CN116311739A - Multi-sensor fire detection method based on long-short-term memory network and environment information fusion - Google Patents
Multi-sensor fire detection method based on long-short-term memory network and environment information fusion Download PDFInfo
- Publication number
- CN116311739A CN116311739A CN202211700578.9A CN202211700578A CN116311739A CN 116311739 A CN116311739 A CN 116311739A CN 202211700578 A CN202211700578 A CN 202211700578A CN 116311739 A CN116311739 A CN 116311739A
- Authority
- CN
- China
- Prior art keywords
- environmental
- time
- representing
- environment
- information
- 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.)
- Pending
Links
- 230000004927 fusion Effects 0.000 title claims abstract description 24
- 238000001514 detection method Methods 0.000 title claims abstract description 23
- 230000007613 environmental effect Effects 0.000 claims abstract description 168
- 230000008859 change Effects 0.000 claims abstract description 61
- 238000000605 extraction Methods 0.000 claims abstract description 33
- 238000012544 monitoring process Methods 0.000 claims abstract description 18
- 238000000034 method Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims abstract description 12
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 claims abstract description 8
- 239000000779 smoke Substances 0.000 claims abstract description 5
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 claims abstract description 4
- 229910002092 carbon dioxide Inorganic materials 0.000 claims abstract description 4
- 239000001569 carbon dioxide Substances 0.000 claims abstract description 4
- 229910002091 carbon monoxide Inorganic materials 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 45
- 230000006870 function Effects 0.000 claims description 39
- 230000004913 activation Effects 0.000 claims description 36
- 230000014509 gene expression Effects 0.000 claims description 24
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000005540 biological transmission Effects 0.000 claims description 5
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 2
- 238000012549 training Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 239000004973 liquid crystal related substance Substances 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000002485 combustion reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/06—Electric actuation of the alarm, e.g. using a thermally-operated switch
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/10—Actuation by presence of smoke or gases, e.g. automatic alarm devices for analysing flowing fluid materials by the use of optical means
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B19/00—Alarms responsive to two or more different undesired or abnormal conditions, e.g. burglary and fire, abnormal temperature and abnormal rate of flow
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
- G08B21/02—Alarms for ensuring the safety of persons
- G08B21/12—Alarms for ensuring the safety of persons responsive to undesired emission of substances, e.g. pollution alarms
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
- G08B21/02—Alarms for ensuring the safety of persons
- G08B21/12—Alarms for ensuring the safety of persons responsive to undesired emission of substances, e.g. pollution alarms
- G08B21/16—Combustible gas alarms
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Environmental & Geological Engineering (AREA)
- General Health & Medical Sciences (AREA)
- Toxicology (AREA)
- Analytical Chemistry (AREA)
- Combustion & Propulsion (AREA)
- Fire Alarms (AREA)
Abstract
The invention discloses a multi-sensor fire detection method based on long-short time memory network and environment information fusion. When detecting fire, the method firstly uses a multi-sensor system to continuously collect temperature, smoke concentration, carbon monoxide concentration and carbon dioxide concentration data in the environment, and constructs a monitoring value time sequence corresponding to environmental indexes; secondly, an environment change degree extraction module and an environment level extraction module in a deep environment information extraction model are constructed based on a long-short-time memory network, and the prediction of the environment index change degree and the environment index level is realized by means of the processing capacity of the long-short-time memory network on the long-distance dependency relationship of time sequence data; then, fusing the acquired multiple environmental information by using a multi-layer perceptron to obtain comprehensive fire state information; and finally, comparing the comprehensive fire state information value with an optimal threshold value learned in the system operation to complete fire judgment.
Description
Technical Field
The invention relates to the field of fire detection and the field of machine learning application, in particular to a multi-sensor fire detection method based on long-short-time memory network and environment information fusion.
