracker, it will abandon the current food and move to a new location. The formula is as follows: , Where: q Only the number of sparrows aware of the danger is traversed, i.e. the number of alerters, not the entire population; X t,b After the tth iteration, the sparrow with the best fitness value is in the global optimal position; B is the step size control parameter, which is a random number that follows a normal distribution with a mean of 0 and a variance of 1; K is a random number, K∈[-1,1]; e is a constant; f q is the fitness value of the sparrow individual; f g and f w are the global best and worst fitness values, respectively. The above updates the positions of the discoverer, follower and sentinel in sequence. If there is no optimal sparrow position, continue updating until the stopping condition is met.

[0011] As a specific implementation method, the specific steps of optimizing the extreme learning machine using the sparrow search algorithm are as follows: S1, normalize the data and divide it into training set and test set; S2, initializes the number of iterations, population size, and the ratio of discoverers to followers of the sparrow optimization algorithm; S3, establish the extreme learning machine network structure and set the network parameters; S4, calculate the fitness value of the sparrow group and find the best fitness value f g and the worst fitness value f w , update the positions of the discoverer, follower and sentinel in order. If the updated position is better, replace it, otherwise return and continue to update the positions of the discoverer, follower and sentinel in order; S5, continuously iteratively updates the position of the sparrows until the set maximum number of iterations is reached, and the sparrow with the lowest fitness is regarded as the optimal solution; S6, after completing the iteration, the obtained optimal weights and thresholds are applied to the extreme learning machine to form the SSA-ELM model.

[0012] As a specific implementation method, the decision layer includes four decision factors, namely, the probability of open fire, the probability of smoldering fire, the duration of fire and the danger level of equipment. By combining the decision factors and fuzzy processing, the final result is determined. The duration of fire T is determined by the following formula: T ( h )=[ T ( h +1)+1]× V [ Y i ( x )- T d ], In the formula, h Indicates the current time period. h +1 indicates the next time period, v represents the probability of open fire and smoldering fire output from the feature layer, V (•) is a step function, T d The threshold for the alarm.

[0013] As a specific implementation method, in the decision-making layer feature fusion process, the input quantity is first fuzzified, and the input quantity includes the probability of open fire, the probability of smoldering fire, the duration of fire and the equipment hazard level. The fuzzy set is divided into positive small, positive medium and positive large, and the membership function adopts trigonometric function. The output fuzzy set is divided into none, small, medium and large, and the membership function adopts trigonometric function.

[0014] As a specific implementation method, the decision layer feature fusion needs to be defuzzified after fuzzification, and the corresponding formula is as follows: , In the formula, U Represents the output value obtained after defuzzification, which is used for further analysis. u ( z ) is the membership function of the output, z is the output variable of the fuzzy rule, and finally the defuzzified U Make multi-level judgments. U When <0.25, it is judged as no fire; When 0.25≤ U When <0.5, it is judged as warning; When 0.5≤ U When <0.75, it is judged as an alarm; when U When ≥0.75, it is judged as a serious alarm.

[0015] Compared with the prior art, the present invention is beneficial in that: 1) The present invention improves the system's ability to perceive fire events and enhances the comprehensiveness and accuracy of fire detection by fusing multi-sensor data; 2) The SSA algorithm is used in the feature layer to optimize the ELM weights and thresholds, which effectively improves the generalization ability of the model and makes the system perform better when dealing with complex data and nonlinear problems. At the same time, the fuzzy reasoning technology is combined in the decision-making layer to flexibly give alarm information based on the laboratory environment feature information and the degree of equipment danger, which improves the flexibility and practicality of the system, allowing the fire detection system to more accurately identify fires and take corresponding measures in a timely manner, thereby effectively ensuring the safety of equipment and personnel in the computer room; 3) The networking method of the laboratory multi-node fire monitoring network is to form a star network with the aggregation node and the monitoring node. It not only has the advantages of large communication radius and wide coverage, but also has a simple networking method. Compared with other networking methods, it has more prominent advantages in power consumption and cost. Based on the improved TDM-LoRa self-organizing network low-power fire monitoring and early warning system, it not only has the advantage of a wide monitoring range, but also solves the problems of channel resource waste and high energy consumption in star networking. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required to be used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. Figure 1 It is a flow chart of a laboratory fire detection method based on multi-node sensor data fusion; Figure 2 It is a diagram of the improved TDM communication data transmission mode of a laboratory fire detection method with multi-node sensor data fusion; Figure 3 A schematic diagram of the ELM network structure of a laboratory fire detection method with multi-node sensor data fusion; Figure 4 Flowchart of an extreme learning machine optimized for a sparrow search algorithm for a laboratory fire detection method with multi-node sensor data fusion.

