CN115569338A - Multi-class fire early warning data distributed training and local fire extinguishing method and system - Google Patents
Multi-class fire early warning data distributed training and local fire extinguishing method and system Download PDFInfo
- Publication number
- CN115569338A CN115569338A CN202211398580.5A CN202211398580A CN115569338A CN 115569338 A CN115569338 A CN 115569338A CN 202211398580 A CN202211398580 A CN 202211398580A CN 115569338 A CN115569338 A CN 115569338A
- Authority
- CN
- China
- Prior art keywords
- fire
- detector
- data
- time
- detectors
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A62—LIFE-SAVING; FIRE-FIGHTING
- A62C—FIRE-FIGHTING
- A62C31/00—Delivery of fire-extinguishing material
- A62C31/02—Nozzles specially adapted for fire-extinguishing
- A62C31/03—Nozzles specially adapted for fire-extinguishing adjustable, e.g. from spray to jet or vice versa
-
- A—HUMAN NECESSITIES
- A62—LIFE-SAVING; FIRE-FIGHTING
- A62C—FIRE-FIGHTING
- A62C37/00—Control of fire-fighting equipment
- A62C37/36—Control of fire-fighting equipment an actuating signal being generated by a sensor separate from an outlet device
- A62C37/38—Control of fire-fighting equipment an actuating signal being generated by a sensor separate from an outlet device by both sensor and actuator, e.g. valve, being in the danger zone
- A62C37/40—Control of fire-fighting equipment an actuating signal being generated by a sensor separate from an outlet device by both sensor and actuator, e.g. valve, being in the danger zone with electric connection between sensor and actuator
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Public Health (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Fire Alarms (AREA)
Abstract
The invention relates to a multi-class fire early warning data distributed training and local fire extinguishing method, which comprises the following steps: the main detector collects data when a fire disaster occurs; the fire-fighting controller wirelessly transmits data to an edge server, and a fire judgment model is established on the edge server and distributed training is carried out; when the trained fire judgment model is not obtained by the edge server, the fire fighting controller judges the fire through the fire judgment threshold table, the fire point detected by the main detector is taken as the circle center, the control valve of the spray head is opened, and fire judgment parameters are sent to surrounding detectors; and when the edge server obtains the trained fire judgment model, respectively judging the fire on all the detectors and the edge server according to the optimal division point to obtain fire judgment parameters. The invention also discloses a multi-class fire early warning data distributed training and local fire extinguishing system. The invention can protect the disaster object to the maximum extent and save the fire extinguishing agent on the premise of high-efficiency and accurate fire extinguishing.
Description
Technical Field
The invention relates to the technical field of local fire extinguishing, in particular to a multi-class fire early warning data distributed training and local fire extinguishing method and system.
Background
Fire extinguishing systems are classified according to the application mode and can be divided into total flooding fire extinguishing systems and partial application fire extinguishing systems. For the fire disaster of large space or long and narrow space, such as cable interlayer, comprehensive pipe gallery and cable trench, the local application fire extinguishing system is adopted to directly spray gas to the protected object at the designed spray rate in the specified time, so as to form local high concentration around the protected object and keep for a certain time to achieve the purpose of fire extinguishing.
The nozzles of the local application fire extinguishing system are uniformly arranged around the protected object, and when a fire disaster occurs, the fire extinguishing agent is directly and intensively sprayed onto the protected object, so that the fire extinguishing agent can cover the outer surface of the whole protected object, namely, the higher concentration of the fire extinguishing agent gas is achieved in the local range around the protected object to extinguish the fire. The existing fire extinguishing agent dosage calculation of the local application fire extinguishing system adopts an action area method or full coverage of protected objects, and the method can simply and quickly realize fire extinguishing, but also has some problems: firstly, the amount of fire extinguishing agent calculated by adopting an action area method or fully covering a protected object is much larger than the required amount for extinguishing fire at an actual ignition point, thereby causing the waste of the fire extinguishing agent; on the other hand, the conventional area method or the protection method in which the object to be protected is covered with a full coating allows the fire extinguishing agent to be directly sprayed to an area not subject to a fire risk, thereby causing contamination of the object not subject to the fire risk. In addition, most of the existing fire extinguishing methods are judged based on sensor threshold values, and the mode can not be well applicable to local fire extinguishing.
Disclosure of Invention
The invention aims to provide a multi-class fire early warning data distributed training and local fire extinguishing method which can protect the affected objects to the maximum extent, save fire extinguishing agents, improve the accuracy of fire judgment and enable the local fire extinguishing method to have higher response speed on the premise of high-efficiency and accurate fire extinguishing.
