CN113436404A - High-sensitivity composite smoke-sensitive low-false-alarm method based on intelligent algorithm - Google Patents
High-sensitivity composite smoke-sensitive low-false-alarm method based on intelligent algorithm Download PDFInfo
- Publication number
- CN113436404A CN113436404A CN202110700540.0A CN202110700540A CN113436404A CN 113436404 A CN113436404 A CN 113436404A CN 202110700540 A CN202110700540 A CN 202110700540A CN 113436404 A CN113436404 A CN 113436404A
- Authority
- CN
- China
- Prior art keywords
- particle
- smoke
- data
- value
- algorithm
- 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
Images
Classifications
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/38—Services specially adapted for particular environments, situations or purposes for collecting sensor information
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a high-sensitivity composite smoke sensing low false alarm method based on an intelligent algorithm, which combines a composite smoke sensing component hardware platform and the intelligent algorithm, wherein the hardware platform adopts an LPC4078 single chip microcomputer system with an REM framework to match with a temperature sensor, an ion type smoke sensor and other various sensors to finish the acquisition of smoke original data, a smoke sensing device is connected with a cloud platform by a wireless network, the acquired original data is transmitted to the cloud platform, the cloud platform utilizes an improved particle swarm algorithm to analyze and solve the data, and when harmful smoke in gas is higher than a standard threshold value, the smoke sensing gives an alarm.
Description
Technical Field
The invention relates to the field of hardware and algorithms of fire alarm and fire sensors, in particular to a high-sensitivity composite smoke-sensing low-false-alarm method based on an intelligent algorithm.
Background
The smoke-sensitive alarm is actually a different name of a smoke-sensitive or smoke-sensitive alarm, the smoke-sensitive alarm realizes fire prevention by monitoring smoke concentration, a separate-from-type smoke sensor is adopted inside, the separate-from-type smoke sensor is a sensor with advanced technology and reliable working stability, and is widely applied to various fire-fighting alarm systems, and the performance of the smoke-sensitive alarm is far superior to that of a gas-sensitive resistance-type fire alarm.
The working principle of the smoke alarm is that the infrared light beam of the infrared transmitting tube is scattered by smoke particles, and the intensity of scattered light is in direct proportion to the smoke concentration, so that the intensity of the infrared light beam received by the photosensitive tube can be changed and converted into an electric signal, and finally, the electric signal is converted into an alarm signal. The smoke induction of the alarm is mainly completed by an optical maze, a group of infrared transmitting and receiving photoelectric tubes are arranged in the maze, and the correlation angle is 135 degrees. When no smoke exists in the environment, the receiving tube cannot receive infrared light emitted by the infrared emitting tube, and the subsequent sampling circuit has no electric signal change; when smoke exists in the environment, smoke particles enter the labyrinth to enable infrared light emitted by the emitting tube to be scattered, the intensity of the scattered infrared light has a certain linear relation with the smoke concentration, the subsequent sampling circuit changes, the main control chip arranged in the alarm judges the variable quantities to determine whether fire occurs or not, once the fire is determined, the alarm sends a fire signal, the fire indicator lamp (red) is turned on, and the buzzer is started to give an alarm.
However, in real life, the smoke sensor is rarely replaced from the beginning of installation, the smoke sensor is often installed in a fixed place for a long time, sometimes, dust is accumulated in the smoke sensor after the smoke sensor is installed for a long time, when airflow passes through the smoke sensor, the dust is easily blown, the smoke sensor can be mistaken for smoke, and due to the fact that the sensor in the smoke sensor is sensitive to tiny smoke particles for safety, the situation of mistaken touch alarm occurs; furthermore, gas such as water vapor may also cause smoke alarms.
In order to solve the problems, the composite smoke-sensing and temperature-sensing fire detector adopts a sensor with high complexity as a smoke sensor for detection, which reduces the false alarm rate of the smoke sensor to a certain extent, but still needs manual judgment.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a high-sensitivity composite smoke-feeling low-false-alarm method based on an intelligent algorithm.
The invention adopts the technical scheme that a high-sensitivity composite smoke sense low-false alarm method based on an intelligent algorithm is divided into a hardware platform and the intelligent algorithm, and the method comprises the following steps:
step 1: the hardware platform adopts a single chip microcomputer system to cooperate with various sensors to finish the acquisition of the smoke original data;
step 2: connecting the smoke sensing device with a cloud platform by using a wireless network, and transmitting the acquired original data to the cloud platform;
and step 3: and (3) performing data analysis and calculation by using an improved particle swarm algorithm in the cloud platform, when harmful smoke quantity in the gas is higher than a standard threshold value, giving an alarm when smoke feeling is generated, and repeating the step (1) when the harmful smoke quantity is lower than the standard threshold value.