[0017] Figure 5 This is a flow chart of the decision-layer fusion model of a laboratory fire detection method based on multi-node sensor data fusion. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention: Figure 1 Flowchart of a laboratory fire detection method based on multi-node sensor data fusion.

[0019] like Figure 1 As shown, a laboratory fire detection method based on multi-node sensor data fusion includes the following steps: Firstly, a laboratory multi-node fire monitoring network was built by using LoRa communication technology; Increase the time slot interval between nodes to improve time division multiplexing (TDM) and incremental transmission communication modes; The extreme learning machine (ELM) is introduced to perform feature layer data fusion on the data collected by each node in the laboratory, including temperature, smoke concentration and CO concentration, so as to output the probability of each fire occurrence; Use the Sparrow Search Algorithm (SSA) to optimize the prediction precision and accuracy of the Extreme Learning Machine (ELM); Finally, with the help of fuzzy reasoning method, combined with the probability of each fire occurrence and the fire duration, feature fusion is performed at the decision-making level to determine the level of fire alarm.

[0020] LoRa communication technology is used to build a laboratory multi-node fire monitoring network system: the STM32F103C8T6 microprocessor is used as the node control core, and the LoRa-SX1278 chip is used as the communication module, which is responsible for building each monitoring node of the overall laboratory environment and realizing the aggregation of data from each monitoring node.

[0021] The networking method of the laboratory fire monitoring network is to form a star network with aggregation nodes and monitoring nodes. It not only has the advantages of large communication radius and wide coverage, but also has a simple networking method. Compared with other networking methods, it has more prominent advantages in power consumption and cost.

[0022] Figure 2 This is a diagram of an improved TDM communication data transmission mode. Figure 2 As shown: The improved time division multiplexing (TDM) and incremental transmission communication mode is different from the traditional TDM. In each signal cycle, the time slot interval between nodes is increased to protect the data security between each information source and solve the risk of data collision caused by communication delays.

[0023] Specifically, the data collected by each node in the laboratory needs to be preprocessed before feature fusion. The preprocessing mainly involves normalizing the sensor data. The normalization formula is as follows: , in c Indicates the number of parameters collected by the sensor, including CO concentration, smoke concentration, and temperature. A c Represents input c The normalized value of the parameter, X c Determined by experimental data c The input value of the parameter, X min and X max are the minimum and maximum input values ​​of the selected parameters respectively.

[0024] Figure 3 This is a schematic diagram of the extreme learning machine (ELM) network structure; The extreme learning machine (ELM) is an efficient single hidden layer feedforward neural network model. Its core feature is to randomly generate the weights and biases of the hidden layer nodes and directly calculate the output weights through analytical solutions. This method significantly improves the training speed, reduces the computational complexity, and maintains good generalization ability.

[0025] The network output model of the extreme learning machine is: T=[ t 1 t 2 … t m ], , Where: T is the output data set, t m represents the network output value of the mth training sample, t j is the column vector set of matrix T, t mj Indicates the output value of the mth output neuron corresponding to the hidden layer neuron j, b im represents the connection weight of hidden layer neuron j to output neuron m, g(•) is the hidden layer neuron activation function, x j is the input sample of hidden layer neuron j, w i is the input sample x corresponding to the hidden layer neuron j j The weight of i is the threshold of hidden layer neuron j.

[0026] The sparrow search algorithm is a new type of swarm intelligence optimization algorithm, which is mainly inspired by the foraging and regurgitation behaviors of sparrows. It has the characteristics of strong optimization ability and fast convergence speed. In the SSA algorithm, the sparrow population is divided into discoverers and followers. The sparrow search algorithm first needs to initialize the sparrow population and fitness. The population uses the method of randomly generating positions to initialize the individual positions of the sparrow population.

[0027] The sparrow search algorithm first needs to initialize the sparrow population and fitness. The population uses the method of randomly generating positions to initialize the individual positions of the sparrow population. The mathematical expression is as follows: , In the formula, s is the number of sparrows, d is the variable dimension to be optimized, x sd Indicates s A sparrow for the d The value of a parameter.