In order to achieve the purpose, the invention adopts the following technical scheme: a multi-class fire early warning data distributed training and local fire extinguishing method comprises the following steps in sequence:
(1) The main detector collects the data of wind speed, CO concentration and temperature when a fire disaster occurs;
(2) The main detector transmits acquired data to the fire-fighting controller, the fire-fighting controller wirelessly transmits the received data to the edge server, and a fire judgment model is established on the edge server for distributed training;
(3) When the trained fire judgment model is not obtained by the edge server, the fire-fighting controller judges the fire through a built-in fire judgment threshold table and sends fire judgment parameters to the main detector, meanwhile, the fire-fighting controller starts a control valve of a nozzle on the main detector by taking an ignition point detected by the main detector as a circle center, the main detector sends the fire judgment parameters to the surrounding detectors, the surrounding detectors collect data of wind speed, CO concentration and temperature when a fire occurs after receiving the fire judgment parameters and send the data to the fire-fighting controller, and the fire-fighting controller judges the fire through the built-in fire judgment threshold table and determines whether to start a control valve of the nozzle on the surrounding detector; when fire extinguishment is carried out, if the fire-fighting controller judges that a fire disaster is extinguished, the fire-fighting controller closes the control valves of the spray heads on all the detectors;
(5) When the edge server obtains the trained fire judgment model, the edge server sends the fire judgment model to all detectors, wherein all the detectors comprise a main detector and peripheral detectors, and fire judgment is respectively carried out on all the detectors and the edge server according to the optimal division point to obtain fire judgment parameters; the edge server sends the fire judgment parameters to the fire-fighting controller; the fire-fighting controller starts a control valve of a spray head on a main detector by taking an ignition point detected by the main detector as a circle center, the main detector sends fire judgment parameters to surrounding detectors, the surrounding detectors collect data of wind speed, CO concentration and temperature when a fire disaster happens after receiving the fire judgment parameters and send the data to the fire-fighting controller, and the fire-fighting controller judges the fire through a built-in fire judgment threshold table and determines whether to start the control valve of the spray head on the surrounding detector; when fire is extinguished, if the fire control controller judges that the fire is extinguished, the fire control controller closes the control valves of the nozzles on all the detectors.
The step (2) specifically comprises the following steps:
(2a) Establishing a fire judgment model, wherein the fire judgment model adopts a non-premixed combustion model, namely, the influence of turbulence is considered through a probability density function, a mixed fraction transport equation and a conservation scalar equation are solved, then the concentration of each component is deduced from the predicted mixed fraction distribution, the fire condition is further judged, and a thermal radiation model is adopted to calculate the fire spreading trend;
(2b) On the basis of establishing a fire judgment model, data of wind speed, CO concentration and temperature collected by a main detector when a fire disaster occurs are input to the fire judgment model for training, and the trained fire judgment model is obtained.
In the step (4), the step of respectively performing fire judgment on all the detectors and the edge server according to the optimal division point to obtain fire judgment parameters specifically comprises the following steps:
(4a) Collecting data of wind speed, CO concentration and temperature when a fire disaster occurs by all detectors, namely first data, selecting a layer of neural network in a fire condition judgment model as a prediction division point, wherein one part of the fire condition judgment model before the layer of neural network is executed by all detectors, and the other part of the fire condition judgment model after the layer is executed by an edge server; inputting first data acquired by all detectors into a fire judgment model on the detectors, executing the first data to a prediction division point of the fire judgment model, stopping executing the first data, outputting a result, namely second data, and simultaneously obtaining first processing time, wherein the first processing time is the time for obtaining the second data from the first data; the detector uploads the second data to the fire-fighting controller, and the fire-fighting controller uploads the received second data to the edge server to obtain first transmission time, wherein the first transmission time refers to the time when the detector uploads the second data until the edge server receives the second data; the edge server receives the uploaded second data, executes a fire judgment model after the prediction division point to obtain fire judgment parameters, and the period from the time when the edge server receives the uploaded second data to the time when the fire judgment model is executed is the second processing time; the edge server transmits the fire judgment parameters back to the fire-fighting controller, and the transmission time is second transmission time;
(4b) The method comprises the steps of obtaining an optimal division point of a fire judgment model based on first processing time, second processing time, first transmission time and second transmission time, and dividing the fire judgment model in all the detectors and the edge server at the optimal division point, wherein the fire judgment model is a first block before the optimal division point and a second block after the optimal division point, the first block of the fire judgment model processes first data in the detectors, and the second block of the fire judgment model processes second data in the edge server.
In step (4 b), the obtaining the optimal division point of the fire determination model based on the first processing time, the second processing time, the first transmission time, and the second transmission time specifically includes:
definition of
It represents in the ith time slot, i =1 … n, the first data produced by the jth detector within the corresponding deployment space, j =1 … m, defining
The total time required for the fire determination by this data is represented, and this total time is composed of four parts: first processing time
First transmission time
Second processing time
And a second transmission time
The minimum is obtained by
The value:
definition of
Representing the optimal division point of a fire judgment model which is taken by a detector j in the corresponding deployment space in the i time slot; definition C i [j]In the time slot i, the minimum time value of the detector j is obtained; first data generated for the first detector j =1 in the first time slot i =1
At the minimum
Selecting the division point on the basis of the first detector to obtain the fire judgment time of the first detector
Optimal division point of fire judgment model
For the first data generated on the second detector i.e. j =2 of the first time slot i =1
On the basis, the optimal division point of the fire condition judgment model is selected
And obtaining the time of judging the fire
Wherein
Means that the first score is satisfiedWith C1[1 ]]Minimum found on premise
By analogy, the first data generated for a certain detector q, j = q of the first time slot