Preferably, the system adopts an LPC4078 single chip microcomputer of an REM framework, and the sensor comprises a temperature sensor and an ionic smoke sensor.
Preferably, a NET Framework + C # heterogeneous system is adopted in the super system of the cloud platform, and NET Framework is applied to rich display components, so that the reliability and stability of the platform are improved. The platform is divided into a service supporting layer, an application system layer, a basic platform layer and a data resource layer. And (4) building a cloud infrastructure of the resource system and the control system, and performing related virtualization and data center network computing. And the system scheduling, safety and flexible management and allocation of resources are realized through a standardized interface.
NET Framework component realizes panoramic visualization display of communication scheduling picture, Visual Studio2010 is utilized to realize development of visualization system, and the layer provides uniform calling interface agent service;
the service support layer mainly provides interface service for service logic access between the application layer and the application system layer. The application system layer realizes development of middleware such as data acquisition, interface management, timing scheduling and the like based on the C # technology;
the application system layer realizes information transmission between the application system layer and the data resource layer and between the application system layer and the service supporting layer through an interface by a NetBEUI protocol;
the basic platform layer provides resources such as calculation, storage and the like required by the user, and realizes the resource allocation and rapid deployment as required through the resource pooling of technologies such as virtualization and the like;
the data resource layer is a visual database, stores information of different application systems, comprises equipment detection test information and equipment operation and fault information, and provides data support for communication scheduling of the platform.
Preferably, the step 3 of improving the particle swarm algorithm to perform data analysis and calculation includes the following steps:
step 3.1: generating a feature vector matrix for the smoke data; generating n particles, and initializing the initial positions of the n particles in the population, wherein the dimension of each particle is the same as that of the search space.
Step 3.2: calculating the average value of each column of features, and then subtracting the average value of the features of the column from each dimension; and (4) carrying out constraint condition processing, calculating the fitness of each particle, forming an individual optimal value solution set and a global optimal value solution set according to the non-domination relation, and storing data.
Step 3.3: calculating a covariance matrix of the features; and determining an individual optimal value and a global optimal value of each particle through the dominance relation and the crowding distance.
Step 3.4: calculating an eigenvalue and an eigenvector aiming at the covariance matrix; and adjusting the position of each particle by combining a position updating formula of the improved particle swarm optimization. And judging whether the position meets the constraint condition, if not, the position needs to be adjusted.
Step 3.5: sorting the calculated characteristic values from large to small; and updating external storage data, storing the non-dominated solution set and deleting the dominated solution.
Step 3.6: and taking out the first K eigenvectors and eigenvalues, and performing backspacing to obtain the dimensionality-reduced eigenvector matrix. And updating the global optimal solution and the individual optimal solution. And (3) stopping iteration after the algorithm reaches the maximum iteration times or meets the iteration requirement, and returning to the step 2 if the iteration ending condition is not met.
Preferably, the step 3.1 of generating the smoke data feature vector matrix includes the following steps:
step 3.1.1: the method comprises the following steps of utilizing an improved particle swarm algorithm to finish scheduling of smoke data from a single target to multiple targets, and determining a fitness function when the multiple targets are scheduled to the multiple targets:
wherein gamma isR(D) For the classification error rate of the selected feature subset R with respect to the decision D, | s | is the size of the selected feature subset. | d | represents the total number of characteristics. Alpha and beta are two parameters corresponding to the classification accuracy and importance of the selected feature size, alpha 0,1]And β ═ 1- α.
Step 3.1.2: the use of the K-NN classifier in the fitness function reduces the error rate. In K-NN, each sample is divided into a class of labels to which its K neighbors mostly belong. In the classification phase, the data set is typically divided into a training subset and a testing subset. To determine the class of each sample in the test data, the nearest k neighbors of each sample must be computed from the training data. K-fold cross-validation with K-10 was used.
Preferably, the step 3.4 of calculating the eigenvalue and eigenvector of the covariance matrix is as follows: feature selection is performed using a multi-population based particle swarm algorithm. In the algorithm, each particle has two solutions, one solution is randomly generated, and the other solution is generated by a Relieff characteristic sorting method.