Background
Fire is a complex combustion process that occurs with a variety of environmental indicators such as temperature, smoke concentration, harmful gas intensity, etc. The traditional single-sensor fire detection method is used for alarming aiming at a single environmental index by comparing the sensor reading at the current moment with a preset threshold value, but because the acquired environmental information is limited, the single-sensor detection method is easy to generate false alarm and missing alarm. The subsequent fire detection method based on the multi-sensor system can monitor various environmental indexes at the same time, and analyze, process, synthesize and fuse the acquired information based on a certain criterion, so as to realize more accurate fire judgment. The multi-sensor detection method overcomes the uncertainty and limitation of a single sensor, improves the effective performance of the whole system, and enhances the reliability of the system due to redundant data among the sensors. However, most of the current multi-sensor fire detection methods only consider the environmental index monitoring value at the current moment, and ignore the deep environmental information corresponding to the environmental index, such as the environmental index change information for quantifying the environmental information change degree of the current moment relative to the previous moment, and the environmental index level information for measuring the environmental level in the time dimension. Therefore, existing multi-sensor fire detection methods are susceptible to false alarms due to non-fire factors, such as short sensor failures, transient electromagnetic interference, etc. In addition, because the multi-sensor system needs to process more environmental information, the problems of complex overall structure and low detection efficiency exist. Therefore, how to comprehensively consider the environmental index monitoring value, the environmental index change degree and the environmental index level and how to efficiently realize the environmental information fusion is an important problem to be solved in the current multi-sensor fire detection.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a multi-sensor fire detection method based on a long-short time memory network and environment information fusion, which utilizes the excellent time sequence prediction capability of the long-short time memory network to extract deep environment information from different environment index time sequence data and combines the information fusion technology to obtain the comprehensive fire state at the current moment for fire judgment.
In order to achieve the above purpose, the invention adopts the following technical scheme:
step S1: the multi-sensor system continuously collects temperature, smoke concentration, carbon monoxide concentration and carbon dioxide concentration data in the environment, builds a time sequence of monitoring values corresponding to environmental indexes, and performs standardization treatment;
step S2: a deep environmental information extraction model is constructed based on a long-short-time memory network, and the deep environmental information extraction model comprises an environmental change degree extraction module and an environmental level extraction module;
step S3: processing the input time sequence data of the environmental index monitoring value by utilizing a long-short time memory network in the deep environmental information extraction model, and predicting the environmental index change degree of the current moment relative to the previous moment and the environmental index level at the current moment;
step S4: in order to improve the system operation efficiency, the multi-layer perceptron is pre-trained by utilizing another set of original fire data under similar conditions, and the optimal hidden layer node number of the multi-layer perceptron is determined;
step S5: using a multi-layer perceptron to fuse multiple environmental information at each moment in time sequence to obtain comprehensive fire state information corresponding to the moment;
step S6: and comparing the comprehensive fire state information value with an optimal threshold value learned in the operation of the system, and if the comprehensive fire state information value is smaller than the optimal threshold value, considering that no fire occurs, otherwise, considering that the fire occurs.
The environment change degree extraction module in the step S2 is a long-short-time memory network, and after the network is subjected to forgetting gate and updating gate processing, the cell state used for transmission is first-class fire state information; the hidden state calculated and output by the output gate is the environmental index change degree information. The core idea of the environment change degree extraction module is to keep the first type fire state transmitted in the neural network chain, forget, update and output the first type fire state information in the transmission process through three unique gate structures in the long-short-term memory network, and further control the output of the environment index change degree information.