The optimal division point of the corresponding fire judgment model is
The time of fire judgment is
The optimal segmentation point set of the fire judgment model in all the detectors in the first time slot is obtained after the implementation is finished once
And total fire determination time
When a second time slot starts, if a certain detector q in the previous time slot carries out fire judgment and then finds that a fire disaster happens in an area in charge of the certain detector q, the fire judgment needs to be preferentially carried out in the round, and the fire judgment is used as a first detector of the time slot; if the fire occurs in the area in charge of the detectors, sequencing the detectors in sequence according to the severity of the fire, and if the fire occurs in the area in charge of the detectors, dividing the severity of the fire according to the threshold range of the fire judgment threshold table, otherwise, executing the operation according to the sequence of the last time slot; therefore, the division point and the fire determination time of the fire determination model of the first i =1 detector in the second time slot i =2 are respectively
And
the second detector, i.e. the best division point and the fire judgment time of the fire judgment model with j =2 are respectively
And
by analogy, the optimal segmentation point set of the fire judgment model in the second time slot is obtained
And total fire determination time
And analogizing the subsequent time slots to obtain the optimal division point set X = { X ] of the fire judgment model of all the detectors in the final total time slot 1 ,X 2 ,……,X n And the fire decision time in the total time slot
Another object of the present invention is to provide a system for distributed training of multi-class fire warning data and local fire extinguishing method, comprising:
the main detector is used for collecting and uploading parameters of wind speed, CO concentration and temperature when a fire disaster occurs, and the detector closest to a fire point is the main detector;
the fire-fighting controller is used for receiving the data uploaded by the detector, uploading the data to the edge server, and transmitting the received information sent by the edge server back to the detector, and the fire-fighting controller is internally provided with a fire condition judgment threshold table;
the edge server is used for receiving parameters of wind speed, CO concentration and temperature and corresponding fire extinguishing operation when a fire disaster occurs and training to obtain a fire condition judgment model capable of judging a fire condition and sending a corresponding fire extinguishing instruction;
the spray head is used for spraying the fire extinguishing agent, has the spraying time and the spraying force of different gears, is controlled by a fire-fighting controller, and is arranged on the main detector and the peripheral detectors;
and the surrounding detector is used for collecting and uploading parameters of wind speed, CO concentration and temperature when a fire disaster occurs.
Setting fire extinguishing subareas according to the potential fire point condition of a space, deploying a main detector, a fire-fighting controller and peripheral detectors in a fire extinguishing area, remotely arranging an edge server, wherein the main detector, the peripheral detectors and the fire-fighting controller are in wired connection with each other, and the fire-fighting controller is in wireless communication with the edge server.
According to the technical scheme, the beneficial effects of the invention are as follows: firstly, compared with the traditional local application fire extinguishing system, the invention arranges the detector and the spray head in a way of arranging the fire extinguishing subareas, implements the corresponding fire extinguishing method according to the actual fire, and can protect the disaster-affected object to the maximum extent and save the fire extinguishing agent on the premise of high-efficiency and accurate fire extinguishing; secondly, the detector can collect actual parameters of a fire extinguishing area to carry out neural network training to obtain a fire judgment model, so that the accuracy of fire judgment is improved; thirdly, the speed of the fire judgment process is accelerated by adopting a model segmentation method, and the designed optimal segmentation point algorithm can enable the fire judgment model to be executed on a detector and an edge server in a blocking mode to accelerate the speed of the overall fire judgment process, so that the local fire extinguishing method has higher response speed.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a schematic view of a large space partial fire;
FIG. 3 is a schematic view of a partial fire in an elongated space;
fig. 4 is a flow chart of the operation of the main and peripheral detectors.
Detailed Description
As shown in fig. 1 and 2, a multi-category fire warning data distributed training and local fire extinguishing method includes the following steps:
(1) The main detector collects the data of wind speed, CO concentration and temperature when a fire disaster occurs;
(2) The main detector transmits the acquired data to the fire-fighting controller, the fire-fighting controller wirelessly transmits the received data to the edge server, and a fire judgment model is established on the edge server for distributed training;
(3) When the trained fire judgment model is not obtained by the edge server, the fire-fighting controller judges the fire through a built-in fire judgment threshold table and sends fire judgment parameters to the main detector, meanwhile, the fire-fighting controller starts a control valve of a nozzle on the main detector by taking an ignition point detected by the main detector as a circle center, the main detector sends the fire judgment parameters to the surrounding detectors, the surrounding detectors collect data of wind speed, CO concentration and temperature when a fire occurs after receiving the fire judgment parameters and send the data to the fire-fighting controller, and the fire-fighting controller judges the fire through the built-in fire judgment threshold table and determines whether to start a control valve of the nozzle on the surrounding detector; when fire is extinguished, if the fire controller judges that the fire is extinguished, the fire controller closes the control valves of the nozzles on all the detectors;
(6) When the edge server obtains the trained fire judgment model, the edge server sends the fire judgment model to all detectors, wherein all the detectors comprise a main detector and peripheral detectors, and fire judgment is respectively carried out on all the detectors and the edge server according to the optimal division point to obtain fire judgment parameters; the edge server sends the fire judgment parameters to the fire-fighting controller; the fire-fighting controller starts a control valve of a nozzle on a main detector by taking an ignition point detected by the main detector as a circle center, the main detector sends fire judgment parameters to surrounding detectors, the surrounding detectors collect data of wind speed, CO concentration and temperature when a fire disaster occurs after receiving the fire judgment parameters and send the data to the fire-fighting controller, and the fire-fighting controller judges the fire through a built-in fire judgment threshold table and determines whether to start the control valve of the nozzle on the surrounding detector; when fire is extinguished, if the fire control controller judges that the fire is extinguished, the fire control controller closes the control valves of the nozzles on all the detectors.
