JP2020161072A - Fire monitoring system - Google Patents

Abstract

To provide a fire monitoring system capable of determining a fire with high accuracy by combining a smell element with fire determination.SOLUTION: In a fire monitoring system, a fire sensor 14 and a smell detector 18 are installed in a target section and are connected to signal lines 12-1 and 12-2 from a receiver 10. The smell detector 18 outputs multiple types of smell detection values in accordance with adsorption of a plurality of smell substances and transmits them to the receiver 10. A smell determination unit 40 determines a smell caused by a burned material, smell of cigarette and smell of cooking as types of smell. A fire determination unit 42 determines a fire when determining, for example, a fire report and the smell caused by a burned material on the basis of the fire report by the fire sensor 14 and the smell type determined by and the smell determination unit 40, and determines a non-fire when the smell caused by a burned material is not determined even if a fire report is determined.SELECTED DRAWING: Figure 1

Description

The present invention relates to a fire monitoring system that determines and warns a fire by a combination of a fire detector and an odor detector.

Conventionally, a fire monitoring system for detecting a fire has been put into practical use by using a sensor such as a smoke detector or a heat detector that monitors a specific physical quantity such as smoke or temperature associated with the fire.

In addition, in addition to smoke sensors and heat sensors, flame sensors, CO sensors, CO ₂ sensors, etc. are provided to prevent false reports (non-fire reports) associated with smoking and cooking other than fire, and the values detected by these sensors. The fire is judged by the combination with, and the fire judgment is made more reliable.

WO 2016-136434 WO2016-031080 WO2017 / 085939A

However, in a fire judgment in which a smoke sensor or a heat sensor is combined with a flame sensor, a CO sensor, a CO ₂ sensor, etc., all of them detect a specific physical quantity such as smoke, heat, or gas associated with the fire. It is a judgment, and there is a limit compared to the fire judgment that humans make through the five senses.

By the way, in recent years, odor sensors capable of detecting and distinguishing a specific odor from a complex odor aggregate, such as the human sense of smell, have been put into practical use (Patent Documents 2 and 3). .. If there is a sensor that can detect and distinguish the odor corresponding to the human sense of smell in this way, it is possible to combine fire judgment with unprecedented judgment factors such as odor, and to quickly and surely judge a fire for cooking and cooking. False alarms due to cigarette smoke can be reliably prevented.

An object of the present invention is to provide a fire monitoring system capable of determining a fire with high accuracy by combining a fire determination with an odor element.

(Fire monitoring system)
The present invention is a fire monitoring system.
A fire detection means that notifies a fire when the smoke concentration or temperature in the monitoring space exceeds or exceeds a predetermined value.
An odor detection means that detects the odor in the surveillance space,
An odor determination means for determining the type of odor by inputting an odor detection value of the odor detection means,
A fire judgment means for judging a fire based on the presence or absence of a fire alarm by the fire detection means and the type of odor output from the odor judgment means,
Is characterized by being provided.

(Judgment of odor type by multi-layer neural network)
The odor detecting means outputs a plurality of types of odor detection values according to the adsorption of a plurality of odor substances.
The odor determination means is composed of a multi-layer neural network, and a combination of a plurality of types of odor detection values of the odor detection means which is input information and a predetermined odor type which is an expected output value is input as learning information by deep learning. When learning and inputting a plurality of types of odor detection values of the odor detecting means as input information, the odor type is estimated and output.

(Type of odor to be judged)
The multi-layer neural network of the odor determination means
Learn the burning odor, cigarette odor, and cooking odor as odor types.
When a plurality of types of odor detection values of the odor detecting means are input, the burning odor of an object, the odor of cigarettes, and the odor of cooking are estimated and output.

(Fire judgment)
The fire determining means determines a fire based on the presence or absence of a fire alarm by the fire detecting means and the presence or absence of a burning odor, a cigarette odor, and a cooking odor output from the odor determining means.

(Judgment of fire and non-fire)
The fire determination means
A fire is judged when it is judged that there is a fire alarm and the smell of burning things is judged, or when the smell of burning things, the smell of cigarettes and the smell of cooking are judged.
Other than the above, non-fire is judged.
A fire monitoring system that features that.

(Judgment of fire caution)
The fire judgment means is further
A fire caution is determined when it is determined whether or not there is a fire alarm and at least one of the smell of cigarettes and the smell of cooking is determined.

(Judgment of odor and fire by multi-layer neural network)
The odor judgment means and the fire judgment means are composed of a multi-layer neural network.
Input information such as the presence or absence of a fire alarm of the fire detection means, multiple types of odor detection values of the odor detection means, and expected output values of fire, fire caution or non-fire and a predetermined odor type as learning information. It learns by deep learning, and when the presence or absence of a fire alarm of the fire detection means and a plurality of types of odor detection values of the odor detection means are input as input information, fire, fire caution or non-fire is estimated and output.

(Place of odor judgment means and fire judgment means)
The fire detection means and the odor detection means are connected to the signal line drawn from the receiver,
The odor determining means is provided in the odor detecting means or the receiver.
The fire determination means is provided in the receiver.

(Fire detection means and odor detection means are installed in the same section)
The fire detecting means are installed in each predetermined section, and the odor detecting means is installed in the same section as the fire detecting means.