The first type of fire state information is a quantitative description of the monitored fire environment based on environmental changes, and the quantitative formula is as follows:
in the above-mentioned formula (1),fire status of the first type, indicating the situation of the environment index n at time k>And->Respectively representing a change forgetting vector and a change update vector of the environmental index n at the moment k, ++>Fire status information of the first type, representing the environmental indicator n at time k-1,/for>First indicating environmental index n at time kA fire-like state candidate vector; wherein, the liquid crystal display device comprises a liquid crystal display device,and->The expression is as follows:
in the above formula (2), σ represents a Sigmoid activation function, W vf Indicating that the environment changes forget the weights,indicating the degree of variation of the environmental indicator n at time k-1,/->Representing the difference of the monitored data of the environmental index n at the time points k and k-1, b vf Indicating a changing environmental forgetting bias;
in the above formula (3), W vu Representing environmental change update weights, b vu Representing the update bias of the environmental change, and the rest of the same symbol expressions are as in formula (2);
in the above formula (4), tanh represents tanh activation function, W v And b v Respectively representing the weight and bias for creating a first type of fire state candidate value vector, and the rest of the same symbol expressions are the same as the formula (2);
the environmental index change degree information is used for predicting the change degree of the environment at the current moment compared with the environment at the previous moment on the time sequence, and the calculation formula is as follows:
in the above-mentioned formula (5),indicating the degree of change of the environmental indicator n at time k, < >>Representing the change output vector of the environmental indicator n at time k, tanh representing tanh activation function,/>A first fire state of the environmental index n at the time k is represented; wherein (1)>The expression of (2) is as follows:
in the above formula (6), σ represents a Sigmoid activation function, wv vo Representing the output weight of the environmental change,indicating the degree of variation of the environmental indicator n at time k-1,/->Representing the difference of the monitored data of the environmental index n at the time points k and k-1, b vo Representing an ambient change output bias.
The environmental index level extraction module in the step S2 is used as another long-short-time memory network, and after the processing of forgetting gate and updating gate in the network, the cell state used for transmitting is the fire state information of the second type; the hidden state calculated and output by the output gate is the environmental index level information. The core idea of the environment index level extraction module is to keep the second type fire state transmitted in the neural network chain, forget, update and output the second type fire state information through the gate structure, and control the output and transmission of the environment index level information.
The second type of fire state information is a quantitative description of the monitored fire environment based on the environment level, and the quantitative formula is as follows:
in the above-mentioned formula (7),a fire status of the second type indicating the situation of the environment index n at time k +.>And->Horizontal forgetting vector and horizontal update vector respectively representing environment index n at k time, and +.>A second fire state, which indicates the situation n at time k-1, ">A second type of fire state candidate vector representing an environmental indicator n at time k; wherein (1)>And->The expression is as follows:
in the above formula (8), σ represents a Sigmoid activation function, W lf Representing the ambient level forgetting weight,level information indicating the environmental indicator n at time k-1,/for>Representing the monitored value of the environmental index n at the time k, b lf Representing an ambient level forget bias;
in the above formula (9), W lu Representing environmental level update weights, b lu Representing the ambient level update bias, the remaining same symbol expressions being as in equation (8);
in the above formula (10), tanh represents tanh activation function, W l And b l Respectively representing the weight and the bias for creating a second type of fire state candidate value vector, and expressing the rest of the same symbols as the formula (8);
the environmental index level information is a measurement result of the environmental level at the current moment in the time dimension, and the calculation formula is as follows:
in the above-mentioned formula (11),level information indicating the environmental indicator n at time k, < >>A horizontal output vector representing the environmental indicator n at time k, tanh representing tanh activation function,/>A second type fire state of the environmental index n at the time k; wherein (1)>The expression of (2) is as follows:
in the above formula (12), σ represents a Sigmoid activation function, W lo Representing the ambient level output weight,level information indicating the environmental indicator n at time k-1,/for>Representing the monitored value of the environmental index n at the time k, b lo Representing the ambient level output bias.