The step (2) specifically comprises the following steps:
(2a) Establishing a fire judgment model, wherein the fire judgment model adopts a non-premixed combustion model, namely, the influence of turbulence is considered through a probability density function, a mixed fraction transport equation and a conservation scalar equation are solved, then the concentration of each component is deduced from the predicted mixed fraction distribution, the fire condition is further judged, and a thermal radiation model is adopted to calculate the fire spreading trend;
(2b) On the basis of establishing a fire judgment model, data of wind speed, CO concentration and temperature collected by a main detector when a fire disaster occurs are input to the fire judgment model for training, and the trained fire judgment model is obtained.
In the step (4), the step of respectively performing fire judgment on all the detectors and the edge server according to the optimal division point to obtain fire judgment parameters specifically comprises the following steps:
(4a) Collecting data of wind speed, CO concentration and temperature when a fire disaster occurs by all detectors, namely first data, selecting a layer of neural network in a fire condition judgment model as a prediction division point, wherein one part of the fire condition judgment model before the layer of neural network is executed by all detectors, and the other part of the fire condition judgment model after the layer is executed by an edge server; inputting first data acquired by all detectors into a fire judgment model on the detectors, executing the first data to a prediction division point of the fire judgment model, stopping executing the first data, outputting a result, namely second data, and simultaneously obtaining first processing time, wherein the first processing time is the time for obtaining the second data from the first data; the detector uploads the second data to the fire-fighting controller, and the fire-fighting controller uploads the received second data to the edge server to obtain first transmission time, wherein the first transmission time refers to the time when the detector uploads the second data until the edge server receives the second data; the edge server receives the uploaded second data, executes a fire judgment model after the prediction division point to obtain fire judgment parameters, and the period from the time when the edge server receives the uploaded second data to the time when the fire judgment model is executed is the second processing time; the edge server transmits the fire judgment parameters back to the fire-fighting controller, and the transmission time is second transmission time;
(4b) The method comprises the steps of obtaining an optimal division point of a fire judgment model based on first processing time, second processing time, first transmission time and second transmission time, and dividing the fire judgment model in all the detectors and the edge server at the optimal division point, wherein the fire judgment model is a first block before the optimal division point and a second block after the optimal division point, the first block of the fire judgment model processes first data in the detectors, and the second block of the fire judgment model processes second data in the edge server.
In the step (4 b), the obtaining of the optimal division point of the fire judgment model based on the first processing time, the second processing time, the first transmission time and the second transmission time specifically includes:
definition of
It represents in the ith time slot, i =1 … n, the first data produced by the jth detector within the corresponding deployment space, j =1 … m, defining
The total time required for the fire determination by this data is expressed, and this total time is composed of four parts: first processing time
First transmission time
Second processing time
And a second transmission time
The minimum is obtained by
The value:
definition of
Representing the optimal division point of a fire judgment model which is taken by a detector j in the corresponding deployment space in the i time slot; definition C i [j]In the time slot i, the minimum time value of the detector j is obtained; first data generated for the first detector j =1 in the first time slot i =1
In the minimization of
Selecting the division point on the basis of the first detector to obtain the fire judgment time of the first detector
Optimal division point of fire judgment model
For the first data generated on the second detector i.e. j =2 in the first time slot i =1
On the basis, the optimal segmentation point of the fire judgment model is selected
And obtaining the time of judging the fire
Wherein
Is expressed in the sense that the first assignment C1[1 ] is satisfied]Minimum found on premise
By analogy, the first data generated for a certain detector q, j = q of the first time slot
The optimal division point of the corresponding fire judgment model is
The time of fire judgment is
The optimal segmentation point set of the fire judgment model in all the detectors in the first time slot is obtained after the implementation is finished once
And total fire determination time
When a second time slot starts, if a certain detector q in the previous time slot carries out fire judgment and then finds that a fire disaster happens in an area in charge of the certain detector q, the fire judgment needs to be preferentially carried out in the round, and the fire judgment is used as a first detector of the time slot; if the fire occurs in the area in charge of the detectors, sequencing the detectors in sequence according to the severity of the fire, and if the fire occurs in the area in charge of the detectors, dividing the severity of the fire according to the threshold range of the fire judgment threshold table, otherwise, executing the operation according to the sequence of the last time slot; thus, the fire for the first, j =1 detector of the second time slot, i =2The division point and the fire judgment time of the judgment model are respectively
And
the second detector, i.e. the best division point and the fire judgment time of the fire judgment model with j =2 are respectively
And
by analogy, the optimal segmentation point set of the fire judgment model in the second time slot is obtained
And total fire determination time
And analogizing the subsequent time slots to obtain the optimal segmentation point set X = { X) of the fire determination model of all detectors in the final total time slot 1 ,X 2 ,……,X n And the fire decision time in the total time slot
The system comprises:
the main detector is used for collecting and uploading parameters of wind speed, CO concentration and temperature when a fire disaster occurs, and the detector closest to a fire point is the main detector;
the fire-fighting controller is used for receiving the data uploaded by the detector, uploading the data to the edge server, and transmitting the received information sent by the edge server back to the detector, and the fire-fighting controller is internally provided with a fire condition judgment threshold table;
the edge server is used for receiving parameters of wind speed, CO concentration and temperature and corresponding fire extinguishing operation when a fire disaster occurs and training to obtain a fire condition judgment model capable of judging a fire condition and sending a corresponding fire extinguishing instruction;
the spray head is used for spraying the fire extinguishing agent, has the spraying time and the spraying strength of different gears, is controlled by a fire-fighting controller, and is arranged on the main detector and the peripheral detectors;
and the surrounding detector is used for collecting and uploading parameters of wind speed, CO concentration and temperature when a fire disaster happens.