[0028] Among them, in the sparrow search algorithm, the finder will give priority to obtaining food during the foraging process, and send out a signal after obtaining the food to provide food directions for other individuals. The position update formula of the finder is as follows: , Where: t is the number of current iterations; X t,qr and X(t+1),qr Respectively represent t In the t and t+1 iterations, the first q The position of the sparrow in dimension r (in this formula q Only traverse the discoverers, not the entire population. For example, if 30% of the population are discoverers, then after sorting according to the fitness value, the first 30% of the individuals are discoverers); t max is the maximum number of iterations in this calculation; R2 is the warning value, R2∈[0,1]; F is the safety value, F∈[0.5,1]; Q is a random number, which obeys the normal distribution; a Represents a random number, a (0,1); L is a 1× d A matrix where all elements have the value 1. when R 2 <F This indicates that there are no predators at this time, and the discoverer further expands the search; R 2 ≥F , indicating that when a predator is found, all sparrows need to fly to other safe areas, where the position of the follower changes with the fitness, the formula is as follows: , Where: X (t+1),p is the optimal position of the finder after the t+1th iteration; X t,w is the global worst position of the tth iteration; X t,qr and X (t+1),qr Respectively represent t In the t and t+1 iterations, the first q The position of the sparrow in dimension r (in this formula q Only traverse the followers, not the entire population. For example, if 70% of the population are followers, then after sorting according to the fitness value, the last 70% of the individuals are followers); A is a 1×d matrix, and each element is randomly selected as 1 or -1. A + = A T ( A A T ) -1 ; T represents the transpose of the matrix; L represents a 1× d Matrix of s 1 is the number of followers, not the population size; when q> s 1 / 2, indicating that the fitness is low q If a follower does not get food, it needs to fly to other places to find food.

[0029] When the sparrows are foraging, some of them will be responsible for vigilance. When danger approaches, they will give up the current food. That is, no matter whether the sparrow is a finder or a follower, it will give up the current food and move to a new position. The formula is as follows: , Where: q Only the number of sparrows that are aware of the danger (i.e., the alerters) is traversed, not the entire population; X t,b The global optimal position of the sparrow with the best fitness value after the tth iteration; B is the step size control parameter, which is a random number that follows a normal distribution with a mean of 0 and a variance of 1; K is a random number, K∈[-1,1]; e is a constant; f q is the fitness value of the sparrow individual; f g and f w are the global best and worst fitness values ​​respectively.

[0030] The above updates the positions of the discoverer, follower and sentinel in sequence. If there is no optimal sparrow position, continue updating until the stopping condition is met.

[0031] Figure 4 Flowchart for optimizing an extreme learning machine for the sparrow search algorithm.

[0032] The steps of optimizing the extreme learning machine using the sparrow search algorithm are: S1, normalize the data and divide it into training set and test set; S2, initialize the number of iterations, population size, and the ratio of discoverers to followers of the sparrow optimization algorithm; S3, establish the extreme learning machine network structure and set the network parameters; S4, calculate the fitness value of the sparrow group and find the best fitness value f g and the worst fitness value f w , according to the step of claim 6, the positions of the discoverer, the follower and the sparrow that is aware of the danger, i.e., the alerter, are updated. If the updated position is better, then the position is replaced. Otherwise, the position of the discoverer, the follower and the alerter is continued to be updated in sequence. S5, continuously iteratively updates the position of the sparrows until the set maximum number of iterations is reached, and the sparrow with the lowest fitness is regarded as the optimal solution; S6, after completing the iteration, the obtained optimal weights and thresholds are applied to the extreme learning machine to form the SSA-ELM model.

[0033] Specifically, the decision layer includes four decision factors, namely, the probability of open fire, the probability of smoldering fire, the duration of fire and the danger level of equipment. By combining the decision factors and fuzzy processing, the final result is determined. The duration of fire T is determined by the following formula: T ( h )=[ T ( h +1)+1]× V [ Y i ( x )- T d ], In the formula, h Indicates the current time period. h +1 indicates the next time period, Y i ( x ) represents the probability of open fire and smoldering fire output from the feature layer, V (•) is a step function, T d The threshold for the alarm.

[0034] The feature fusion of the decision layer first fuzzifies the input quantity, which includes the probability of open fire, the probability of smoldering fire, the duration of fire and the hazard level of equipment. The fuzzy set is divided into positive small, positive medium and positive large, and the membership function adopts trigonometric function. The output fuzzy set is divided into none, small, medium and large, and the membership function adopts trigonometric function.

[0035] The decision layer feature fusion needs to be defuzzified after fuzzification. The corresponding formula is as follows: , In the formula, U Represents the output value obtained after defuzzification, which is used for further analysis. u ( z ) is the membership function of the output, z is the output variable of the fuzzy rule, and finally the defuzzified U Make multi-level judgments. U When <0.25, it is judged as no fire; When 0.25≤ U When <0.5, it is judged as warning; When 0.5≤ U <0.75, it is judged as an alarm; U When ≥0.75, it is judged as a serious alarm.