The information fusion in the step S5 is to analyze and integrate the multi-element fire monitoring values obtained according to time sequence by utilizing a computer and neural network technology under a certain criterion. In a multi-sensor fire detection system, environmental index information provided by each information source has a certain degree of uncertainty, and the fusion of the uncertainty information belongs to an uncertainty reasoning process. The neural network can meet the processing requirements of the multi-sensor data fusion technology by virtue of self-learning, self-adaption and nonlinear processing capabilities, and the classification standard is determined according to the sample similarity accepted by the current system, and the determination method is mainly expressed on the weight distribution of the network. The multi-layer perceptron fuses the functional expression for processing various environmental information as follows:
in the above-mentioned formula (13),output vector representing input layer of multi-layer perceptron at k moment, sigma represents Sigmoid activation function, W in And b in Respectively representing the weight and bias of the input layer, +.>Representing the monitoring of the environment index n at time kThe measured value, the environmental index change degree information and the environmental index level information form a transpose of a vector group;
the output vector of the input layer is activated by adopting a Sigmoid function, and the activation expression is as follows:
in the above-mentioned formula (14),output vector representing hidden layer of multi-layer perceptron at k moment, sigma represents Sigmoid activation function Whi id And b hid Respectively representing the weight and bias of the hidden layer;
further, activating the hidden layer output vector to obtain the comprehensive fire state, wherein the calculation formula is as follows:
in the above formula (15), F k Represents the comprehensive fire state at time k, sigma represents the Sigmoid activation function, W out And b out Respectively representing the weight and bias of the output layer.
Drawings
FIG. 1 is a diagram of a system architecture according to the present invention
FIG. 2 is a block diagram of an environment index change degree extraction module according to the present invention
FIG. 3 is a block diagram of an environmental indicator level extraction module according to the present invention
FIG. 4 is a block diagram of an environment information fusion module according to the present invention
Detailed Description
The present invention will be further described with reference to the accompanying drawings and examples, which are not intended to limit the scope of the invention.
Examples:
FIG. 1 is a system architecture diagram of the present invention, which is composed of an environmental index monitor processor, a deep environmental information extraction model and an information fusion module. The multi-sensor system constructs corresponding time series data by continuously collecting the environmental index monitoring values. The environment index change degree extraction module and the environment level extraction module in the deep environment information extraction model respectively process input environment index time sequence data by utilizing a long-short-time memory network, and output change degree information and level information corresponding to different environment indexes at the current moment by combining a cell state and a hidden state at the last moment transmitted in the network. And then, the information fusion module analyzes and processes various environmental information at the current moment to obtain comprehensive fire state information for fire judgment. Specifically, the environment information corresponding to each type of environment index comprises an environment index monitoring value, environment index change degree information and environment index level information.
Specifically, to illustrate an embodiment of the present invention, in the example, the raw time-series data portion of the collected environmental indicator monitor value construction is shown in table 1:
TABLE 1 original time series data (part of) for different environmental indicator monitor values
In this example, the monitored environmental indicators include temperature, smoke concentration, carbon monoxide concentration, and carbon dioxide concentration, a relative ignition time of 10 indicating a moment when a fire has occurred, less than 10 indicating a moment when no fire has occurred, and greater than or equal to 10 indicating a moment when a fire has occurred. The original time sequence data of the environmental indexes are standardized, and the standardized formula is as follows:
in the above-mentioned formula (1),representing the monitored value of the environmental index n after standardized treatment at time k,/for the environmental index n>Representing the original monitored value of the environmental indicator n at time k,/-, for example>Representing the maximum raw monitored value of the environmental indicator n over the time series T.
The normalized environmental indicator monitoring values are shown in table 2:
table 2 time series data (part) of different environmental index monitoring values after normalization treatment
Further, the standardized environmental index time series data is input into a deep environmental information extraction model, and the deep environmental information extraction model comprises an environmental index change degree extraction module and an environmental index level extraction module.