Setting fire extinguishing subareas according to the potential fire point condition of the space, deploying a main detector, a fire-fighting controller and peripheral detectors in a fire extinguishing area, remotely arranging an edge server, and performing wired connection among the main detector, the peripheral detectors and the fire-fighting controller, and performing wireless communication between the fire-fighting controller and the edge server.
The invention is further described below with reference to fig. 1 to 4.
The detector collects and uploads parameters such as wind speed, CO concentration and temperature when a fire disaster occurs, a single chip microcomputer is arranged in the detector, inference operation, namely fire judgment operation can be executed, and the detectors can be linked; the main detector can only broadcast the fire to the surrounding detectors without interaction with the surrounding detectors, all with completely independent detection and algorithm execution.
As shown in fig. 4, the detector that detects the fire first serves as a main detector, and in a time period between when the main detector detects that the fire is reduced to a threshold value from the room temperature, the main detector broadcasts the fire determination parameter of the main detector at a frequency of once per second; and the surrounding detectors determine the control of the surrounding detectors on the nozzles according to the weighted average of the received fire judgment parameters and the fire judgment parameters obtained by calculation.
The fire-fighting controller can automatically receive the detector signal and can automatically clear redundant information under normal conditions; when the fire controller receives an abnormal signal of a certain detector, the fire controller can automatically identify the true and false signals according to signals of other parallel detectors and automatically clear false signals. When a fire occurs, fire judgment parameters obtained from a fire judgment threshold table, namely table 1 or an edge server are transmitted to a detector, and the actions of pre-starting, closing and the like of a fire extinguishing nozzle are controlled until the temperature is reduced to below 50 ℃ and the control valve of the nozzle is kept closed for 10 minutes, wherein the table 1 is as follows:
TABLE 1
The spray head spray is controlled by two factors: firstly, the spraying force of the spray heads is adjustable in multiple gears, and more spraying schemes are provided for combined spraying of different spray heads; and secondly, during the injection time, all the nozzles can be closed when the temperature is lower than the threshold value, the fire behavior can be controlled, and the injection gear can be adjusted down or part of the nozzles can be closed at a proper time according to the fire condition.
During the algorithm execution, the data transmission between the probe and the edge server and the data processing process of the edge server bring certain time consumption. In practical situations, in order to ensure the timeliness of the fire response, it is desirable to keep the response time of the fire extinguishing apparatus as short as possible, which requires that the time of both parts be as short as possible. Since the detector and the fire controller are connected by wires, the transmission time between the detector and the fire controller can be ignored, so the data uploading and returning time only considers the wireless transmission time between the fire controller and the edge server. The time of wireless transmission is mainly related to the network situation, and it is more important to reduce the data processing time as much as possible compared with the wireless transmission time. In order to reduce the time consumption of the data processing process, it is considered to accelerate the inference process by using a model segmentation method in the inference stage, i.e., the fire decision stage. Specifically, the fire judgment model needing calculation is divided into two parts, one part is executed on a detector with weak calculation power, the other part is executed on an edge server with strong calculation power, and the calculation of the two parts is parallel.
In summary, the following time consumption needs to be considered in the case of introducing model segmentation. I.e. the processing time t of the data acquired at the detector (first processing time) s Data upload (first transmission time) t u And a backtransmission time (second transmission time) t d And a processing time (second processing time) t of the data at the edge server w 。
In the process of model segmentation, when the original data volume or the intermediate data volume is large and complex, a detector or an edge server may not be able to process in time, and thus the resulting data waiting time may affect the total time of the inference process. Therefore, it is necessary to find a suitable segmentation point for model segmentation, so as to minimize the sum of the above time.
Example one
For a large-space local fire, as shown in fig. 2, when a fire occurs at a fire point, a detector sends out an alarm control signal, the system receives the position of the fire, firstly, a c2/d2/d3/d4/c4/c3/c2 spray head is started to extinguish the fire, meanwhile, the detector in the area monitors that the temperature of the area is reduced in real time and is maintained to be lower than 100 ℃, the fire is considered to be controlled, the spreading phenomenon does not occur, the detector monitors that the temperature of the area is maintained to be lower than 50 ℃ in real time for 10 minutes, the fire is considered to be extinguished, and the control system sends out an instruction to stop spraying the fire extinguishing agent. If the detector in the area monitors that the problem in the area continuously rises in real time, and the fire is indicated to have a spreading trend, the b1/c1/d1/e1/e2/e3/e4/e5/d5/c5/b5/b4/b3/b2 spray head is started to start fire extinguishing, when the temperature of the area is kept below 50 ℃ for 10 minutes, the fire is considered to be extinguished, the control system sends an instruction to stop spraying the fire extinguishing agent, and the like.
Example two
For a local fire in a narrow and long space, as shown in fig. 3, when a fire occurs at a fire point, a detector sends out an alarm control signal, the system receives the position of the fire, firstly, a b/c spray head is opened to extinguish the fire, meanwhile, the detector in the area monitors that the temperature of the area is reduced in real time and is kept below 100 ℃, the fire is considered to be controlled and not to be spread, the detector monitors that the temperature of the area is kept below 50 ℃ in real time for 10 minutes, the fire is considered to be extinguished, and the control system sends out an instruction to stop spraying the fire extinguishing agent. If the problem of the area is monitored by a detector in the area in real time and continuously rises, and the situation that the fire disaster tends to spread is indicated, the a … d spray head is opened to extinguish the fire, the fire disaster is considered to be extinguished when the temperature of the area is kept for 10 minutes below 50 ℃, the control system sends an instruction to stop spraying the fire extinguishing agent, and the like.