(Smell detection means installed in some sections)
Fire detection means are installed in each designated section,
Smell detection means are installed in a part of multiple compartments,
When the odor detecting means is not installed in the section where the fire is reported, the odor determining means of the receiver selects the odor detecting means installed in the surrounding section to determine the type of odor.

(Construction and learning of multi-layer neural network)
A multi-layer neural network is a coupled neural network,
Learning is performed by backpropagation based on the error between the value output when the learning information is input and the predetermined expected value.

(Basic effect)
The present invention is a fire monitoring system, wherein a fire detecting means for reporting a fire when the smoke concentration or temperature in the monitoring space exceeds or exceeds a predetermined value, and an odor detecting means for detecting an odor in the monitoring space. , A fire is determined based on the odor determination means for inputting the odor detection value of the odor detection means and determining the type of odor, the presence or absence of a fire alarm by the fire detection means, and the type of odor output from the odor determination means. Since a fire judgment means is provided, when a smoke detector or heat detector that functions as a fire detection means detects a fire, in addition to this, by judging the peculiar smell associated with the fire, tobacco, cooking, etc. It is possible to prevent erroneous fire judgment due to non-fire factors and judge fire with high accuracy.

(Effect of odor type judgment by multi-layer neural network)
Further, the odor detecting means outputs a plurality of types of odor detecting values according to the adsorption of a plurality of odor substances, and the odor determining means is composed of a multi-layered neural network and has a plurality of types of odor detecting means which are input information. When a combination of a predetermined odor type, which is an odor detection value and an expected output value, is input as learning information and learned by deep learning, and a plurality of types of odor detection values of the odor detection means are input as input information, the odor type is input. Since it is estimated and output, it is possible to detect a specific odor from a complex odor aggregate and distinguish it for fire judgment, like the human odor.

(Effect of the type of odor to be judged)
In addition, the multi-layered neural network of the odor determination means learns the burning odor, the odor of tobacco, and the odor of cooking as odor types, and when a plurality of types of odor detection values of the odor detection means are input, the object Since the burning odor, tobacco odor, and cooking odor are estimated and output, the burning odor, tobacco odor, and cooking odor of things necessary for fire judgment can be judged, and highly accurate fire judgment can be made. it can.

(Effect of fire judgment)
Further, the fire determining means determines a fire based on the presence or absence of a fire alarm by the fire detecting means, the burning odor of an object output from the odor determining means, the odor of tobacco, and the presence or absence of the odor of cooking. , Fire judgment means
(1) A fire occurs when it is determined that there is a fire alarm and the burning odor of an object is determined, or when it is determined that there is a fire alarm and the burning odor of an object, the smell of cigarettes and cooking are determined. Judging,
(2) When it is determined whether there is a fire alarm or not, and at least one of the smell of cigarettes and the smell of cooking is determined, the fire caution is determined.
(3) Except for (1) and (2), non-fire is judged.
Therefore, for example, when smoke is detected and the burning smell of an object is determined, it is definitely a fire, and when smoke is detected and the burning smell of an object is determined, the smell of cigarettes and cooking is also determined. If it is determined, it is definitely a fire, and therefore, even if there is cigarette or cooking smoke, the fire can be reliably determined.

In addition, when smoke is detected and the smell of burning things and the smell of cigarettes or cooking are determined, for example, it is assumed that the ashtray is smoldering, so it is judged that it is a fire caution and a caution alarm is output.

In addition, even if smoke is not detected, if the smell of cigarettes or cooking is determined in addition to the smell of burning things, there is a high possibility that it will lead to a fire, so it is judged that it is a fire caution and a caution alarm is output. ..

Furthermore, although smoke is detected, the burning odor of the object is not determined, and when the odor of cigarettes or cooking is determined, it is a normal smoking or cooking state, and even if smoke is detected, it is judged to be non-fire. This makes it possible to reliably prevent false reports (non-fire reports) caused by cigarettes and cooking smoke.

(Effect of odor and fire judgment by multi-layer neural network)
In addition, the odor determination means and the fire determination means are composed of a multi-layered neural network, and the presence or absence of a fire alarm of the fire detection means which is input information, and a plurality of types of odor detection values and output expected values of the odor detection means. , Fire caution or non-fire and a combination of a predetermined odor type are input as learning information and learned by deep learning, and the presence or absence of a fire alarm of the fire detection means and multiple types of odor detection values of the odor detection means are used as input information. Since a fire, fire caution or non-fire is estimated and output when input, the functions of the odor judgment means and the fire judgment means are realized by learning the multi-layered neural network, and the smoke or heat associated with the fire is reduced. In addition, by determining a predetermined odor related to a fire, a highly accurate fire determination can be made.

(Effects depending on the location of the odor judgment means and the fire judgment means)
Further, since the fire detecting means and the odor detecting means are connected to the signal line drawn from the receiver, the odor detecting means is provided in the odor detecting means or the receiver, and the fire detecting means is provided in the receiver, the fire is detected. An odor detector that functions as an odor detection means is installed in the caution area together with a smoke detector and a heat sensor that function as a means, and by transmitting an odor detection signal or an odor judgment signal to the receiver, smoke or heat can be detected. Highly accurate fire judgment can be made by the combination of.

Further, when the odor detecting means is provided in the odor detecting means, it is possible to judge the odor type provided with the multi-layer neural network on the edge side and transmit it to the receiver, and the configuration on the receiver side is simple. Can be.