Step 3.4.1: the Relieff algorithm sorts the features by calculating the distance of each feature to the target. In this method, each feature is assigned a weight, ranging from 1 to + 1. It is desirable that the relevant features have a higher weight. The algorithm searches the solution space with both solutions simultaneously. The particle then adjusts its position using the gBest and pBest solutions.
In this algorithm, the initial velocity value of the particle is set to zero. There are two initialization solutions per particle. Initial solution, xi,kWhere i is the index k of the particle is the initial type. x is the number ofi,0Representing two initial solutions xi,kThe best solution of (1). The random selection selects a number from 0 to 1 based on a uniform distribution. If the value is greater than 0.5, the position is set to 1, otherwise it is set to 0. The terrain ordering method assigns a value of [1,1 ] to each feature]. These correlation values are converted to [0,1 ]]An equation.
Wherein, theta to U (0, 1), rw is the correlation weight matrix,is the probability of occurrence of the ith feature, xi,2Initial position generated using Relieff. If the correlation value of any characteristic is negative, the correlation value of that characteristic is assumed to be 0. And calculating a global optimal solution of the group and a personal optimal solution of each particle according to the fitness value of the initialized particle.
Step 3.4.2: for each particle, a time-varying mirror sigmoid transfer function is used to adjust the next position of the particle.
In the formula, I represents the I particle, k is the initial type, j is the dimension of the I particle characteristic, and w is the inertia weight for adjusting the current sum of the particlesEquilibrium between the next positions. c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers between 0 and 1.Is pBestiThe jth position of (1), gBestjIs the jth position of gBest.
Step 3.4.3: the position of the particle is a continuous value that cannot be used directly for feature selection due to its binary nature. To solve this problem, a transfer function is used to convert successive values into binary values. The transfer function is as follows.
Wherein σ is σ at the first iterationminσ as σ at the last iterationmaxLinear increase, switching smoothly from exploration to development. σ is given as
The next binary position of each transfer function is obtained using the equation. The objective function is inAndgreedy selection is made between them. Then, the optimal position is selected as the particleThe next binary position. Will be provided withIs given to
If the new fitness value is better than the human best experience value for the particle, then the pairAssigned value of pBesti. If pBestiBetter than the current global best, then pBestiAnd assigning a value to the gBest, thereby updating the gBest.
The invention combines a hardware platform of a composite smoke sensing component and an intelligent algorithm, wherein the hardware platform adopts a single chip system to cooperate with various sensors to finish the acquisition of the original smoke data; connecting the smoke sensing device with a cloud platform by using a wireless network, and transmitting the acquired original data to the cloud platform; the improved particle swarm algorithm is utilized to analyze and solve data in the cloud platform, when harmful smoke in gas is higher than a standard threshold value, the smoke is sensed and an alarm is given, the method can effectively improve the sensitivity of composite smoke, and the false alarm rate is reduced.
Drawings
FIG. 1 is a flow chart of the overall steps of the present invention
FIG. 2 is a system architecture diagram of the present invention
FIG. 3 is a flow chart of feature vector extraction according to the present invention
FIG. 4 is a flow chart of the algorithm operation of the present invention
Detailed Description
It should be noted that the embodiments and features of the embodiments can be combined with each other without conflict, and the present application will be further described in detail with reference to the drawings and specific embodiments.
As shown in fig. 1, a high-sensitivity composite smoke-sensing low-false-alarm method based on an intelligent algorithm is divided into a hardware platform and an intelligent algorithm, and the method comprises the following steps as shown in fig. 1:
step 1: the hardware platform adopts a single chip microcomputer system to cooperate with various sensors to finish the acquisition of the smoke original data;
step 2: connecting the smoke sensing device with a cloud platform by using a wireless network, and transmitting the acquired original data to the cloud platform;
and step 3: and (3) performing data analysis and calculation by using an improved particle swarm algorithm in the cloud platform, when harmful smoke quantity in the gas is higher than a standard threshold value, giving an alarm when smoke feeling is generated, and repeating the step (1) when the harmful smoke quantity is lower than the standard threshold value.
In the invention, the system adopts an LPC4078 singlechip with REM architecture, and the sensors comprise a temperature sensor and an ion type smoke sensor.
In the invention, as shown in fig. 2, a super system of the cloud platform adopts a NET Framework + C # heterogeneous system, and NET Framework is applied to abundant display components, so that the reliability and stability of the platform are improved. The platform is divided into a service supporting layer, an application system layer, a basic platform layer and a data resource layer. And (4) building a cloud infrastructure of the resource system and the control system, and performing related virtualization and data center network computing. And the system scheduling, safety and flexible management and allocation of resources are realized through a standardized interface.