In conclusion, the invention arranges the detectors and the spray heads in a way of arranging the fire extinguishing subareas, and implements the corresponding fire extinguishing method according to the actual fire situation, thereby protecting the object from the disaster to the maximum extent and saving the fire extinguishing agent on the premise of high-efficiency and accurate fire extinguishing; the detector can collect the actual parameters of the fire extinguishing area to carry out neural network training to obtain a fire judgment model, so that the accuracy of fire judgment is improved; the invention adopts a model segmentation method to accelerate the speed of the fire judgment process, and the designed optimal segmentation point algorithm can lead the fire judgment model to be executed on the detector and the edge server in a blocking way to accelerate the speed of the whole fire judgment process, so that the local fire extinguishing method has higher response speed.
Claims (6)
1. A multi-class fire early warning data distributed training and local fire extinguishing method is characterized in that: the method comprises the following steps in sequence:
(1) The main detector collects the data of wind speed, CO concentration and temperature when a fire disaster occurs;
(2) The main detector transmits the acquired data to the fire-fighting controller, the fire-fighting controller wirelessly transmits the received data to the edge server, and a fire judgment model is established on the edge server for distributed training;
(3) When the trained fire judgment model is not obtained by the edge server, the fire-fighting controller judges the fire through a built-in fire judgment threshold table and sends fire judgment parameters to the main detector, meanwhile, the fire-fighting controller starts a control valve of a nozzle on the main detector by taking an ignition point detected by the main detector as a circle center, the main detector sends the fire judgment parameters to the surrounding detectors, the surrounding detectors collect data of wind speed, CO concentration and temperature when a fire occurs after receiving the fire judgment parameters and send the data to the fire-fighting controller, and the fire-fighting controller judges the fire through the built-in fire judgment threshold table and determines whether to start a control valve of the nozzle on the surrounding detector; when fire is extinguished, if the fire controller judges that the fire is extinguished, the fire controller closes the control valves of the nozzles on all the detectors;
(4) When the edge server obtains the trained fire judgment model, the edge server sends the fire judgment model to all detectors, wherein all the detectors comprise a main detector and peripheral detectors, and fire judgment is respectively carried out on all the detectors and the edge server according to the optimal division point to obtain fire judgment parameters; the edge server sends the fire judgment parameters to the fire-fighting controller; the fire-fighting controller starts a control valve of a spray head on a main detector by taking an ignition point detected by the main detector as a circle center, the main detector sends fire judgment parameters to surrounding detectors, the surrounding detectors collect data of wind speed, CO concentration and temperature when a fire disaster happens after receiving the fire judgment parameters and send the data to the fire-fighting controller, and the fire-fighting controller judges the fire through a built-in fire judgment threshold table and determines whether to start the control valve of the spray head on the surrounding detector; when fire is extinguished, if the fire control controller judges that the fire is extinguished, the fire control controller closes the control valves of the nozzles on all the detectors.
2. The multi-class fire early warning data distributed training and local fire extinguishing method according to claim 1, wherein: the step (2) specifically comprises the following steps:
(2a) Establishing a fire judgment model, wherein the fire judgment model adopts a non-premixed combustion model, namely, the influence of turbulence is considered through a probability density function, a mixed fraction transport equation and a conservation quantity equation are solved, then the concentration of each component is deduced from the predicted mixed fraction distribution, the fire condition is further judged, and a heat radiation model is adopted to calculate the fire spreading trend;
(2b) On the basis of establishing a fire judgment model, data of wind speed, CO concentration and temperature collected by a main detector when a fire disaster occurs are input to the fire judgment model for training, and the trained fire judgment model is obtained.
3. The multi-class fire early warning data distributed training and local fire extinguishing method according to claim 1, wherein: in the step (4), the step of respectively performing fire judgment on all the detectors and the edge server according to the optimal division point to obtain fire judgment parameters specifically comprises the following steps:
(4a) Collecting data of wind speed, CO concentration and temperature when a fire disaster occurs by all detectors, namely first data, selecting a layer of neural network in a fire judgment model as a prediction division point, wherein one part of the fire judgment model before the layer of neural network is executed by all the detectors, and the other part of the fire judgment model after the layer is executed by an edge server; inputting first data acquired by all detectors into a fire judgment model on the detectors, executing the first data to a prediction division point of the fire judgment model, stopping executing the first data, outputting a result, namely second data, and simultaneously obtaining first processing time, wherein the first processing time is the time for obtaining the second data from the first data; the detector uploads the second data to the fire-fighting controller, and the fire-fighting controller uploads the received second data to the edge server to obtain first transmission time, wherein the first transmission time refers to the time when the detector uploads the second data until the edge server receives the second data; the edge server receives the uploaded second data, executes a fire judgment model after the prediction division point to obtain fire judgment parameters, and the period from the time when the edge server receives the uploaded second data to the time when the fire judgment model is executed is the second processing time; the edge server transmits the fire judgment parameters back to the fire-fighting controller, and the transmission time is second transmission time;
(4b) The method comprises the steps of obtaining an optimal division point of a fire judgment model based on first processing time, second processing time, first transmission time and second transmission time, and dividing the fire judgment model in all the detectors and the edge server at the optimal division point, wherein the fire judgment model is a first block before the optimal division point and a second block after the optimal division point, the first block of the fire judgment model processes first data in the detectors, and the second block of the fire judgment model processes second data in the edge server.