(Effect of installing fire detection means and odor detection means in the same section)
In addition, since the fire detecting means is installed in each predetermined section and the odor detecting means is installed in the same section as the fire detecting means, in addition to the smoke or heat in the same section, the type of odor is determined to determine the type of fire with high accuracy. I can judge.

(Effect of installing odor detection means in some detector installation areas)
Further, the fire detecting means is installed in each predetermined section, the odor detecting means is installed in a part of a plurality of sections, and the odor detecting means of the receiver is not installed in the section where the fire is reported. In this case, the odor detection means installed in the surrounding compartments are selected to determine the type of odor, so even if odor detection means are not provided in all compartments, fire alarms and high-precision fires based on odor types It is possible to make a judgment, and it is possible to reduce the number of odor detectors that function as odor detection means compared to the number of smoke detectors and heat sensors that function as fire detection means, and reduce the system cost.

(Effects of multi-layer neural network configuration and learning)
In addition, the multi-layer neural network is a coupled neural network, and learning is performed by back propagation based on the error between the value output when the learning information is input and the predetermined expected value. Therefore, for example, the odor type In the judgment, when multiple odor detection values output from the odor detection means are input as input information, the estimation of the odor type such as burning odor, cigarette odor, or cooking odor as the expected output value is estimated. Weights and biases in a multi-layer neural network are learned by giving values and back-propagation processing to minimize the error between the output value and the expected value, and the odor of burning with higher accuracy from the input information. It can be used for highly accurate fire judgment by discriminating the smell of tobacco or the smell of cooking.

Explanatory drawing which showed embodiment of fire monitoring system Explanatory diagram showing the odor detection value of 16 channels by the odor detector of FIG. 1 on a radar chart. Explanatory drawing showing the functional configuration of fire judgment by the fire monitoring system Explanatory drawing which showed the functional structure of the odor determination part of FIG. Explanatory diagram showing the functional configuration of the multi-layer neural network shown in FIG. Explanatory drawing showing a list of fire judgment processing by the fire judgment part of FIG. Explanatory drawing showing the functional configuration of a fire monitoring system in which an odor judgment unit and a fire judgment unit are composed of a multi-layer neural network.

[Basic concept of the embodiment]
FIG. 1 is an explanatory diagram showing an embodiment of a fire monitoring system. The basic concept of the fire monitoring system according to the present embodiment is that in addition to the fire detector 14 which is a fire detecting means for issuing a fire when the smoke concentration or temperature in the monitoring space exceeds or exceeds a predetermined value, the monitoring space An odor detector 18 which is an odor detecting means for detecting an odor is provided, and an odor detection value of the odor detector 18 is transmitted to, for example, an odor determining unit 40 which is an odor determining means provided in a receiver 10, and a type of odor, for example, , The odor of burning things, the odor of cigarettes, and the odor of cooking are discriminated, and the fire determination unit 42, which is a fire determination means, determines the presence or absence of a fire alarm by the fire detector 14, and the odor determination unit 40 determines the type of odor. In addition to this, when a fire detector 14 detects a fire and issues a fire, it determines a peculiar odor associated with the fire, for example, a burning odor of an object. , It is possible to prevent erroneous fire judgment caused by non-fire factors such as cigarettes and cooking, and to judge a fire with high accuracy.

Further, the odor detector 18 outputs a plurality of types of odor detection values according to the adsorption of a plurality of odor substances, and the odor determination unit 40 is composed of a multi-layered neural network and is input information odor detector 18 Smell detection by inputting a combination of multiple types of odor detection values and predetermined odor types (burnt odor, cigarette odor, cooking odor), which are expected output values, as learning information and learning by deep learning. When a plurality of types of odor detection values of the vessel 18 are input as input information, the odor type can be estimated and output, and a specific odor is detected from a complex odor aggregate like a human odor. It can be distinguished and used for fire judgment.

[Overview of fire alarm system]
As shown in FIG. 1, this embodiment takes an R-type fire monitoring system as an example. For example, a receiver 10 is installed in a manager's room on the first floor of a building, and a system is provided from the receiver 10 to a caution area. Signal lines 12-1 to 12-3 are pulled out separately for each. Here, the signal lines 12-1 and 12-2 serve as a fire monitoring system, and the signal lines 12-3 serve as a terminal control system. The signal lines 12-1 to 12-3 may be referred to as a signal line 12 when it is not necessary to distinguish them.

A plurality of analog fire detectors 14 having a transmission function in which unique addresses are set are connected to the signal lines 12-1 and 12-2. Further, in addition to the analog fire detector 14, the odor detector 18 is connected to the signal lines 12-1 and 12-2 via a repeater 16 having a unique address and a transmission function. There is.

In the signal line 12-1, a fire detector 14 and an odor detector 18 are installed in each of the monitoring sections A11 and A12 to monitor fires and odors in the same monitoring area. On the other hand, in the signal line 12-2, the fire detector 14 is installed in all the monitoring sections A21 to AA23, but the odor detector 18 is a part of the monitoring section as shown in the monitoring section A22, for example. In addition to the installation section A22, the odor detectors 18 in the monitoring section A22 monitor the odors of the adjacent monitoring sections A21 and A23, reducing the number of odor detectors 18 installed.

The odor detector 18 may be installed according to the usage status of the monitoring section. For example, the odor detector 18 is installed together with the fire detector 14 in a section where a non-fire factor exists, such as a smoking room, a cooking room, and a storage section for valuables and important documents.