In the invention, an application layer mainly realizes panoramic Visual display of a communication scheduling picture through a NET Framework component, and utilizes Visual Studio2010 to realize development of a Visual system, and the application layer provides a uniform calling interface agent service;
in the invention, the service support layer mainly provides interface service for service logic access between the application layer and the application system layer. The application system layer realizes development of middleware such as data acquisition, interface management, timing scheduling and the like based on the C # technology;
in the invention, an application system layer realizes information transmission between a data resource layer and a service supporting layer through an interface by a NetBEUI protocol;
in the invention, the basic platform layer provides resources such as calculation, storage and the like required by a user, and realizes the resource allocation and rapid deployment as required by pooling of the resources through technologies such as virtualization and the like;
in the invention, the data resource layer is a visual database, stores information of different application systems, including equipment detection test information, equipment operation and fault information, and provides data support for communication scheduling of the platform.
In the invention, the improved particle swarm algorithm for data analysis and calculation in step 3 comprises the following steps, as shown in fig. 3:
step 3.1: generating a feature vector matrix for the smoke data; generating n particles, and initializing the initial positions of the n particles in the population, wherein the dimension of each particle is the same as that of the search space.
Step 3.2: firstly, calculating the average value of each column of characteristics; then each dimension needs to subtract the characteristic mean value of the column; and (4) carrying out constraint condition processing, calculating the fitness of each particle, forming an individual optimal value solution set and a global optimal value solution set according to the non-domination relation, and storing data.
Step 3.3: calculating a covariance matrix of the features; and determining an individual optimal value and a global optimal value of each particle through the dominance relation and the crowding distance.
Step 3.4: calculating an eigenvalue and an eigenvector aiming at the covariance matrix; and adjusting the position of each particle by combining a position updating formula of the improved particle swarm optimization. And judging whether the position meets the constraint condition, if not, the position needs to be adjusted.
Step 3.5: sorting the calculated characteristic values from large to small; and updating external storage data, storing the non-dominated solution set and deleting the dominated solution.
Step 3.6: taking out the first K eigenvectors and eigenvalues, and performing backspacing to obtain a dimensionality-reduced eigenvector matrix; and updating the global optimal solution and the individual optimal solution, stopping iteration after the algorithm reaches the maximum iteration times or meets the iteration requirement, and returning to the step 2 if the iteration end condition is not met.
In the invention, step 3.1 of generating the smoke data feature vector matrix comprises the following steps:
step 3.1.1: particle Swarm Optimization (PSO) is a meta-heuristic search technique that mimics the motion of a flock of birds to find the food and location of each particle. In particle swarm optimization, each particle represents a population solution and is evaluated by a predefined fitness function. One group is a set of solutions containing N particles. Each particle is a form of a candidate solution having a vector,d, i and j represent the quantitative characteristics, the population of the index, and the index of the property, respectively. XiIs a binary vector with a value of 1 or 0.
As shown in fig. 4, in BPSO, each particle is generally randomly initialized. In the present invention, two initialization techniques, z, are evaluated, namely random initialization and correlation value initialization based on the Relieff sorting method. Particle swarm optimization algorithms find the best solution by adjusting each particle with the personal best and global best information. gBest is the solution with the highest fitness value in the population, and pBest is the solution with the highest fitness value in the particle. The rate of change of position (velocity) of the ith particle is denoted Vi={vi 1,vi 2,vi j,...,vi d}。vi jIs at a predefined value of VmaxAnd VminTo avoid local optimality.
The next velocity vector is calculated using the current velocity, the local best position and the best position of the population (as shown in equation (1)).
Inertia weighted value wmaxInitialization, gradually decreasing to w by equation (2)minWhereinFor the position of the ith particle of the jth dimension solution, the present invention sets c1 and c2 to 2.
To perform feature selection using BPSO, the value of the velocity vector must be converted into a binary string by equation (6). r represents the shapes (s-type and v-type) of the transfer functions in the formulas (3) and (4), and the transfer functions are classified into v-type and s-type according to the shapes.
Tv(φ)=|tanh(φ)| (4)
The feature selection can be regarded as a multi-objective optimization problem, and the goal of the feature selection is to realize two mutually contradictory goals; higher classification accuracy and a smaller number of selected features. The goal is to achieve better accuracy with a minimum number of features, usually with classification error as the evaluation function. However, for feature selection, the number of selected features should also be considered in the merit function. To this end, the present invention employs the fitness function given in equation (7).