[0036] The present invention improves the system's perception of fire events by fusing multi-sensor data, and enhances the comprehensiveness and accuracy of fire detection; the SSA algorithm is used in the feature layer to optimize the ELM weights and thresholds, effectively improving the generalization ability of the model, so that the system performs better when processing complex data and nonlinear problems. At the same time, the fuzzy reasoning technology is combined in the decision-making layer, and alarm information is flexibly given according to the laboratory environment feature information and the degree of equipment danger, which improves the flexibility and practicality of the system, so that the fire detection system can more accurately identify fires and take corresponding measures in time, thereby effectively ensuring the safety of equipment and personnel in the computer room; the laboratory multi-node fire monitoring network networking method is to form a star network with the aggregation node and the monitoring node, which not only has the advantages of large communication radius and wide coverage, but also has a simple networking method. Compared with other networking methods, power consumption and cost have more prominent advantages. Based on the improved TDM-LoRa self-organizing network low-power fire monitoring and early warning system, it not only has the advantages of wide monitoring range, but also solves the problems of channel resource waste and high energy consumption in star networking.

[0037] It should be emphasized that the above are only preferred embodiments of the present invention and do not limit the present invention in any form. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the case.

Claims

1. A laboratory fire detection method based on multi-node sensor data fusion, characterized in that: The following steps are involved: A multi-node fire monitoring network in the laboratory was built by using LoRa communication technology, and the time slot interval between nodes was increased to improve the time division multiplexing and incremental transmission communication mode; The extreme learning machine is introduced to perform feature layer data fusion on the data collected by each node in the laboratory, including temperature, smoke concentration and CO concentration, so as to output the probability of each fire occurrence. Use the sparrow search algorithm to optimize the extreme learning machine; With the help of fuzzy reasoning method, combined with the probability of each fire occurrence and the fire duration, feature fusion is performed at the decision-making level to determine the level of fire alarm.

2. The laboratory fire detection method of multi-node sensor data fusion according to claim 1 is characterized in that: The process of building a laboratory multi-node fire monitoring network using LoRa communication technology is as follows: the STM32 series microprocessor is used as the node control core, the LoRa-SX1278 chip is used as the communication module, and the LoRa-SX1278 chip is responsible for building each monitoring node of the overall laboratory environment and realizing the aggregation of data from each monitoring node.

3. The laboratory fire detection method of multi-node sensor data fusion according to claim 1 or 2, characterized in that: The networking method of the laboratory multi-node fire monitoring network is to form a star network with the aggregation nodes and the monitoring nodes.

4. The laboratory fire detection method of multi-node sensor data fusion according to claim 1 is characterized in that: The data collected by each node in the laboratory needs to be preprocessed before feature fusion. The preprocessing includes normalizing the sensor data. The normalization formula is as follows: ， in c Indicates the number of parameters collected by the sensor, including CO concentration, smoke concentration, and temperature. A c Represents input c The normalized value of the parameter, X c Determined by experimental data c The input value of the parameter, X min and X max are the minimum and maximum input values ​​of the selected parameters respectively.

5. The laboratory fire detection method of multi-node sensor data fusion according to claim 1 is characterized in that: The network output model of the extreme learning machine is: T＝[ t 1 t 2 … t m ]， ， Where: T is the output data set, t m represents the network output value of the mth training sample, l j represents the output vector of the 1st to mth output neurons corresponding to the hidden layer neuron j, t mj Indicates the output value of the mth output neuron corresponding to the hidden layer neuron j, b im represents the connection weight of hidden layer neuron j to output neuron m, g(•) is the hidden layer neuron activation function, x j is the input sample of hidden layer neuron j, w i is the input sample x corresponding to the hidden layer neuron j j The weight of i is the threshold of hidden layer neuron j.