Further, on the signal lines 12-1 and 12-2, an on / off type fire detector can be connected to the sensor line drawn from the repeater as needed, and further, gas can be connected via the repeater. A leak alarm or transmitter can also be connected.

A terminal control device 15 such as a fire door or a fire damper is connected to the signal line 12-3 via a repeater 16 having a unique address and a transmission function.

Here, the maximum number of addresses set in the terminal devices such as the fire detector 14 and the repeater 16 connected to the signal lines 12-1 to 12-3 is, for example, 255, and the signal lines 12-1 to 12-3. A terminal device including a maximum of 255 fire detectors 14 can be connected to each of the above.

[Functional configuration of receiver]
(Overview of receiver)
The receiver 10 is provided with a main CPU 20 and a plurality of sub CPU boards 22-1 to 22-3, and the sub CPU boards 22-1 to 22-3 are provided with a sub CPU 24 and a transmission circuit unit 26.

The main CPU 20 and the sub CPU 24 are connected by a serial transfer bus 25, and transmit and receive data to and from each other. As the serial transfer bus 25, for example, a control area network (CAN) is used, and data is transmitted / received in packet units called frames (also referred to as messages).

The main CPU 20 includes a display 28 with a touch panel using a liquid crystal display panel or the like, a display unit 30 equipped with a fire representative light, a gas leak representative light, a failure representative light, etc., and an operation unit 32 provided with a predetermined operation switch. An acoustic alarm unit 34 provided with a speaker and a signal transfer unit 36 are connected.

(Fire monitoring control)
The sub CPU 24 provided on the sub CPU boards 22-1, 22-2 follows a predetermined communication protocol between the fire detector 14 and the repeater 16 to which the odor detector 18 is connected by instructing the transmission circuit unit 26. Fire monitoring and control are performed by transmitting and receiving signals.

The downlink signal from the transmission circuit unit 26 to the repeater 16 to which the fire detector 14 and the odor detector 18 are connected is transmitted in the voltage mode. The signal in this voltage mode is transmitted as a voltage pulse that changes the line voltage of the signal line 12 between, for example, 18 volt and 30 volt.

On the other hand, the uplink signal from the repeater 16 to which the fire detector 14 and the odor detector 18 are connected to the transmission circuit unit 26 is transmitted in the current mode. In this current mode, a signal current is passed through the signal line 12 at the timing of bit 1 of the transmission data, and an uplink signal is transmitted to the receiver 10 as a so-called current pulse train.

During normal monitoring, the fire monitoring control by the sub CPU 24 instructs the transmission circuit unit 26 at regular intervals to transmit a broadcast batch AD conversion signal including a batch AD conversion command. The fire detector 14 that has received the batch AD conversion signal detects and holds the smoke concentration or temperature and the odor detection value as sensor data, and the repeater 16 instructs the odor detector 18 to use the odor detection value as sensor data. Detect and retain. Subsequently, the sub CPU 24 transmits a call signal including a polling command in which terminal addresses are sequentially specified.

When the repeater 16 to which the fire detector 14 and the odor detector 18 are connected receives a call signal having an address matching its own address, it transmits a response signal including the sensor data held at that time to the transmission circuit unit 26. ..

For example, when the smoke concentration exceeds a predetermined caution level or exceeds the caution level, the fire detector 14 transmits a fire interrupt signal to the receiver 10.

When the sub CPU 24 receives the fire interruption signal from the fire detector 14 via the transmission circuit unit 26, the sub CPU 24 transmits a group search signal to identify the group including the fire detector 14 detecting the fire, and then the group. The address of the fire detector 14 that detects a fire by transmitting an internal search signal is specified, and the smoke concentration of the fire detector 14 that interrupted the fire and the odor detector 18 installed in the same monitoring section or surrounding section The odor detection values are intensively collected, and the detector address where the fire is detected, the smoke concentration, and the odor detection value are transmitted to the main CPU 20 via the serial transfer bus 25 as fire detection information.

The main CPU 20 is provided with a monitoring control unit 38, an odor determination unit 40, and a fire determination unit 42 as functions realized by executing a program. When the monitoring control unit 38 of the main CPU 20 receives the fire detection information from the sub CPU 24, it determines a fire in cooperation with the odor determination unit 40 and the fire determination unit 42, and when the fire determination is confirmed, the fire is detected in the display device 28. Fire alarm information including the location of the fire is displayed based on the detector address, a predetermined main acoustic alarm indicating the occurrence of a fire is output from the speaker of the acoustic alarm unit 34, and the fire transfer signal is further transmitted by the transfer unit 36. Output to the outside to perform predetermined interlocking control.

[Smell detector]
FIG. 2 is an explanatory diagram showing the odor detection values of 16 channels by the odor detector on a radar chart. The odor detector 18 connected to the signal lines 12-1 and 12-2 of FIG. 1 via a repeater 16 outputs a plurality of types of odor detection values x1 to xn according to the adsorption of a plurality of odorous substances.

The odor detector 18 is provided with, for example, a plurality of gas sensors having a sensitive film formed on a crystal substrate to select and adsorb different odor molecules. The sensitive film of the gas sensor uses a material such as ammonia, toluene, acetone, and ethanol that reacts with each odor molecule and selectively adsorbs the gas component. The gas sensor outputs a detection value xi having a level corresponding to, for example, the frequency of the crystal oscillator changing in proportion to the amount of adsorption of odor molecules. In the present embodiment, for example, a 16-channel odor detector 18 that outputs detected values x1 to x16 in response to 16 types of odor molecules is used.