Wherein gamma isR(D) For the classification error rate of the selected feature subset R with respect to the decision D, | s | is the size of the selected feature subset. | d | represents the total number of characteristics. Alpha and beta are two parameters corresponding to the classification accuracy and importance of the selected feature size, alpha 0,1]And β ═ 1- α. In the present invention, the significant impact weight is assigned to the classification accuracy rather than the number of attributes.
Step 3.1.2: in the present invention, a K-NN classifier is used in the fitness function. In K-NN, each sample is divided into a class of labels to which its K neighbors mostly belong. In the classification phase, the data set is typically divided into a training subset and a testing subset. To determine the class of each sample in the test data, the nearest k neighbors of each sample must be computed from the training data. In the present invention, K-fold cross-validation with K-10 was used.
Conventional particle swarm algorithms typically start with a randomly generated population. However, particle swarm algorithms are sensitive to initialization and tend to fall into local optima, especially in high-dimensional feature spaces. In the present invention, in order to make the search space more diversified, we propose a particle swarm algorithm based on multiple clusters to perform feature selection. In the algorithm, each particle has two solutions, one solution is randomly generated, and the other solution is generated by a Relieff characteristic sorting method.
Step 3.4.1: the Relieff algorithm is a method of sorting features by calculating the distance of each feature to a target. In this method, each feature is assigned a weight, ranging from 1 to + 1. It is desirable that the relevant features have a higher weight. The algorithm searches the solution space with both solutions simultaneously. The particle then adjusts its position using the gBest and pBest solutions. FIG. 1 summarizes the proposed concept, where PiThe ith particle is shown. Xi,0And vi,0Current position and velocity, respectively, and Xi,1And Xi,2Is a Chinese character' tongCompeting solutions are initialized by different initialization techniques. Vi,1And Vi,2Is the speed of the solution. The algorithm 3 also gives details of the algorithm.
In this algorithm, the initial velocity value of the particle is set to zero. There are two initialization solutions per particle. Initial solution, xi,kWhere i is the index k of the particle is the initial type. x is the number ofi,0Representing two initial solutions xi,kThe best solution of (1). The random selection selects a number from 0 to 1 based on a uniform distribution. If the value is greater than 0.5, the position is set to 1, otherwise it is set to 0. The terrain ordering method assigns a value of [1,1 ] to each feature]. These correlation values are converted to [0,1 ]]An equation. (8) And (9).
Wherein, theta to U (0, 1), rw is the correlation weight matrix,is the probability of occurrence of the ith feature, xi,2Initial position generated using Relieff. Note that if the correlation value of any characteristic is negative, the correlation value of the characteristic is assumed to be 0.
And calculating a global optimal solution of the group and a personal optimal solution of each particle according to the fitness value of the initialized particle. The initial solution generation mechanism of the method is given.
Step 3.4.2: for each particle, the new velocity is calculated according to equation (10). In order to adjust the next position of the particle, a time-varying mirror sigmoid transfer function is proposed, and good results are obtained for the binary problem. In the present invention, we modify the function to be able to handle the feature selection of multiple populations.Andare obtained from the formulae (11) and (12), respectively. By selecting of formula (16)Andthe best position in between as the next position. The suboptimal solution (gBest) is updated based on these new particle fitness values.
In the formula, I represents the ith particle, k is the initial type, j is the dimension of the characteristic of the ith particle, and w is the inertia weight, and is used for adjusting the balance between the current position and the next position of the particle. c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers between 0 and 1.Is pBestiThe jth position of (1), gBestjIs the jth position of gBest.
Step 3.4.3: the position of the particle is a continuous value that cannot be used directly for feature selection due to its binary nature. To solve this problem, a transfer function is used to convert successive values into binary values. The transfer function is shown in equations (13) and (14).
Wherein σ is σ at the first iterationminσ as σ at the last iterationmaxLinear increase, switching smoothly from exploration to development. σ is given as
The next binary position of each transfer function is obtained using the equation. Respectively (11) and (12). Then, based on the objective function as shown in equation (16), inAndgreedy selection is made between them. Then, the optimal position is selected as the particleThe next binary position. Will be provided withIs given to
If the new fitness value is better than the human best experience value for the particle, then the pairAssigned value of pBesti. If pBestiBetter than the current global best, then pBestiThe value is assigned to the gBest to which,thereby updating the gBest.