FIG. 2 is a diagram showing the structure of the environmental index change degree extraction module of the present invention, which is used as a long-short-term memory network, wherein after the processing of forgetting gate and updating gate, the cell state used for transmitting is the first type of fire state information; the hidden state calculated and output by the output gate is the environmental index change degree information. The network is input into the environmental index monitoring data difference value between the current moment and the previous moment and the environmental index change degree information of the previous moment, so as to calculate the environmental index change degree information of the current moment; and the first fire state information and the environmental index change degree information obtained in the extraction process are transmitted to the next cell unit. The quantification formula of the first fire state is as follows:
in the above-mentioned formula (2),fire status of the first type, indicating the situation of the environment index n at time k>And->Respectively representing a change forgetting vector and a change update vector of the environmental index n at the moment k, ++>A fire state of the first type, which indicates the situation of the environment index n at time k-1, ">A first type of fire state candidate value vector representing an environmental indicator n at time k; wherein (1)>Andthe expression is as follows:
in the above formula (3), σ represents a Sigmoid activation function, W vf Indicating that the environment changes forget the weights,indicating the degree of variation of the environmental indicator n at time k-1,/->Representing the difference of the monitored data of the environmental index n at the time points k and k-1, b vf Indicating an environmental change forgetting bias;
in the above formula (4), W vu Representing environmental change update weights, b vu Representing environmental change update biasThe rest of the same symbols are expressed as in formula (3);
in the above formula (5), tanh represents tanh activation function, W v And b v Respectively representing the weight and bias for creating a first type of fire state candidate value vector, and the rest of the same symbol expressions are the same as the formula (3);
the environmental index change degree information is used for predicting the change degree of the environment at the current moment compared with the environment at the previous moment on the time sequence, and the calculation formula is as follows:
in the above-mentioned formula (6),indicating the degree of change of the environmental indicator n at time k, < >>Representing the change output vector of the environmental indicator n at time k, tanh representing tanh activation function,/>A first fire state of the environmental index n at the time k is represented; wherein (1)>The expression of (2) is as follows:
in the above formula (7), σ represents a Sigmoid activation function, W vo Representing the output weight of the environmental change,indicating the degree of variation of the environmental indicator n at time k-1,/->Representing the difference of the monitored data of the environmental index n at the time points k and k-1, b vo Representing an ambient change output bias.
FIG. 3 is a schematic diagram of an environment index level extraction module structure of the present invention, as an alternative structure, a long-short-term memory network in which the cell status for transmission is the second type of fire status information after being subjected to forgetting gate and update gate processing; the hidden state calculated and output by the output gate is the environmental index level information. The network input is the standardized environmental index monitoring value at the current moment and the environmental index level information at the previous moment, so as to calculate the environmental index level information at the current moment; and the second type of fire state information and the environmental index level information obtained in the extraction process are transmitted to the next cell unit. Wherein, the quantization formula of the second fire state is as follows:
in the above-mentioned formula (8),a fire status of the second type indicating the situation of the environment index n at time k +.>And->Horizontal forgetting vector and horizontal update vector respectively representing environment index n at k time, and +.>A second fire state, which indicates the situation n at time k-1, ">A second type fire state information candidate value vector representing an environmental index n at the time k; wherein, the liquid crystal display device comprises a liquid crystal display device,and->The expression is as follows:
in the above formula (9), σ represents a Sigmoid activation function, W lf Representing the ambient level forgetting weight,level information indicating the environmental indicator n at time k-1,/for>Representing the monitored value of the environmental index n at the time k, b lf Representing an ambient level forget bias;
in the above formula (10), W lu Representing environmental level update weights, b lu Representing the ambient level update bias, the remaining same symbol expressions being as in equation (9);
in the above formula (11), tanh represents tanh activation function, W l And b l Respectively representing the weight and the bias for creating a second type of fire state candidate value vector, and expressing the rest of the same symbols as the formula (9);
the environmental indicator level information is a measurement result in the time dimension from the environmental level at the current time, and the calculation formula is as follows:
in the above-mentioned formula (12),level information indicating the environmental indicator n at time k, < >>A horizontal output vector representing the environmental indicator n at time k, tanh representing tanh activation function,/>A second type fire state of the environmental index n at the time k; wherein (1)>The expression of (2) is as follows:
in the above formula (13), σ represents a Sigmoid activation function, W lo Representing the ambient level output weight,level information indicating the environmental indicator n at time k-1,/for>Representing the monitored value of the environmental index n at the time k, b lo Representing the ambient level output bias.