Further, in the present embodiment, as the types of odors used for fire determination, for example, the burning odor of an object, the odor of tobacco, and the odor of cooking are discriminated, and the burning odor of the object to be discriminated is determined. When the odor of cigarettes or the odor of cooking comes into contact with the odor detector 18, for example, the unique detection values x1 to x16 shown in the radar chart of FIG. 2 are output.

FIG. 2A shows a radar chart of detected values x1 to x16 when the burning odor of an object comes into contact, and FIG. 2B shows a radar chart of detected values x1 to x16 when the odor of tobacco comes into contact. , FIG. 2C shows a radar chart of detected values x1 to x16 when the smell of cooking comes into contact. The radar chart of FIG. 2 is an example, and has various patterns depending on the type and arrangement of the 16-channel gas sensor.

As is clear from the radar chart of FIG. 2, the detection values x1 to x16 of the odor detector 18 in the multi-channel configuration (multiple gas sensor configuration) have different patterns of burning odor, tobacco odor, and cooking odor. This difference in visual pattern means that a specific odor is detected and distinguished from a complex collection of odors, like the human sense of smell.

Here, the burning odor of an object includes a wide variety of odors that differ depending on the type of combustibles caused by the fire, and the cooking odor is the odor of the food being cooked, depending on the type of cooking. A wide variety of odors are included, and as a result, the patterns are not limited to the burning odor and the cooking odor pattern shown in FIGS. 2 (A) and 2 (C), and a wide variety of patterns are obtained. On the other hand, the tobacco odor is a characteristic odor containing nicotine, and has a pattern similar to the tobacco odor pattern shown in FIG. 2 (B).

In order to determine the type of odor such as burning odor, cigarette odor, and cooking odor from the detection values x1 to x16 of the odor detector 18, in the present embodiment, the odor determination unit 40 has multiple layers. A neural network of equations is provided.

As the odor detector 18, in addition to the 16 channels of the present embodiment, a higher resolution odor detector such as 32 channels or 64 channels may be used.

[Functional configuration of fire judgment]
FIG. 3 is an explanatory diagram showing the functional configuration of fire judgment by the fire monitoring system, FIG. 4 is an explanatory diagram showing the functional configuration of the odor determination unit of FIG. 3, and FIG. 5 is the function of the multi-layer neural network shown in FIG. It is explanatory drawing which showed the structure.

As shown in FIG. 3, the fire judgment function by the fire monitoring system shown in FIG. 1 is composed of a fire detector 14, an odor detector 18, an odor judgment unit 40, and a fire judgment unit 42.

The fire detector 14 is, for example, a smoke detector or a heat detector, and reports a fire when the smoke concentration or temperature exceeds a predetermined fire threshold value or exceeds the fire threshold value, and sends a fire warning signal S to the fire determination unit 42. It can be regarded as outputting.

The odor detector 18 detects the odor in the same monitoring space as the fire detector 14, and outputs the odor detection values x1 to x16 to the odor determination unit 40, taking the 16-channel odor detector 18 as an example.

The odor determination unit 40 determines the type of odor by inputting the odor detection values x1 to x16 from the odor detector 18, and in the present embodiment, the burnt odor estimation value y1 and the tobacco odor estimation The value y2 and the estimated cooking odor value y3 are output as the odor type.

The odor determination unit 40 includes a multi-layer neural network 46. As shown in FIG. 4, the odor determination unit 40 provided with the multi-layer neural network 46 is composed of a multi-layer neural network 46, an input unit 56, a learning control unit 58, and a learning information holding unit 60. It is realized by executing a program by the CPU of a computer circuit corresponding to the processing of the neural network.

The odor detection values x1 to x16 from the odor detector 18 are input to the multi-layer neural network 46 via the input unit 56, and the burning odor estimation value y1, the tobacco odor estimation value y2, and the cooking odor estimation value y3 are input. Output.

The learning control unit 58 uses learning information of a plurality of sets of odor detection values x1 to x16 of the odor detector 18 detected by generating a burning odor, a cigarette odor, and a cooking odor by a fire experiment or the like. Learning with a teacher that the holding unit 60 temporarily stores and stores the combination of the odor detection values x1 to x16 stored in the learning information holding unit 60 and the expected output values of burning odor, cigarette odor, and cooking odor. Information is input to the multi-layer neural network 46 via the input unit 56, and the weight and bias of the multi-layer neural network 46 are learned by a learning method such as a back propagation method (error back propagation method).

When arbitrary odor detection values x1 to x16 are input from the odor detector 18 into the multi-layer neural network 46 that has been trained using this supervised learning information, the burnt odor estimation value y1 and the cigarette odor estimation value are input. y2 and the estimated cooking odor y3 are output.

Here, the burning odor estimation value y1, the cigarette odor estimation value y2, and the cooking odor estimation value y3 are ideally as follows.
In the case of the burning smell of things, (y1, y2, y3) = (1,0,0)
In the case of the smell of cigarettes (y1, y2, y3) = (0,1,0)
In the case of the smell of cooking, (y1, y2, y3) = (0,0,1)
When the odor detection values x1 to x16 of the odor detector 18 are actually input to the multi-layer neural network 46, the sum of the estimated values y1, y2, and y3 is 1, and each value is between 0 and 1.