According to the invention, a hardware platform and an intelligent algorithm are combined, and the hardware platform adopts a single chip microcomputer system to cooperate with various sensors to complete the acquisition of smoke original data; connecting the smoke sensing device with a cloud platform by using a wireless network, and transmitting the acquired original data to the cloud platform; the improved particle swarm algorithm is utilized to analyze and solve data in the cloud platform, when harmful smoke in gas is higher than a standard threshold value, the smoke is sensed and an alarm is given, the method can effectively improve the sensitivity of composite smoke, and the false alarm rate is reduced.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made herein without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims (8)
1. A high-sensitivity composite smoke-sensing low-false-alarm method based on an intelligent algorithm is characterized by comprising a hardware platform and the intelligent algorithm, and the method comprises the following steps:
step 1: the hardware platform adopts a single chip microcomputer system to cooperate with various sensors to finish the acquisition of the smoke original data;
step 2: connecting the smoke sensing device with a cloud platform by using a wireless network, and transmitting the acquired original data to the cloud platform;
and step 3: and (3) performing data analysis and calculation by using an improved particle swarm algorithm in the cloud platform, when harmful smoke quantity in the gas is higher than a standard threshold value, giving an alarm when smoke feeling is generated, and repeating the step (1) when the harmful smoke quantity is lower than the standard threshold value.
2. The high-sensitivity composite smoke sensing low-false alarm method based on the intelligent algorithm as claimed in claim 1, wherein the system adopts an LPC4078 single chip microcomputer with an REM framework, and the sensors comprise a temperature sensor and an ionic smoke sensor.
3. The high-sensitivity composite smoke sensing low-false-alarm method based on the intelligent algorithm as claimed in claim 1, wherein a NET Framework + C # heterogeneous system is adopted in the super system of the cloud platform, a NET Framework rich presentation component is applied, and meanwhile, the reliability and stability of the platform are improved. The platform is divided into a service supporting layer, an application system layer, a basic platform layer and a data resource layer. And (4) building a cloud infrastructure of the resource system and the control system, and performing related virtualization and data center network computing. And the system scheduling, safety and flexible management and allocation of resources are realized through a standardized interface.
4. The high-sensitivity composite smoke-sensing low-false-alarm method based on the intelligent algorithm as claimed in claim 3, wherein the application layer mainly realizes panoramic Visual display of a communication scheduling picture through a NET Framework component, realizes development of a Visual system by using Visual Studio2010, and provides a uniform call interface agent service;
the service support layer mainly provides interface service for service logic access between the application layer and the application system layer. The application system layer realizes development of middleware such as data acquisition, interface management, timing scheduling and the like based on the C # technology;
the application system layer realizes information transmission between the application system layer and the data resource layer and between the application system layer and the service supporting layer through an interface by a NetBEUI protocol;
the basic platform layer provides resources such as calculation, storage and the like required by the user, and realizes the resource allocation and rapid deployment as required through the resource pooling of technologies such as virtualization and the like;
the data resource layer is a visual database, stores information of different application systems, comprises equipment detection test information and equipment operation and fault information, and provides data support for communication scheduling of the platform.
5. The high-sensitivity composite smoke-feeling low-false-alarm method based on the intelligent algorithm as claimed in claim 1, wherein the step 3 of improving the particle swarm algorithm to analyze and solve the data comprises the following steps:
step 3.1: generating a feature vector matrix for the smoke data; generating n particles, and initializing the initial positions of the n particles in the population, wherein the dimension of each particle is the same as that of the search space.
Step 3.2: calculating the average value of each column of features, and then subtracting the average value of the features of the column from each dimension; and (4) carrying out constraint condition processing, calculating the fitness of each particle, forming an individual optimal value solution set and a global optimal value solution set according to the non-domination relation, and storing data.
Step 3.3: calculating a covariance matrix of the features; and determining an individual optimal value and a global optimal value of each particle through the dominance relation and the crowding distance.
Step 3.4: calculating an eigenvalue and an eigenvector aiming at the covariance matrix; and adjusting the position of each particle by combining a position updating formula of the improved particle swarm optimization. And judging whether the position meets the constraint condition, if not, the position needs to be adjusted.
Step 3.5: sorting the calculated characteristic values from large to small; and updating external storage data, storing the non-dominated solution set and deleting the dominated solution.
Step 3.6: and taking out the first K eigenvectors and eigenvalues, and performing backspacing to obtain the dimensionality-reduced eigenvector matrix. And updating the global optimal solution and the individual optimal solution. And (3) stopping iteration after the algorithm reaches the maximum iteration times or meets the iteration requirement, and returning to the step 2 if the iteration ending condition is not met.