FIG. 4 is a block diagram of an environmental information fusion module of the present invention, which is used for analyzing and fusing a plurality of fire monitoring values obtained according to time sequence under a certain criterion, and the function expression is as follows:
in the above-mentioned formula (14),output vector representing input layer of multi-layer perceptron at k moment, sigma represents Sigmoid activation function, W in And b in Respectively representing the weight and bias of the input layer,/>Indicating the transposition of a vector group consisting of the monitoring value of the environmental index n at the k moment, the environmental index change degree information and the environmental index level information;
the output vector of the input layer is activated by adopting a Sigmoid function, and the expression of the activation function is as follows:
in the above-mentioned formula (15),output vector representing hidden layer of multi-layer perceptron at k moment, sigma represents Sigmoid activation function, W hid And b hid Respectively representing the weight and bias of the hidden layer;
further, activating the hidden layer output vector to obtain the comprehensive fire state, wherein the calculation formula is as follows:
in the above formula (16), F k Represents the comprehensive fire state at time k, sigma represents the Sigmoid activation function, W out And b out Respectively representing the weight and bias of the output layer.
Example the test results are shown in table 3, the test results are good and the results are accurate.
TABLE 3 actual test results for the system
The foregoing description is only of the preferred embodiments of the invention, and all changes and modifications that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Claims (8)
1. A multi-sensor fire detection method based on long-short-time memory network and environment information fusion is characterized by comprising the following steps:
step S1: the multi-sensor system continuously collects temperature, smoke concentration, carbon monoxide concentration and carbon dioxide concentration data in the environment, builds a time sequence of monitoring values corresponding to environmental indexes, and performs standardization treatment;
step S2: a deep environmental information extraction model is constructed based on a long-short-time memory network, and the deep environmental information extraction model comprises an environmental change degree extraction module and an environmental level extraction module;
step S3: processing the input time sequence data of the environmental index monitoring value by utilizing a long-short time memory network in the deep environmental information extraction model, and predicting the environmental index change degree of the current moment relative to the previous moment and the environmental index level at the current moment;
step S4: pre-training the multi-layer perceptron by using another set of original fire data under similar conditions, and determining the optimal hidden layer node number of the multi-layer perceptron;
step S5: using a multi-layer perceptron to fuse multiple environmental information at each moment in time sequence to obtain comprehensive fire state information corresponding to the moment;
step S6: and comparing the comprehensive fire state information value with an optimal threshold value learned in the operation of the system, and if the comprehensive fire state information value is smaller than the optimal threshold value, considering that no fire occurs, otherwise, considering that the fire occurs.
2. The multi-sensor fire detection method based on long-short-time memory network and environment information fusion according to claim 1, wherein the method is characterized in that: the environment change degree extraction module in the step S2 is a long-short-time memory network, and after the network is subjected to forgetting gate and updating gate processing, the cell state used for transmission is first-class fire state information; the hidden state calculated and output by the output gate is the environmental index change degree information.
3. The multi-sensor fire detection method based on long-short-time memory network and environment information fusion according to claim 2, wherein the method is characterized in that: the first type of fire state information is a quantitative description of the monitored fire environment based on environmental changes, and the quantitative formula is as follows:
in the above-mentioned formula (1),fire status of the first type, indicating the situation of the environment index n at time k>And->Respectively representing a change forgetting vector and a change update vector of the environmental index n at the moment k, ++>A fire state of the first type, which indicates the situation of the environment index n at time k-1, ">A first type of fire state candidate value vector representing an environmental indicator n at time k; wherein (1)>And->The expression is as follows:
in the above formula (2), σ represents a Sigmoid activation function, W vf Indicating that the environment changes forget the weights,indicating the degree of variation of the environmental indicator n at time k-1,/->Representing the difference of the monitored data of the environmental index n at the time points k and k-1, b vf Indicating an environmental change forgetting bias;
in the above formula (3), W vu Representing environmental change update weights, b vu Representing the update bias of the environmental change, and the rest of the same symbol expressions are as in formula (2);
in the above formula (4), tanh represents tanh activation function, W v And b v The weight and bias for creating the first type fire state candidate value vector are respectively expressed, and the rest of the same symbols are expressed as in the formula (2).