6. The laboratory fire detection method of multi-node sensor data fusion according to claim 1 is characterized in that: The sparrow search algorithm first needs to initialize the sparrow population and fitness. The population uses the method of randomly generating positions to initialize the individual positions of the sparrow population. The mathematical expression is as follows: , In the formula, s is the number of sparrows, d is the variable dimension to be optimized, x sd Indicates s A sparrow for the d The value of the parameter, Among them, in the sparrow search algorithm, the finder will give priority to obtaining food during the foraging process, and send out a signal after obtaining the food to provide food directions for other individuals. The position update formula of the finder is as follows: ， Where: t is the number of current iterations; X t,qr and X (t+1),qr Respectively represent t In the t and t+1 iterations, the first q The position of the sparrow in dimension r; t max is the maximum number of iterations in this calculation; R2 is the warning value, R2∈[0,1]; F is the safety value, F∈[0.5,1]; Q is a random number, which obeys the normal distribution; a Represents a random number, a (0,1); L is a 1× d A matrix where all elements have the value 1. when R 2 <F This indicates that there are no predators at this time, and the discoverer further expands the search; R 2 ≥F , indicating that when a predator is found, all sparrows need to fly to other safe areas, where the position of the follower changes with the fitness, the formula is as follows: , Where: X (t+1),p is the optimal position of the finder after the t+1th iteration; X t,w is the global worst position of the tth iteration; X t,qr and X (t+1),qr Respectively represent t In the t and t+1 iterations, the first q The position of the sparrow in dimension r; A is a 1×d matrix, each element is randomly selected as 1 or -1, A + = A T ( AA T ) -1 ; T represents the transpose of the matrix; L represents a 1× d Matrix of s 1 is the number of followers, not the population size; when q>s 1 / 2, indicating that the fitness is low q A follower did not get food and had to fly to other places to find food; The alerter is the sparrow that is partially responsible for alerting during the foraging process. That is, when danger approaches, no matter the sparrow is a discoverer or a tracker, it will abandon the current food and move to a new location. The formula is as follows: , Where: q Only the number of sparrows aware of the danger is traversed, i.e. the number of alerters, not the entire population; X t,b After the tth iteration, the sparrow with the best fitness value is in the global optimal position; B is the step size control parameter, which is a random number that follows a normal distribution with a mean of 0 and a variance of 1; K is a random number, K∈[-1,1]; e is a constant; f q is the fitness value of the sparrow individual; f g and f w are the global best and worst fitness values, respectively. The above updates the positions of the discoverer, follower and sentinel in sequence. If there is no optimal sparrow position, continue updating until the stopping condition is met.

7. The laboratory fire detection method of multi-node sensor data fusion according to claim 6 is characterized in that: The specific steps of optimizing the extreme learning machine using the sparrow search algorithm are as follows: S1, normalize the data and divide it into training set and test set; S2, initializes the number of iterations, population size, and the ratio of discoverers to followers of the sparrow optimization algorithm; S3, establish the extreme learning machine network structure and set the network parameters; S4, calculate the fitness value of the sparrow group and find the best fitness value f g and the worst fitness value f w , update the positions of the discoverer, follower and sentinel in order. If the updated position is better, replace it, otherwise return and continue to update the positions of the discoverer, follower and sentinel in order; S5, continuously iteratively updates the position of the sparrows until the set maximum number of iterations is reached, and the sparrow with the lowest fitness is regarded as the optimal solution; S6, after completing the iteration, the obtained optimal weights and thresholds are applied to the extreme learning machine to form the SSA-ELM model.

8. The laboratory fire detection method of multi-node sensor data fusion according to claim 1 is characterized in that: The decision layer includes four decision factors, namely, the probability of open fire, the probability of smoldering fire, the duration of fire and the danger level of equipment. By combining the decision factors and fuzzy processing, the final result is determined. The duration of fire T is determined by the following formula: T ( h )＝[ T ( h +1)+1]× V [ Y i ( x )- T d ]， In the formula, h Indicates the current time period. h +1 indicates the next time period, v represents the probability of open fire and smoldering fire output from the feature layer, V (•) is a step function, T d The threshold for the alarm.

9. The laboratory fire detection method of multi-node sensor data fusion according to claim 1 is characterized in that: In the decision-making layer feature fusion process, the input quantity is first fuzzified. The input quantity includes the probability of open fire, the probability of smoldering fire, the duration of fire and the equipment hazard level. The fuzzy set is divided into positive small, positive medium and positive large. The membership function adopts trigonometric function. The output fuzzy set is divided into none, small, medium and large. The membership function adopts trigonometric function.

10. The laboratory fire detection method of multi-node sensor data fusion according to claim 1, characterized in that: The decision layer feature fusion needs to be defuzzified after fuzzification, and the corresponding formula is as follows: ， In the formula, U Represents the output value obtained after defuzzification, which is used for further analysis. u ( z ) is the membership function of the output, z is the output variable of the fuzzy rule, and finally the defuzzified U Make multi-level judgments. U When <0.25, it is judged as no fire; When 0.25≤ U When <0.5, it is judged as warning; When 0.5≤ U When <0.75, it is judged as an alarm; when U When ≥0.75, it is judged as a serious alarm.

Images