6. A high-sensitivity composite smoke-feeling low false alarm method based on intelligent algorithm as claimed in claim 5, wherein said step 3.1 is to generate the smoke data eigenvector matrix, comprising the following steps:
step 3.1.1: the method comprises the following steps of utilizing an improved particle swarm algorithm to finish scheduling of smoke data from a single target to multiple targets, and determining a fitness function when the multiple targets are scheduled to the multiple targets:
wherein gamma isR(D) For the classification error rate of the selected feature subset R with respect to the decision D, | s | is the size of the selected feature subset. | d | represents the total number of characteristics. Alpha and beta are two parameters corresponding to the classification accuracy and importance of the selected feature size, alpha 0,1]And β ═ 1- α.
Step 3.1.2: the use of the K-NN classifier in the fitness function reduces the error rate. In K-NN, each sample is divided into a class of labels to which its K neighbors mostly belong. In the classification phase, the data set is typically divided into a training subset and a testing subset. To determine the class of each sample in the test data, the nearest k neighbors of each sample must be computed from the training data. K-fold cross-validation with K-10 was used.
7. A high-sensitivity composite smoke sense low false alarm method based on an intelligent algorithm is characterized in that the step 3.4 of calculating the eigenvalue and the eigenvector of a covariance matrix comprises the following steps: feature selection is performed using a multi-population based particle swarm algorithm. In the algorithm, each particle has two solutions, one solution is randomly generated, and the other solution is generated by a Relieff characteristic sorting method.
Step 3.4.1: the Relieff algorithm sorts the features by calculating the distance of each feature to the target. In this method, each feature is assigned a weight, ranging from 1 to + 1. It is desirable that the relevant features have a higher weight. The algorithm searches the solution space with both solutions simultaneously. The particle then adjusts its position using the gBest and pBest solutions.
In this algorithm, the initial velocity value of the particle is set to zero. There are two initialization solutions per particle. Initial solution, xi,kWhere i is the index k of the particle is the initial type. x is the number ofi,0Representing two initial solutions xi,kThe best solution of (1). The random selection selects a number from 0 to 1 based on a uniform distribution. If the value is greater than 0.5, the position is set to 1, otherwise it is set to 0. The terrain ordering method assigns a value of [1,1 ] to each feature]. These correlation valuesIs converted into [0,1 ]]An equation.
Wherein, theta to U (0, 1), rw is the correlation weight matrix,is the probability of occurrence of the ith feature, xi,2Initial position generated using Relieff. If the correlation value of any characteristic is negative, the correlation value of that characteristic is assumed to be 0. And calculating a global optimal solution of the group and a personal optimal solution of each particle according to the fitness value of the initialized particle.
8. The high-sensitivity composite smoke-feeling low-false-alarm method based on the intelligent algorithm as claimed in claim 6, characterized in that:
step 3.4.2: for each particle, a time-varying mirror sigmoid transfer function is used to adjust the next position of the particle.
In the formula, I represents the ith particle, k is the initial type, j is the dimension of the characteristic of the ith particle, and w is the inertia weight, and is used for adjusting the balance between the current position and the next position of the particle. c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers between 0 and 1.Is pBestiThe jth position of (1), gBestjIs the jth position of gBest.
Step 3.4.3: the position of the particle is a continuous value that cannot be used directly for feature selection due to its binary nature. To solve this problem, a transfer function is used to convert successive values into binary values. The transfer function is as follows.