4. The multi-sensor fire detection method based on long-short-time memory network and environment information fusion according to claim 2, wherein the method is characterized in that: the environmental index change degree information is used for predicting the change degree of the environment at the current moment compared with the environment at the previous moment on the time sequence, and the calculation formula is as follows:
in the above-mentioned formula (5),indicating the degree of change of the environmental indicator n at time k, < >>Representing the change output vector of the environmental indicator n at time k, tanh representing tanh activation function,/>A first fire state of the environmental index n at the time k is represented; wherein (1)>The expression of (2) is as follows:
in the above formula (6), σ represents a Sigmoid activation function, W vo Representing the output weight of the environmental change,indicating the degree of variation of the environmental indicator n at time k-1,/->Representing the difference of the monitored data of the environmental index n at the time points k and k-1, b vo Representing an ambient change output bias.
5. The multi-sensor fire detection method based on long-short-time memory network and environment information fusion according to claim 1, wherein the method is characterized in that: the environment level extraction module in the step S2 is a long-short-time memory network, and after the processing of forgetting gate and updating gate in the network, the cell state used for transmitting is the fire state information of the second type; the hidden state calculated and output by the output gate is the environmental index level information.
6. The multi-sensor fire detection method based on long-short-term memory network and environment information fusion according to claim 5, wherein the method is characterized in that: the second type of fire state information is a quantitative description of the monitored fire environment based on the environment level, and the quantitative formula is as follows:
in the above-mentioned formula (7),a fire status of the second type indicating the situation of the environment index n at time k +.>And->Horizontal forgetting vector and horizontal update vector respectively representing environment index n at k time, and +.>A second fire state, which indicates the situation n at time k-1, ">A second type of fire state candidate vector representing an environmental indicator n at time k; wherein (1)>And->The expression is as follows:
in the above formula (8), σ represents a Sigmoid activation function, W lf Representing the ambient level forgetting weight,level information indicating the environmental indicator n at time k-1,/for>Representing the monitored value of the environmental index n at the time k, b lf Representing an ambient level forget bias;
in the above formula (9), W lu Representing environmental level update weights, b lu Representing the ambient level update bias, the remaining same symbol expressions being as in equation (8);
in the above formula (10), tanh represents tanh activation function, W l And b l The weights and offsets for creating the second type fire state candidate vectors are respectively expressed, and the rest of the same symbols are expressed as in formula (8).
7. The multi-sensor fire detection method based on long-short-term memory network and environment information fusion according to claim 5, wherein the method is characterized in that: the environmental index level information is a measurement result of the environmental level at the current moment in the time dimension, and the calculation formula is as follows:
in the above-mentioned formula (11),level information indicating the environmental indicator n at time k, < >>A horizontal output vector representing the environmental indicator n at time k, tanh representing tanh activation function,/>A second type fire state of the environmental index n at the time k; wherein (1)>The expression of (2) is as follows:
in the above formula (12), σ represents a Sigmoid activation function, W lo Representing the ambient level output weight,level information indicating the environmental indicator n at time k-1,/for>Representing the monitored value of the environmental index n at the time k, b lo Representing the ambient level output bias.