4. The multi-class fire early warning data distributed training and local fire extinguishing method according to claim 3, wherein: in step (4 b), the obtaining the optimal division point of the fire determination model based on the first processing time, the second processing time, the first transmission time, and the second transmission time specifically includes:
definition of
It represents in the ith time slot, i =1 … n, the first data produced by the jth detector in the corresponding deployment space, j =1 … m, defining
The total time required for the fire determination by this data is represented, and this total time is composed of four parts: first processing time
First transmission time
Second processing time
And a second transmission time
The minimum is obtained by
The value:
definition of
Representing the optimal division point of a fire judgment model which is taken by a detector j in the corresponding deployment space in the i time slot; definition C i [j]In the time slot i, the minimum time value of the detector j is obtained; first data generated for the first detector j =1 in the first time slot i =1
In the minimization of
Selecting the division point on the basis of the first detector to obtain the fire judgment time of the first detector
Optimal division point of fire judgment model
For the first data generated on the second detector i.e. j =2 in the first time slot i =1
On the basis, the optimal division point of the fire condition judgment model is selected
And obtaining the time of judging the fire
Wherein
Is expressed in the sense that the first assignment C1[1 ] is satisfied]Minimum found on premise
By analogy, the first data generated for a certain detector q, j = q of the first time slot
The optimal division point of the corresponding fire judgment model is
The time of fire judgment is
The optimal segmentation point set of the fire judgment model in all the detectors in the first time slot is obtained after the implementation is finished once
And total fire determination time
When a second time slot starts, if a certain detector q in the previous time slot carries out fire judgment and then finds that a fire disaster happens in an area in charge of the certain detector q, the fire judgment needs to be preferentially carried out in the round, and the fire judgment is used as a first detector of the time slot; if the fire occurs in the area where the detectors are responsible, the detectors are sequenced according to the severity of the fire, and the severity of the fire is determined according to the severityDividing the threshold range of the fire judgment threshold table, otherwise, executing the operation according to the sequence of the last time slot; therefore, the division point and the fire determination time of the fire determination model of the first i =1 detector of the second time slot i =2 are respectively
And
the second detector, namely the best division point of the fire judgment model with j =2 and the fire judgment time are respectively
And
by analogy, the optimal segmentation point set of the fire judgment model in the second time slot is obtained
And total fire determination time
And analogizing the subsequent time slots to obtain the optimal division point set X = { X ] of the fire judgment model of all the detectors in the final total time slot 1 ,X 2 ,……,X n And the fire decision time in the total time slot
5. A system for implementing the multi-class fire early warning data distributed training and local fire extinguishing method according to any one of claims 1 to 4, wherein: the method comprises the following steps:
the main detector is used for collecting and uploading parameters of wind speed, CO concentration and temperature when a fire disaster occurs, and the detector closest to a fire point is the main detector;
the fire-fighting controller is used for receiving the data uploaded by the detector, uploading the data to the edge server, and transmitting the received information sent by the edge server back to the detector, and the fire-fighting controller is internally provided with a fire condition judgment threshold table;
the edge server is used for receiving parameters of wind speed, CO concentration and temperature and corresponding fire extinguishing operation when a fire disaster occurs and training to obtain a fire condition judgment model capable of judging a fire condition and sending a corresponding fire extinguishing instruction;
the spray head is used for spraying the fire extinguishing agent, has the spraying time and the spraying force of different gears, is controlled by a fire-fighting controller, and is arranged on the main detector and the peripheral detectors;
and the surrounding detector is used for collecting and uploading parameters of wind speed, CO concentration and temperature when a fire disaster occurs.