Wherein σ is σ at the first iterationminσ as σ at the last iterationmaxLinear increase, switching smoothly from exploration to development. σ is given as
The next binary position of each transfer function is obtained using the equation. The objective function is inAndgreedy selection is made between them. Then, the optimal position is selected as the particleThe next binary position. Will be provided withIs given to
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110700540.0A CN113436404A (en) | 2021-06-23 | 2021-06-23 | High-sensitivity composite smoke-sensitive low-false-alarm method based on intelligent algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110700540.0A CN113436404A (en) | 2021-06-23 | 2021-06-23 | High-sensitivity composite smoke-sensitive low-false-alarm method based on intelligent algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113436404A true CN113436404A (en) | 2021-09-24 |
Family
ID=77753605
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110700540.0A Pending CN113436404A (en) | 2021-06-23 | 2021-06-23 | High-sensitivity composite smoke-sensitive low-false-alarm method based on intelligent algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113436404A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103699943A (en) * | 2013-12-27 | 2014-04-02 | 长春工业大学 | GA-PSOBP algorithm-based geological disaster risk evaluation method |
CN111613037A (en) * | 2020-04-30 | 2020-09-01 | 杭州拓深科技有限公司 | Method for reducing composite smoke sense false alarm based on intelligent algorithm |
US20200324117A1 (en) * | 2019-04-15 | 2020-10-15 | Advanced Neuromodulation Systems, Inc. | Systems and methods of generating stimulation patterns |
CN112153131A (en) * | 2020-09-15 | 2020-12-29 | 山东钢铁集团日照有限公司 | Iron and steel quality private cloud platform construction method based on super-fusion technology |
CN112887994A (en) * | 2021-01-20 | 2021-06-01 | 华南农业大学 | Wireless sensor network optimization method based on improved binary particle swarm and application |
-
2021
- 2021-06-23 CN CN202110700540.0A patent/CN113436404A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103699943A (en) * | 2013-12-27 | 2014-04-02 | 长春工业大学 | GA-PSOBP algorithm-based geological disaster risk evaluation method |
US20200324117A1 (en) * | 2019-04-15 | 2020-10-15 | Advanced Neuromodulation Systems, Inc. | Systems and methods of generating stimulation patterns |
CN111613037A (en) * | 2020-04-30 | 2020-09-01 | 杭州拓深科技有限公司 | Method for reducing composite smoke sense false alarm based on intelligent algorithm |
CN112153131A (en) * | 2020-09-15 | 2020-12-29 | 山东钢铁集团日照有限公司 | Iron and steel quality private cloud platform construction method based on super-fusion technology |
CN112887994A (en) * | 2021-01-20 | 2021-06-01 | 华南农业大学 | Wireless sensor network optimization method based on improved binary particle swarm and application |
Non-Patent Citations (2)
Title |
---|
刘平山: "A Multi-objective Particle Swarm Optimization Data Scheduling Algorithm for Peer-to-Peer Video Streaming", 《2017 13TH INTERNATIONAL CONFERENCE ON NATURAL COMPUTATION, FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY》 * |
毛恒: "粒子群优化算法的改进及应用研究", 《中国优秀博士学位论文全文数据库 》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111626350B (en) | Target detection model training method, target detection method and device | |
JP5515647B2 (en) | Positioning device | |
KR101081051B1 (en) | A method for detecting fire-flame using fuzzy finite automata | |
CN111062429A (en) | Chef cap and mask wearing detection method based on deep learning | |
JP2004280812A (en) | Method or system for selecting attribute type used for classifier | |
CN112200021B (en) | Target crowd tracking and monitoring method based on limited range scene | |
CN113936413B (en) | Early fire monitoring and early warning method and device | |
CN110309887A (en) | Based on the Fuzzy C-Means Clustering method for detecting abnormality for improving flower pollination | |
CN109859236A (en) | Mobile object detection method, calculates equipment and storage medium at system | |
CN113223264A (en) | QPSO-BP neural network-based intelligent fire early warning system and method | |
Yang et al. | Multiple human location in a distributed binary pyroelectric infrared sensor network | |
CN108809989A (en) | A kind of detection method and device of Botnet | |
CN116664359A (en) | Intelligent fire early warning decision system and method based on multi-sensor fusion | |
CN110335313A (en) | Audio collecting device localization method and device, method for distinguishing speek person and system | |
CN113837302B (en) | City safety management monitoring method and system based on Internet of things and electronic equipment | |
CN116665001A (en) | Intelligent park multisource data dynamic monitoring and real-time analysis system and method | |
CN113436404A (en) | High-sensitivity composite smoke-sensitive low-false-alarm method based on intelligent algorithm | |
CN114141385A (en) | Early warning method and system for infectious diseases and readable storage medium | |
Liu et al. | A new multi-sensor fire detection method based on LSTM networks with environmental information fusion | |
CN109002746A (en) | 3D solid fire identification method and system | |
CN108668254B (en) | WiFi signal characteristic area positioning method based on improved BP neural network | |
CN114120208A (en) | Flame detection method, device, equipment and storage medium | |
CN107172062B (en) | A kind of intrusion detection method based on biological immune T cell receptor mechanism | |
CN115510900A (en) | Automatic circuit fault diagnosis method and system | |
CN113884807A (en) | Power distribution network fault prediction method based on random forest and multi-layer architecture clustering |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210924 |
|
RJ01 | Rejection of invention patent application after publication |