(Fully coupled neural network)
As shown in FIG. 5, the multi-layer neural network 46 is a fully connected neural network, and is composed of an input layer 62, a fully connected 64, a repetition of an intermediate layer 66 and a fully connected 64, and an output layer 68.

Here, since the fully-coupled neural network classifies the input information odor detection values x1 to x16 into three classes of burning odor, tobacco odor, and cooking odor, the final output layer. In 68, the same three units as the target three classes are arranged, and the inputs to these units are set to output y1, y2, y3 with a total of 1 by using a softmax function. The output y1, y2, y3 of the unit will indicate the probability of belonging to that class.

(Learning by backpropagation)
The multi-layer neural network 46 composed of an input layer 62, a plurality of intermediate layers 66, and an output layer 68 shown in FIG. 5 is provided with a plurality of units in each layer and connected to a plurality of units in the other layers, and each unit is connected. Weights and bias values are set in, the vector product of multiple input values and weights is obtained, the bias values are added to obtain the sum, and this is passed through a predetermined activation function and output to the unit of the next layer. Forward propagation is performed in which the value propagates until the final layer is reached.

To change the weights and biases of such neural networks, a learning algorithm known as backpropagation is used. In backpropagation, supervised learning when a data set of input value x and expected output value (expected value) y is given to the network, and unsupervised learning when only input value x is given to the network. In this embodiment, supervised learning is performed.

When backpropagation is performed in supervised learning, for example, a function of mean squared error is used as an error for comparing the estimated value y * and the expected value y, which are the results of forward propagation that has passed through the network. To do.

In backpropagation, the magnitude of the error between the estimated value y * and the expected value y is used to propagate the value from the rear to the front of the network while correcting the weight and bias. The corrected amount for each weight and bias is treated as a contribution to the error, calculated by the steepest descent method, and the value of the error function is minimized by changing the weight and bias values.

The procedure for learning by backpropagation for a neural network is as follows.
(1) The input value x is input to the neural network, forward propagation is performed, and the estimated value y * is obtained.
(2) Calculate the error with an error function based on the estimated value y * and the expected value y.
(3) Backpropagation is performed on the network while updating the weight and bias.

This procedure is repeated using different combinations of input values x and expected values y to minimize the value of the error function until the neural network weight and bias errors are minimized as much as possible.

In the supervised learning control of the multi-layer neural network 46 shown in FIG. 4,
In the case of the burning smell of things, (y1, y2, y3) = (1,0,0)
In the case of the smell of cigarettes (y1, y2, y3) = (0,1,0)
In the case of the smell of cooking, (y1, y2, y3) = (0,0,1)
Is used as the output value (expected value) to perform the backpropagation described above.

(Mounting position of odor judgment unit)
The learning of the multi-layer neural network 46 shown in FIG. 3 is not performed by the receiver 10, but is learned by another computer device, and the trained multi-layer neural network 46 is mounted on the receiver 10 as an odor determination unit 40. become.

Further, the odor determination unit 40 provided with the learned multi-layer neural network 46 is mounted on the odor detector 18 installed in the caution area, and the odor type is determined by the multi-layer neural network 46 by edge processing. You can do it. As a result, the processing load on the receiver 10 can be significantly reduced.

[Fire judgment unit]
As shown in FIG. 3, the fire determination unit 42 includes comparators 48, 50, 52 and a determination unit 54. The burnt odor estimation value y1, the tobacco odor estimation value y2, and the cooking odor estimation value y3 output from the multi-layer neural network 46 of the odor determination unit 40 have a total of 1 and are values between 0 and 1, respectively. Therefore, the comparators 48, 50, and 52 are binarized so as to indicate either 0 or 1.

A reference value 1/3 is set in the comparators 48, 50, 52 by the reference voltage sources 48a, 50a, 52a, and the estimated values y1, y2, y3 are set to 3 bit estimated values (Y1, Y2, Y3) by 2. It will be binarized to indicate the presence or absence of burning odors, tobacco odors and cooking odors.

The determination unit 54 is based on the binarized values of the fire alarm signal S, the burnt odor estimation value Y1, the tobacco odor estimation value Y2, and the cooking odor estimation value Y3, in other words, 4-bit input information (S, Y1, Based on Y2 and Y3), fire D1, fire caution D2 or non-fire D3 is determined and output.

FIG. 6 is an explanatory diagram showing a list of fire determination processes by the fire determination unit of FIG. 3, classified into modes 1 to 16, and corresponds to bit 1 for each 4-bit input information (S, Y1, Y2, Y3). “Yes” is indicated by a 〇 mark, and “None” corresponding to bit 0 is indicated by a cross mark.

(Fire judgment)
The determination output of modes 1 and 4 in FIG. 6 is "fire", and a fire alarm is output from the receiver 10.

Mode 1
(Fire alarm = Yes) (Smell of burning things = Yes) (Smell of cigarettes = Yes) (Smell of cooking = Yes)
In this case as well, the judgment output is "fire". That is, in mode 1, there is a fire alarm, the burning odor of an object is determined, and the odor of cigarettes and cooking is also determined. Therefore, it can be definitely determined that the fire is also in this case.

Mode 4 (fire alarm = yes), (burning smell = yes),
(Tobacco smell = none), (cooking smell = none)
The judgment output is "fire". That is, in mode 4, since there is a fire alarm and the burning odor of an object is determined, it can be definitely determined to be a fire.

(Fire caution judgment)
The determination output of modes 2, 3, 9 to 12 in FIG. 6 is "fire caution", and a fire caution alarm (also referred to as a pre-alarm or fire sign alarm) is output from the receiver 10. Fire caution means that it is not a fire, but there is a high possibility of a fire and caution is required.

Mode 2 is (fire alarm = yes), (burning smell = yes), (cigarette smell = yes),
(Cooking smell = none)
For example, it is assumed that cigarette butts are smoldering in an ashtray and a large amount of smoke is emitted, and there is a high possibility that a fire will develop. Therefore, it is necessary to call attention.

Mode 3 is
(Fire alarm = Yes), (Smell of burning things = Yes), (Smell of cooking = Yes),
(Smell of cigarette = none)
For example, it is assumed that the pot is overheated during cooking and the cooked food is smoldering and a large amount of smoke is emitted, and there is a high possibility that a fire will occur. Therefore, it is necessary to call attention.

Modes 9 to 12 are cases in which there is no fire alarm, but there is a burning odor, and neither tobacco nor cooking odor is detected, or one or both of them are detected. Since it corresponds to the previous stages of a few, even in this case, the judgment output is set to "fire caution" to call attention. Since there is no fire alarm in modes 10 to 14, the determination output may be set to "non-fire".

The remaining determination outputs of modes 5 to 8, 13 to 16 in FIG. 6 are "non-fire". Of these, modes 5 to 8 have a fire alarm, but at least there is no burning odor, and even if there is one or both of the cigarette odor and the cooking odor, this is due to smoke from normal smoking or cooking. Since it can be determined to be smoke or heat, the determination output can be set to "non-fire", and a non-fire alarm in which a fire alarm is erroneously output from the receiver 10 can be reliably prevented.

The fire determination by the fire determination unit 42 shown in FIG. 6 by combining the fire alarm and the type of odor is an example, and the fire caution may not be determined and only the fire and the non-fire may be determined.

When a fire or fire caution is determined, the display 28 of the receiver 10 displays a burning odor, a cigarette odor, a cooking odor, etc. as the type of odor used for determining the fire or fire caution. The person in charge may be able to determine what is the cause of the fire or what the combusted material is.

[Control operation of fire monitoring system]
The control operation by the fire monitoring system of FIG. 1 will be described as follows. Now, if a fire breaks out in the monitoring section A11 from which the signal line 12-1 is pulled out, the receiver 10 receives the fire alarm of the fire detector 14 installed in the monitoring section A11, and also in the monitoring area A11. The odor detection values x1 to x16 of the installed odor detector 18 are received.

The odor determination unit 40 of the receiver 10 inputs the odor detection values x1 to x16 to determine the burning odor of the object, and the fire determination unit 42 is based on the fire alarm and the burning odor of the object determined by the odor determination unit 40. A fire is determined, and in response to this, the monitoring control unit 38 performs alarm control to output a fire alarm.

On the other hand, if a fire breaks out in the monitoring area A21 from which the signal line 12-2 is pulled out, the receiver 10 receives the fire alarm of the fire detector 14 installed in the monitoring area A21, and also in the monitoring area A21. The odor detection values x1 to x16 of the odor detector 18 installed in the adjacent monitoring section A22 are received. The odor detection value x1 to x16 by the odor determination unit 40 installed in the monitoring section A22 around the receiver 10 is input to determine the burning odor of an object, and the fire determination unit 42 is a fire detector installed in the monitoring area A21. A fire is determined based on the fire alarm of 14 and the burning odor of an object in the adjacent monitoring area A22 determined by the odor determination unit 40, and in response to this, the monitoring control unit 38 performs alarm control to output a fire alarm. ..

[Embodiment in which an odor determination unit and a fire determination unit are configured by a multi-layer neural network]
FIG. 7 is an explanatory diagram showing a functional configuration of a fire monitoring system in which an odor determination unit and a fire determination unit are configured by a multi-layer neural network.

As shown in FIG. 7, the present embodiment is a multi-layer type in which the fire alarm signal S from the fire detector 14 and the odor detection values x1 to x16 from the odor detector 18 are all provided in the odor / fire determination unit 70. It is input to the neural network 46, the fire estimated value y1 and the non-fire estimated value y2 are output, and the fire determination unit 72 determines whether it is a fire or a non-fire.

The functional configuration of the odor / fire determination unit 70 provided with the multi-layer neural network 46 is the same as that in FIG. 3, and the multi-layer neural network 46 uses the learning control unit 58 to input information such as a fire alarm signal S and an odor detection value. Supervised learning information that combines x1 to x16 and expected output values of fire and non-fire is input to the multi-layer neural network 46 via the input unit 56, for example, backpropagation method (error backpropagation method), etc. The weight and bias of the multi-layer neural network 46 are learned by the learning method.

As a result, when the fire alarm signal S from the fire detector 14 and the odor detection values x1 to x16 from the odor detector 18 are actually input to the multi-layer neuronetwork 46, a fire having a value in the range of 0 to 1 is actually input. The estimated value y1 and the non-fire estimated value y2 are output and binarized with the reference value 1/2 set by the reference voltage sources 74a and 76a of the comparators 74 and 76 provided in the fire judgment unit 72 to determine the fire. As the value Y1 and the non-fire determination value Y2, the fire determination result (Y1, Y2) = (1,0) or the non-fire determination result (Y1, Y2) = (0,1) is output.

Since the odor / fire determination unit 70 of the present embodiment cannot be provided in the odor detector 18 installed in the caution area shown in FIG. 1 to perform edge processing, it will be provided in the receiver 10. Further, the learning of the multi-layer neural network 46 is not performed by the receiver 10, but is learned by another computer device, and the trained multi-layer neural network 46 is mounted on the receiver 10.

[Modification of the present invention]
(Fire alarm system)
The above embodiment is an example of an R-type fire alarm system capable of acquiring a terminal address of a fire detector or the like to identify the location of a fire, but the present invention is not limited to this and is pulled out from a receiver. The same can be applied to a P-type fire alarm system that detects and warns a fire in each signal line by providing an odor detector in each line and determining the type of odor with a receiver.

(Other)
In addition, the present invention includes appropriate modifications that do not impair its purpose and advantages, and is not further limited by the numerical values shown in the above embodiments.

10: Receivers 12-1 to 12-3: Signal line 14: Fire detector 16: Repeater 18: Smell detector 20: Main CPU
22-1 to 22-3: Sub CPU board 24: Sub CPU
25: Serial transfer bus 26: Transmission circuit unit 28: Display device 30: Display unit 32 :: Operation unit 34: Acoustic alarm unit 36: Transfer unit 38: Monitoring control unit 40: Smell determination unit 42: Fire determination unit 46: Multi-layer neural network 48, 50, 52, 74, 76: Comparator 48a, 50a, 52a, 74a, 76a: Reference voltage source 54: Judgment unit 56: Input unit 58: Learning control unit 60: Learning information holding unit 62: Input layer 64: Fully coupled 66: Intermediate layer 68: Output layer 70: Smell / fire judgment unit 72: Fire judgment unit

Claims (11)

A fire detecting means for issuing a fire when the smoke concentration or temperature in the monitoring space exceeds a predetermined value or exceeds the predetermined value.
An odor detecting means for detecting the odor in the monitoring space and
An odor determining means for inputting an odor detecting value of the odor detecting means to determine the type of odor, and
A fire determining means for determining a fire based on the presence or absence of the fire alarm by the fire detecting means and the type of the odor output from the odor determining means.
A fire monitoring system characterized by the fact that
In the fire monitoring system according to claim 1,
The odor detecting means outputs a plurality of types of odor detecting values according to the adsorption of a plurality of odor substances.
The odor determination means is configured by a multi-layer neural network, and a combination of the plurality of types of odor detection values of the odor detection means, which is input information, and a predetermined odor type, which is an expected output value, is input as learning information. A fire monitoring system characterized by learning by deep learning and estimating and outputting the odor type when the plurality of types of odor detection values of the odor detecting means are input as input information.
In the fire monitoring system according to claim 2.
The multi-layer neural network of the odor determining means
Learn the burning odor, cigarette odor, and cooking odor as the odor types.
A fire monitoring system characterized in that, when the plurality of types of odor detection values of the odor detecting means are input, the burning odor of an object, the odor of cigarettes, and the odor of cooking are estimated and output.
In the fire monitoring system according to claim 3,
The fire determining means determines a fire based on the presence or absence of the fire alarm by the fire detecting means, the burning odor of the thing output from the odor determining means, the odor of the cigarette, and the presence or absence of the odor of cooking. A fire monitoring system characterized by
In the fire monitoring system according to claim 1,
The fire determination means
When it is determined that there is a fire alarm and the burning odor of the product is determined, or when the burning odor of the product, the cigarette odor and the cooking odor are determined, the fire is determined.
Other than the above, non-fire is judged.
A fire monitoring system that features that.
In the fire monitoring system according to claim 1,
The fire determination means further
A fire monitoring system comprising determining whether or not there is a fire alarm, and determining fire caution when at least one of the cigarette odor and the cooking odor is determined.
In the fire monitoring system according to claim 1,
The odor determination means and the fire determination means are composed of a multi-layer neural network.
Deep learning by inputting as learning information a combination of the presence or absence of the fire alarm of the fire detection means, the plurality of types of odor detection values of the odor detection means, and the expected output value of fire or non-fire, which are input information. When the presence or absence of the fire alarm of the fire detecting means and the plurality of types of odor detection values of the odor detecting means are input as input information, a fire or non-fire is estimated and output. Fire monitoring system.
In the fire monitoring system according to claim 1,
The fire detecting means and the odor detecting means are connected to a signal line drawn from the receiver.
The odor determining means is provided in the odor detecting means or the receiver.
The fire monitoring system is characterized in that the fire determination means is provided in the receiver.
In the fire monitoring system according to claim 8,
A fire alarm system, characterized in that the fire detecting means is installed in a predetermined detector installation section, and the odor detecting means is installed in the same detector setting section as the fire detecting means.
In the fire monitoring system according to claim 8,
The fire detection means is installed in each predetermined detector installation section, and is installed.
The odor detecting means is installed in a part of the detector installation section, and the odor detecting means is installed.
When the odor detecting means is not installed in the fire alarm section, the odor determining means of the receiver selects the odor detecting means installed in the surrounding section to determine the type of odor. A featured fire alarm system.
In the fire monitoring system according to claim 2 or 7.
The multi-layer neural network is a coupled neural network.
A fire monitoring system characterized in that learning is performed by backpropagation based on an error between a value output when the learning information is input and an expected value of a predetermined value.

Images