8. The multi-sensor fire detection method based on long-short-time memory network and environment information fusion according to claim 1, wherein the method is characterized in that: in the step S5, a plurality of environmental information at each time on the time sequence is processed by using a multi-layer perceptron fusion, wherein the environmental information corresponding to each type of environmental index includes a monitored value at the time, environmental index change degree information at the time relative to the previous time and environmental index level information at the time, and the fusion function is as follows:
in the above-mentioned formula (13),output vector representing input layer of multi-layer perceptron at k moment, sigma represents Sigmoid activation function, W in And b in Respectively are provided withRepresenting the weight and bias of the input layer, +.>Indicating the transposition of a vector group consisting of the monitoring value of the environmental index n at the k moment, the environmental index change degree information and the environmental index level information; the output vector of the input layer is activated by adopting a Sigmoid function, and the activation expression is as follows:
in the above-mentioned formula (14),output vector representing hidden layer of multi-layer perceptron at k moment, sigma represents Sigmoid activation function, W hid And b hid Respectively representing the weight and bias of the hidden layer;
activating the hidden layer output vector to obtain the comprehensive fire state, wherein the calculation formula is as follows:
in the above formula (15), F k Represents the comprehensive fire state at time k, sigma represents the Sigmoid activation function, W out And b out Respectively representing the weight and bias of the output layer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211700578.9A CN116311739A (en) | 2022-12-27 | 2022-12-27 | Multi-sensor fire detection method based on long-short-term memory network and environment information fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211700578.9A CN116311739A (en) | 2022-12-27 | 2022-12-27 | Multi-sensor fire detection method based on long-short-term memory network and environment information fusion |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116311739A true CN116311739A (en) | 2023-06-23 |
Family
ID=86836650
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211700578.9A Pending CN116311739A (en) | 2022-12-27 | 2022-12-27 | Multi-sensor fire detection method based on long-short-term memory network and environment information fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116311739A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117558116A (en) * | 2024-01-11 | 2024-02-13 | 山东奥深智能工程有限公司 | Fire control early warning system based on big data |
-
2022
- 2022-12-27 CN CN202211700578.9A patent/CN116311739A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117558116A (en) * | 2024-01-11 | 2024-02-13 | 山东奥深智能工程有限公司 | Fire control early warning system based on big data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11314242B2 (en) | Methods and systems for fault detection and identification | |
CN114757309B (en) | Multi-physical-field monitoring data collaborative fusion engineering disaster early warning method and system | |
CN116625438B (en) | Gas pipe network safety on-line monitoring system and method thereof | |
Johansson et al. | Detection of vessel anomalies-a Bayesian network approach | |
CN102013148B (en) | Multi-information fusion fire hazard detection method | |
CN115169479A (en) | Remote monitoring method, system and storage medium for sewage treatment process | |
CN113762329A (en) | Method and system for constructing state prediction model of large rolling mill | |
CN115688035A (en) | Time sequence power data anomaly detection method based on self-supervision learning | |
KR102427205B1 (en) | Apparatus and method for generating training data of artificial intelligence model | |
Son et al. | Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge | |
CN113516837B (en) | Urban fire judgment method and system based on multi-source information fusion and storage medium thereof | |
Vinothini et al. | IoT based flood detection and notification system using decision tree algorithm | |
Jiang et al. | A multisensor cycle-supervised convolutional neural network for anomaly detection on magnetic flux leakage signals | |
CN116311739A (en) | Multi-sensor fire detection method based on long-short-term memory network and environment information fusion | |
Baptista et al. | Classification prognostics approaches in aviation | |
CN111191855B (en) | Water quality abnormal event identification and early warning method based on pipe network multi-element water quality time sequence data | |
Ye et al. | A deep learning-based method for automatic abnormal data detection: Case study for bridge structural health monitoring | |
Liu et al. | Deep learning approach for sensor data prediction and sensor fault diagnosis in wind turbine blade | |
Dang et al. | seq2graph: Discovering dynamic non-linear dependencies from multivariate time series | |
Gámiz et al. | Dynamic reliability and sensitivity analysis based on HMM models with Markovian signal process | |
Zhang et al. | A flexible monitoring framework via dynamic-multilayer graph convolution network | |
Zhang et al. | An unsupervised leak detection method with aggregating prediction and reconstruction along projection pathway for natural gas gathering pipelines | |
CN112418398A (en) | Safety monitoring method for power information equipment | |
CN117970533B (en) | Lightning primary forecasting method based on atmospheric electric field | |
Kunnathettu et al. | Comparative Analysis of Neural Network and Machine Learning Techniques for Air Quality Prediction |
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 |