6. The system of claim 5, wherein: setting fire extinguishing subareas according to the potential fire point condition of the space, deploying a main detector, a fire-fighting controller and peripheral detectors in a fire extinguishing area, remotely arranging an edge server, and performing wired connection among the main detector, the peripheral detectors and the fire-fighting controller, and performing wireless communication between the fire-fighting controller and the edge server.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211398580.5A CN115569338B (en) | 2022-11-09 | 2022-11-09 | Multi-category fire early warning data distributed training and local extinguishing method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211398580.5A CN115569338B (en) | 2022-11-09 | 2022-11-09 | Multi-category fire early warning data distributed training and local extinguishing method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115569338A true CN115569338A (en) | 2023-01-06 |
| CN115569338B CN115569338B (en) | 2023-10-13 |
Family
ID=84589338
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211398580.5A Active CN115569338B (en) | 2022-11-09 | 2022-11-09 | Multi-category fire early warning data distributed training and local extinguishing method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115569338B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB1474680A (en) * | 1974-12-19 | 1977-05-25 | Cerberus Ag | Fire extinguishing installations |
| JP2005168645A (en) * | 2003-12-09 | 2005-06-30 | Ohbayashi Corp | Fireproof compartment system |
| CN110046837A (en) * | 2019-05-20 | 2019-07-23 | 北京唐芯物联网科技有限公司 | A kind of fire management system based on artificial intelligence |
| CN111035872A (en) * | 2020-01-02 | 2020-04-21 | 中车青岛四方车辆研究所有限公司 | Battery box fire prevention and control system and method |
| CN111127811A (en) * | 2019-12-25 | 2020-05-08 | 上海船舶电子设备研究所(中国船舶重工集团公司第七二六研究所) | Fire early detection alarm system and method |
| CN111686392A (en) * | 2020-06-23 | 2020-09-22 | 海南科技职业大学 | Artificial intelligence fire extinguishing system is surveyed to full scene of vision condition |
| CN112156399A (en) * | 2020-07-22 | 2021-01-01 | 浙江蓝盾电工新材料科技有限公司 | Fire-fighting system for echelon early warning and multiple times of accurate discharge of lithium battery energy storage units |
| CN114528987A (en) * | 2022-02-15 | 2022-05-24 | 中国科学院上海微***与信息技术研究所 | Neural network edge-cloud collaborative computing segmentation deployment method |
| CN216676774U (en) * | 2021-12-22 | 2022-06-07 | 福建永福电力设计股份有限公司 | Water spray fire extinguishing system for transformer |
| CN114699701A (en) * | 2022-04-08 | 2022-07-05 | 中车大连机车研究所有限公司 | Fire prevention and control device capable of realizing accurate fire extinguishing |
-
2022
- 2022-11-09 CN CN202211398580.5A patent/CN115569338B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB1474680A (en) * | 1974-12-19 | 1977-05-25 | Cerberus Ag | Fire extinguishing installations |
| JP2005168645A (en) * | 2003-12-09 | 2005-06-30 | Ohbayashi Corp | Fireproof compartment system |
| CN110046837A (en) * | 2019-05-20 | 2019-07-23 | 北京唐芯物联网科技有限公司 | A kind of fire management system based on artificial intelligence |
| CN111127811A (en) * | 2019-12-25 | 2020-05-08 | 上海船舶电子设备研究所(中国船舶重工集团公司第七二六研究所) | Fire early detection alarm system and method |
| CN111035872A (en) * | 2020-01-02 | 2020-04-21 | 中车青岛四方车辆研究所有限公司 | Battery box fire prevention and control system and method |
| CN111686392A (en) * | 2020-06-23 | 2020-09-22 | 海南科技职业大学 | Artificial intelligence fire extinguishing system is surveyed to full scene of vision condition |
| CN112156399A (en) * | 2020-07-22 | 2021-01-01 | 浙江蓝盾电工新材料科技有限公司 | Fire-fighting system for echelon early warning and multiple times of accurate discharge of lithium battery energy storage units |
| CN216676774U (en) * | 2021-12-22 | 2022-06-07 | 福建永福电力设计股份有限公司 | Water spray fire extinguishing system for transformer |
| CN114528987A (en) * | 2022-02-15 | 2022-05-24 | 中国科学院上海微***与信息技术研究所 | Neural network edge-cloud collaborative computing segmentation deployment method |
| CN114699701A (en) * | 2022-04-08 | 2022-07-05 | 中车大连机车研究所有限公司 | Fire prevention and control device capable of realizing accurate fire extinguishing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115569338B (en) | 2023-10-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN209728979U (en) | A kind of intelligent building fire plant based on BIM | |
| CN110097728B (en) | Fire-fighting remote monitoring system | |
| KR101180411B1 (en) | Automatic Control Apparatus of a fire door | |
| CN110237477A (en) | A kind of firefighting monitoring system based on artificial intelligence | |
| CN110011883B (en) | Intelligent building Internet of things information marginalization analysis and local control system | |
| CN110518487A (en) | A kind of method and system for being monitored to electrification safety work | |
| CN113034833A (en) | Artificial intelligence information management system based on multisensor fuses | |
| CN117482445A (en) | Intelligent simple spraying fire early-stage suppression system based on Internet of things | |
| CN115569338B (en) | Multi-category fire early warning data distributed training and local extinguishing method and system | |
| CN114748813B (en) | Quick fire extinguishing method, system, equipment and storage medium suitable for high-rise building | |
| CN115310882A (en) | Visual fire-fighting strategy planning method and system | |
| CN111888675B (en) | Automatic fire-fighting air cushion control platform and method | |
| CN116485165B (en) | Forest fire control strategy formulation method, system and storage medium based on fusion factors | |
| CN117495110A (en) | Fire rescue risk assessment method, device, equipment and readable storage medium | |
| CN117668570A (en) | Device, method and equipment for selecting laying of fireproof cable | |
| CN110486898A (en) | Air conditioning control method, device and air-conditioning system based on User Activity rule | |
| CN115662041A (en) | Fire extinguishing system based on fire sensing and consequence prediction of transformer substation | |
| CN116822778A (en) | Intelligent fire emergency management method based on image recognition technology | |
| CN110624193B (en) | Fire control method for preventing afterburning based on Internet of things | |
| CN208506834U (en) | A kind of building security system | |
| CN111973925B (en) | Method, device and system for fire extinguishing by robot cooperation | |
| CN117592768B (en) | Fire disaster management method and system for internal places of enterprises based on big data processing | |
| CN117236671B (en) | Dynamic real-time monitoring method and system for mountain fire of power transmission line | |
| CN108325136A (en) | A kind of three-level management and control control system | |
| CN117282054B (en) | Tunnel fire-fighting linkage system based on edge calculation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |