CN108903936B - Intelligent mining helmet control method based on human body information and environment information fusion - Google Patents

Intelligent mining helmet control method based on human body information and environment information fusion Download PDF

Info

Publication number
CN108903936B
CN108903936B CN201810715885.1A CN201810715885A CN108903936B CN 108903936 B CN108903936 B CN 108903936B CN 201810715885 A CN201810715885 A CN 201810715885A CN 108903936 B CN108903936 B CN 108903936B
Authority
CN
China
Prior art keywords
wavelet
neural network
frequency
brain wave
electroencephalogram
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.)
Active
Application number
CN201810715885.1A
Other languages
Chinese (zh)
Other versions
CN108903936A (en
Inventor
汪梅
牛钦
翟珂
王刚
张佳楠
张思明
张松志
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Science and Technology
Original Assignee
Xian University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian University of Science and Technology filed Critical Xian University of Science and Technology
Priority to CN201810715885.1A priority Critical patent/CN108903936B/en
Publication of CN108903936A publication Critical patent/CN108903936A/en
Application granted granted Critical
Publication of CN108903936B publication Critical patent/CN108903936B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/369—Electroencephalography [EEG]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
    • A61B5/165—Evaluating the state of mind, e.g. depression, anxiety
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
    • A61B5/18—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state for vehicle drivers or machine operators
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/369—Electroencephalography [EEG]
    • A61B5/372—Analysis of electroencephalograms
    • A61B5/374—Detecting the frequency distribution of signals, e.g. detecting delta, theta, alpha, beta or gamma waves
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
    • A61B5/6802—Sensor mounted on worn items
    • A61B5/6803—Head-worn items, e.g. helmets, masks, headphones or goggles
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
    • A61B5/6802—Sensor mounted on worn items
    • A61B5/681—Wristwatch-type devices
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7253—Details of waveform analysis characterised by using transforms
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/74—Details of notification to user or communication with user or patient ; user input means
    • A61B5/746—Alarms related to a physiological condition, e.g. details of setting alarm thresholds or avoiding false alarms
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2503/00—Evaluating a particular growth phase or type of persons or animals
    • A61B2503/20—Workers
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2560/00—Constructional details of operational features of apparatus; Accessories for medical measuring apparatus
    • A61B2560/02—Operational features
    • A61B2560/0242—Operational features adapted to measure environmental factors, e.g. temperature, pollution

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Pathology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Veterinary Medicine (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Public Health (AREA)
  • Psychiatry (AREA)
  • Psychology (AREA)
  • Artificial Intelligence (AREA)
  • Physiology (AREA)
  • Child & Adolescent Psychology (AREA)
  • Hospice & Palliative Care (AREA)
  • Social Psychology (AREA)
  • Educational Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Developmental Disabilities (AREA)
  • Signal Processing (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention discloses an intelligent mining helmet control method based on human body information and environment information fusion, which comprises the steps of obtaining an original brain wave signal; preprocessing original brain waves; wavelet decomposition reconstruction of the original brain wave signal; extracting brain wave signal features; optimizing a BP neural network by an artificial fish swarm algorithm to identify fatigue characteristics; identifying the fatigue degree of miners, monitoring the environmental information of the miners and carrying out alarm operation on dangerous conditions. The method extracts four sub-band energies, ApEn approximate entropy, KC complexity and C in the high-precision brain wave signal obtained after preprocessing (filtering and denoising) on the original brain wave signal0The complexity is used as an input layer of the artificial fish swarm optimization BP neural network, the artificial fish swarm optimization BP neural network is trained to identify the fatigue of miners, the identification accuracy is high, and the intelligent wearable bracelet worn on the wrist of the miners can give an alarm in three modes of sound, light and vibration to the safety states with different danger levels, so that the safety coefficient of underground operation is improved.

Description

Intelligent mining helmet control method based on human body information and environment information fusion
Technical Field
The invention relates to the technical field of miner fatigue identification, in particular to an intelligent mining helmet control method based on human body information and environment information fusion.
Background
China is the largest coal producing country around the world. On the basis of the technical background of the existing mine, the traditional mine helmet cannot meet the market demand, so that the mine helmet needs to be intelligently upgraded by combining a digital mine monitoring system. The intelligent helmet for the mine is researched, monitoring and early warning of complex underground environments can be completed, specific information of miners is provided for an aboveground monitoring room, and reliable data is provided for accident elimination, rescue and accident reason analysis. The existing mining helmet monitoring system is as follows:
1. the mine alarm mode is mostly that the mine alarm mode is installed in a roadway to carry out sound-light alarm, and for personnel deviating from an alarm source (such as under the environment of super noise and much dust in the same line of tunneling), the situations that alarm information is not received in time exist, namely the real-time performance of the information cannot be ensured.
2. Colliery sensor often monitors environmental information etc. in fixed position, when personnel walked this position, just can't know the information etc. of self environment, had the potential safety hazard.
3. Most of the existing mining helmets only complete monitoring, voice communication and the like of the environment where miners are located, basically, the miners are not monitored, and detection and evaluation on the aspects of the fatigue degree of the miners are not performed.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides an intelligent mining helmet control method based on human body information and environment information fusion.
In order to achieve the purpose, the invention is implemented according to the following technical scheme:
an intelligent mining helmet control method based on human body information and environment information fusion comprises the following steps:
s1, acquisition of original brain wave signals: acquiring an original brain wave signal of a miner through an electroencephalogram signal acquisition device mounted on a mining helmet, preprocessing the original brain wave signal to obtain an original brain wave signal, and then sending the original brain wave signal to an electroencephalogram signal processor mounted on the rear side of a mining body; meanwhile, the infrared composite five-in-one gas sensor and the DHT11 temperature and humidity sensor are adopted to measure the concentrations of carbon dioxide, methane, carbon monoxide, formaldehyde and volatile organic compounds, and gas parameters are transmitted to the main controller;
the electroencephalogram signal acquisition device comprises a first electroencephalogram electrode, a second electroencephalogram electrode, a third electroencephalogram electrode and an electroencephalogram signal processor, wherein the first electroencephalogram electrode is arranged on the inner surface of the mining helmet body and used for acquiring mental states of the forehead part on the right side of the brain, the second electroencephalogram electrode is used for acquiring electric potentials at earlobes and shielding reference signals, the third electroencephalogram electrode is used for shielding artifact signals below the brain, the electroencephalogram signal processor is arranged on the rear side of the mining helmet body and used for preprocessing signals acquired by the first electroencephalogram electrode, the second electroencephalogram electrode and the third electroencephalogram electrode, and the signal output end of the electroencephalogram signal processor is connected with the signal input;
s2, denoising the original brain wave signal, wherein the process is as follows:
s21, selecting a wavelet basis function most similar to the original signal in the electroencephalogram signal processor, determining the decomposition scale of wavelet transformation, and performing wavelet transformation on the signal by using a Mallat tower method to obtain high-frequency coefficient components and low-frequency coefficient components with different decomposition scales;
s22, calculating a wavelet threshold value on the j scale, comparing the wavelet entropies of n subband signals, selecting the wavelet coefficient of the subband with the largest wavelet entropy value, considering that the wavelet coefficient of the subband is caused by noise, calculating the median value of the wavelet coefficient of the subband as the noise variance of the j scale, and thus calculating the wavelet threshold value of the j scale;
s23, because the wavelet coefficient values of the noise on different scales are different, the wavelet coefficient of the noise is smaller and smaller with the increase of the decomposition scale, the wavelet threshold values of different scales are respectively calculated according to S22, and thresholding processing is carried out on the high-frequency coefficient component of each scale to obtain an approximate high-frequency wavelet coefficient;
s24, utilizing the low-frequency coefficient component of the highest wavelet decomposition layer and the approximate high-frequency wavelet coefficient components of different scales after threshold processing to form the coefficient component required by signal reconstruction, and reconstructing according to the reconstruction mode of multi-resolution analysis to obtain a pure brain wave signal;
s3, extracting brain wave features, wherein the process is as follows:
s31, selecting a wavelet basis to perform wavelet decomposition reconstruction on the brain wave signals: the electroencephalogram signal processor performs wavelet decomposition on electroencephalogram signals acquired by the electroencephalogram signal acquisition module by using a sym5 wavelet basis function, wherein the signal decomposition expression is as follows:
Figure BDA0001717396450000031
the detail coefficients in the formula are:
Figure BDA0001717396450000032
the approximation coefficient in the formula is:
Figure BDA0001717396450000033
the mesoscale function is:
Figure BDA0001717396450000034
the wavelet function in the formula is:
Figure BDA0001717396450000035
the final signal reconstruction expression is then:
Figure BDA0001717396450000036
Figure BDA0001717396450000037
s32, decomposing brain wave signals by six layers by adopting a sym5 wavelet function to obtain six layers of low-frequency a6, six layers of high-frequency d6, five layers of high-frequency d5 and four layers of high-frequency d4 of the signals respectively; through frequency detection of FFT, six layers of low-frequency a6, six layers of high-frequency d6, five layers of high-frequency d5 and four layers of high-frequency d4 are found to be respectively approximate to the frequencies of delta waves, theta waves, alpha waves and beta waves, so that signals of wavelet decomposition are adopted to represent the delta waves, the theta waves, the alpha waves and the beta waves;
s33, carrying out wavelet decomposition on the original data to obtain delta (x), theta (x), alpha (x) and beta (x) sub-band signals of different frequency bands, wherein the sub-band energy is as follows:
Figure BDA0001717396450000041
Figure BDA0001717396450000042
calculate the energy ratio of each signal: eall＝E(δ)+E(θ)+E(α)+E(β)，En(δ)＝E(δ)/Eall，En(θ)＝E(θ)/Eall，En(α)＝E(α)/Eall，En(β)＝E(β)/EallTherefore, the ratio of sub-signals of EEG of miners in a fatigue state, a waking state and a concentration state can be quantitatively analyzed;
s34, calculating ApEn approximate entropy, KC complexity and C0The complexity is compared with the change conditions of the three characteristic values of the brain wave signals in a fatigue state, a waking state and a concentration state, and a plurality of groups of fatigue characteristic values, a plurality of groups of waking characteristic values and a plurality of groups of concentration characteristic values are respectively listed and stored and are subsequently used as input data of a neural network;
s4, optimizing the BP neural network by the artificial fish swarm algorithm to identify fatigue characteristics, wherein the process is as follows:
s41, obtaining the four sub-band energies obtained in S3, ApEn approximate entropy, KC complexity and C0The complexity is used as the input layer data of the neural network, and the fatigue, the wakefulness and the concentration states are represented by (1, 0, 0), (0,1, 0) and (0, 0,1) respectively as the output layer of the neural network;
s42, setting M neurons in the input layer, J neurons in the hidden layer and K neurons in the output layer in the BP neural network, and expressing the weight value from the input layer to the hidden layer as wijThe threshold of hidden layer neurons is denoted bjThe weight value from the hidden layer to the output layer is represented as vjkThe threshold of the output layer neurons is denoted as ak. The parameter to be adjusted in the BP neural network is the weight wij、vjkAnd a threshold value bj、akIf the parameters to be adjusted are set as the state of the artificial fish, the artificial fish x can be expressed as a vector with dimensions M × J + J × K + K:
x＝(w11,…,wM1,b1,…,wiJ,…,wMJ,bJ,v11,…,vJ1,a1…,v1k,…,vJK,aK) Wherein w is11,…,wM1Representing the weight of the input layer neuron to the first hidden layer neuron, b1 representing the threshold of the first neuron in the hidden layer, wiJ,…wMJRepresenting the weights of input layer neurons to the J-th hidden layer neuron, bJ representing the threshold of the J-th hidden layer neuron, v11,…,vJ1Representing the weight of the hidden layer neuron to the first output layer neuron, a1 representing the threshold of the first neuron in the output layer, v1k,…,vJKRepresenting weights from hidden layer neuron to Kth output layer neuron, aKThe valve value of the Kth neuron of the output layer is represented, the food concentration FC of the artificial fish is set to be the reciprocal of the total error E of the BP neural network, namely FC is 1/E, the point with the largest food concentration searched by the artificial fish is the point with the smallest error of the BP neural network, and any two artificial fish x are selected1、x2The euclidean distance d between them is expressed as:
Figure BDA0001717396450000051
in the formula x1、x2The elements in the BP neural network are subtracted in a one-to-one correspondence mode according to the dimensionality, the state of the artificial fish x is changed after foraging, clustering and rear-end collision are carried out on the artificial fish x, and meanwhile initial weight and threshold value are adjusted once for the BP neural network;
s43, selecting the neuron of the middle hidden layer as 20, inputting the sample as 17-dimensional data, and outputting the network as 1-dimensional data by the main controller according to the characteristics of the BP neural network error function and the design experience of predecessors in combination with experiments, namely setting the structure of the neural network as 17-20-1;
s44, selecting the number of artificial fishes as 10 in the artificial fish optimization algorithm, the visible area as 0.5, the crowding factor as 0.618, the step length as 0.1, the iteration frequency as 50 and the repeated exploration frequency as 50; the BP neural network activation function 1 selects a tansig function, the activation function 2 selects a logsig function, the training function adopts a trandx function, the maximum training time is set to 20000 times, the learning rate is 0.05,required accuracy of error is 10-4Momentum factor 0.9;
s45, pair of binary functions
Figure BDA0001717396450000061
-10 ≦ x, y ≦ 10) fitting to test the performance of the artificial fish school optimized BP network: taking 40000 data with the step length of 0.1, wherein 35000 data are taken as training samples, 5000 data are taken as testing samples, and the data are trained for 100 times;
s46, selecting 2100 groups of data, 600 groups of centralized state data, 600 groups of awake states and 600 groups of fatigue states from the electroencephalogram database, wherein 1800 groups are used for network training, the other 300 groups are used for network testing, and grouping is performed in a crossed mode. Recognizing the state of the brain wave signal by utilizing the neural network trained and tested in the steps to realize the monitoring of the fatigue of underground personnel;
s5, the main controller processes the concentration data of various sensors to judge the overrun state of the gas, fuses the current safety state and the danger level, sends the data to the interface displayed by the monitoring host through the existing underground communication node, and carries out triple alarm in three modes of sound, light and vibration on the safety state of different danger levels through the intelligent bracelet which is connected with the main controller through the Bluetooth module and is worn on the wrist of the miner.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention utilizes the maximum wavelet entropy to adaptively select the denoising threshold value of the electroencephalogram signal under the strong background noise, obtains a weak signal detection method based on the wavelet entropy, and realizes the extraction of fatigue characteristics in the sub-band signal energy information of delta waves, theta waves, alpha waves and beta waves.
2. The invention optimizes the initial weight of the BP neural network by utilizing the global optimization function of the artificial fish swarm algorithm, and obtains an artificial fish swarm-BP neural network classifier model. Simulation comparison shows that the identification accuracy of the artificial fish swarm-BP neural network model is improved by 2.3% compared with that of the BP neural network model.
3. The invention builds a mining intelligent helmet system based on a brain-computer interface, and completes the safety situation evaluation algorithm design based on environmental parameters and mental states; and the intelligent wearable bracelet is introduced into the underground of the coal mine, so that triple alarming of sound and light vibration and real-time information display are realized, the safety early warning level is greatly improved, the occurrence of accidents can be effectively reduced, and the intelligent wearable bracelet has great significance for mine safety production.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a pure electroencephalogram signal diagram extracted by wavelet entropy of the present invention: (a) the image is a brain wave signal image annihilated in a strong noise background, and the image (b) is a brain wave signal wavelet entropy dryness-removing result image.
FIG. 3 is a graph of the energy occupied by wavelets in different states of the present invention: (a) fatigue state, (b) waking state, (c) concentrating state.
FIG. 4 shows ApEn, KC and C under different states of the brain wave signal of the present invention0A value; (a) approximate entropy ApEn in different states, (b) KC value in different states, (C) C in different states0。
Fig. 5 is a structural diagram of the BP neural network for artificial fish school optimization according to the present invention.
FIG. 6 is a flow chart of the BP neural network optimization by the artificial fish swarm algorithm of the invention.
FIG. 7 is a three-dimensional diagram of a binary function of the present invention.
FIG. 8 is a comparison graph of the artificial fish swarm BP neural network algorithm of the present invention with the actual values and the test results of the conventional BP neural network algorithm.
FIG. 9 is a comparison graph of training results of the BP neural network and the AF-BP neural network of the present invention: (a) the error is a BP neural network training error, (b) the state is a BP neural network training state, (c) the error is an AF-BP neural network training error, and (d) the state is an AF-BP neural network training state.
Fig. 10 shows the gas explosion disaster risk early warning level of the smart band according to the embodiment of the present invention.
Fig. 11 is a design flow of processing the whole data of the data stream of the smart band according to the embodiment of the present invention.
FIG. 12 is a flow chart of a methane hazard zone bit according to an embodiment of the present invention.
Fig. 13 is a bubble sorting of the danger level flag bits of the smart band according to the present invention.
Fig. 14 is a flowchart illustrating the process of arranging the danger level flag bits of the smart band according to the present invention.
Fig. 15 is a flowchart illustrating a danger early warning operation of the smart bracelet according to the present invention.
Fig. 16 is a flowchart illustrating the operation of the intelligent bracelet danger alarm according to the present invention.
Fig. 17 is a circuit structure block diagram of the integrated intelligent mining helmet of the present invention.
Detailed Description
The present invention will be further described with reference to specific examples, which are illustrative of the invention and are not to be construed as limiting the invention.
The invention discloses an intelligent mining helmet control method based on human body information and environment information fusion, which is shown in figure 1 and comprises the following steps:
s1, acquisition of original brain wave signals: acquiring an original brain wave signal of a miner by an electroencephalogram signal acquisition device arranged on the mining helmet, preprocessing the original brain wave signal to obtain the original brain wave signal, and then sending the original brain wave signal to an electroencephalogram signal processor arranged on the rear side of the mining body, wherein the electroencephalogram signal processor supplies power by a rechargeable power supply arranged in the mining helmet; meanwhile, the infrared composite five-in-one gas sensor and the DHT11 temperature and humidity sensor are adopted to measure the concentrations of carbon dioxide, methane, carbon monoxide, formaldehyde and volatile organic compounds, and gas parameters are transmitted to the main controller;
as shown in fig. 17, the electroencephalogram signal acquisition device comprises a first electroencephalogram electrode, a second electroencephalogram electrode, a third electroencephalogram electrode and an electroencephalogram signal processor, wherein the first electroencephalogram electrode is mounted on the inner surface of the mining helmet body and used for acquiring mental states of the right forehead part of the brain, the second electroencephalogram electrode is used for acquiring electric potentials at earlobes and shielding reference signals, the third electroencephalogram electrode is used for shielding artifacts signals below the brain, the electroencephalogram signal processor is mounted on the rear side of the mining helmet body and used for preprocessing signals acquired by the first electroencephalogram electrode, the second electroencephalogram electrode and the third electroencephalogram electrode, and a signal output end of the electroencephalogram signal processor is connected with a signal input end;
s2, denoising the original brain wave signal, wherein the process is as follows:
the wavelet entropy theory is a theory of establishing similar information entropy based on a wavelet analysis method, and can quantitatively describe the energy distribution characteristics in the time-frequency domain. The coefficient matrix of wavelet transform is processed into a probability distribution sequence, and the entropy value calculated by the probability distribution sequence reflects the sparsity of the coefficient matrix. The wavelet entropy theory is to utilize the sparsity of the wavelet transform matrix to suppress the extraneous components.
According to the framework theory of wavelet transform, when the wavelet basis functions are a set of orthogonal bases, the wavelet transform has the property of energy conservation, namely:
Figure BDA0001717396450000091
the wavelet energy at a single scale is defined as the sum of the squares of the wavelet coefficients at that scale.
Figure BDA0001717396450000092
As can be seen from the characteristics of the orthogonal wavelet transform, the total power of the signal is equal to the sum of the powers of the components within a certain time window.
Figure BDA0001717396450000101
The high-frequency information amount of each decomposition scale is regarded as an independent signal source, each layer of high-frequency wavelet coefficient is divided into n equal small intervals, the wavelet entropy of each small interval is calculated, the median value of the small interval with the largest entropy value is selected as the variance of noise, and threshold value self-adaption selection based on the wavelet entropy is achieved.
Let the high-frequency wavelet coefficient of the j-th layer be dj(k) And if the sampling points are N and the wavelet coefficients on the sampling points are divided into N equal parts, the energy corresponding to the wavelet coefficient of the kth subinterval is as follows:
Figure BDA0001717396450000102
the total energy of the j-th layer high-frequency wavelet coefficient is expressed as:
Figure BDA0001717396450000103
the probability of the signal energy contained in the kth sub-interval existing in the total energy on the scale is set as:
Figure BDA0001717396450000104
defining the wavelet entropy of the signal corresponding to the kth sub-interval as:
Figure BDA0001717396450000105
s21, selecting a wavelet basis function which is most similar to the original signal, determining the decomposition scale of wavelet transformation, and performing wavelet transformation on the signal by using a Mallat tower method to obtain high-frequency coefficient components and low-frequency coefficient components of different decomposition scales;
s22, calculating a wavelet threshold value on the j scale, comparing the wavelet entropies of n subband signals, selecting the wavelet coefficient of the subband with the largest wavelet entropy value, considering that the wavelet coefficient of the subband is caused by noise, calculating the median value of the wavelet coefficient of the subband as the noise variance of the j scale, and thus calculating the wavelet threshold value of the j scale;
s23, because the wavelet coefficient values of the noise on different scales are different, the wavelet coefficient of the noise is smaller and smaller with the increase of the decomposition scale, the wavelet threshold values of different scales are respectively calculated according to S22, and thresholding processing is carried out on the high-frequency coefficient component of each scale to obtain an approximate high-frequency wavelet coefficient;
s24, utilizing the low-frequency coefficient component of the highest wavelet decomposition layer and the approximate high-frequency wavelet coefficient components of different scales after threshold processing to form the coefficient component required by signal reconstruction, and reconstructing according to the reconstruction mode of multi-resolution analysis to obtain a pure brain wave signal; the clean brain wave signal extracted by using wavelet entropy is shown in fig. 2.
S3, extracting brain wave features, wherein the process is as follows:
s31, selecting a wavelet basis to perform wavelet decomposition on the brain wave signal: the electroencephalogram signal processor carries out wavelet decomposition reconstruction on electroencephalogram signals collected by the electroencephalogram signal acquisition module by utilizing a sym5 wavelet basis function, and the signal decomposition expression is as follows:
Figure BDA0001717396450000111
the detail coefficients in the formula are:
Figure BDA0001717396450000112
the approximation coefficient in the formula is:
Figure BDA0001717396450000113
the mesoscale function is:
Figure BDA0001717396450000114
the wavelet function in the formula is:
Figure BDA0001717396450000115
the final signal reconstruction expression is then:
Figure BDA0001717396450000121
Figure BDA0001717396450000122
s32, decomposing brain wave signals by six layers by adopting a sym5 wavelet function to obtain six layers of low-frequency a6, six layers of high-frequency d6, five layers of high-frequency d5 and four layers of high-frequency d4 of the signals respectively; through frequency detection of FFT, six layers of low-frequency a6, six layers of high-frequency d6, five layers of high-frequency d5 and four layers of high-frequency d4 are found to be respectively approximate to the frequencies of delta waves, theta waves, alpha waves and beta waves, so that signals of wavelet decomposition are adopted to represent the delta waves, the theta waves, the alpha waves and the beta waves;
s33, carrying out wavelet decomposition on the original data to obtain delta (x), theta (x), alpha (x) and beta (x) sub-band signals of different frequency bands, wherein the sub-band energy is as follows:
Figure BDA0001717396450000123
Figure BDA0001717396450000124
calculate the energy ratio of each signal: eall＝E(δ)+E(θ)+E(α)+E(β)，En(δ)＝E(δ)/Eall，En(θ)＝E(θ)/Eall，En(α)＝E(α)/Eall，En(β)＝E(β)/EallTherefore, the ratio of sub-signals of EEG of miners in a fatigue state, a waking state and a concentration state can be quantitatively analyzed; the subband energies in the different states are shown in table 1.
TABLE 1 energy of each sub-band under different states
Figure BDA0001717396450000125
The results of the normalization process of the energy values in table 1 are shown in fig. 3. It can be seen from fig. 3 that the delta wave is dominant in the three states. Under the fatigue state, the sub-band energy of the brain wave signal is mainly concentrated on delta waves; the energy of delta wave is reduced in the waking state, and the energy of other waves is increased; in the concentrated state, the alpha wave energy is relatively outstanding. From the frequency perspective, as mental states increase, low frequency signal energy decreases and high frequency signal energy increases. The ratio of sub-signals of brain waves in different states can be quantitatively analyzed by utilizing wavelet decomposition, and the extracted features are obvious and easy to distinguish.
S34, calculating ApEn approximate entropy, KC complexity and C0Complexity, as shown in fig. 4. Comparing the change conditions of the three characteristic values of the brain wave signals in the fatigue state, the waking state and the concentration state, respectively listing and storing a plurality of groups of fatigue characteristic values, a plurality of groups of waking characteristic values and a plurality of groups of concentration characteristic values, and subsequently using the values as input data of a neural network; wherein ApEn approximate entropy and KC are complexDegree and C0The complexity calculation method specifically comprises the following steps:
(1) ApEn approximate entropy
Two parameters are artificially set in the process of approximate entropy calculation, one is a mode dimension m for reflecting the complexity change characteristics of the signal sequence under different dimensions, and the other is a similar tolerance r which is adjusted according to the situation in the process of programming, but once an ideal r value is obtained, the two parameters are kept constant in the subsequent program. The approximate entropy algorithm steps are as follows:
1) let the original data set be s(1),s(2),…,s(N)A total of N data points.
2) Will sequence s(i)Are sequentially arranged to form m-dimensional vectors S(i)，S(i)From S (1) to S(N-m+1)Wherein:
S(i)＝[s(i),s(i+1),…,s(i+m-1)],i＝1～N-m+1
for each value of i defined above, a vector S is calculated(i)With the remaining vector S(j)The distance between:
Figure BDA0001717396450000146
3) setting a similar tolerance r, counting d [ S ] for each value of i ≦ N-m +1(i),S(j)]The number smaller than r and the ratio of this number to the total number of distances N-m +1 are designated
Figure BDA0001717396450000141
Namely:
Figure BDA0001717396450000142
4) will be provided with
Figure BDA0001717396450000143
Taking the logarithm, and averaging the logarithm with all i, and recording as phim(r) is:
Figure BDA0001717396450000144
5) adding 1 to the dimension m, and repeating the steps (1) to (4) to obtain the product
Figure BDA0001717396450000145
And phim+1(r), then the approximate entropy may be defined as:
ApEn(m,r)＝Φm(r)-Φm+1(r)
in the present experiment, m is 2, r is 0.1 Std, where Std is the standard deviation of the raw data s (i).
Before analyzing the brain wave signals, firstly, denoising the original brain waves, wherein the sampling frequency of the brain wave signals is 512Hz, namely 512 data are collected per second, the brain wave signals are respectively extracted in three states, the total time of collecting 30s is 512 x 30 to 15360 data, the approximate entropy per second is calculated according to the algorithm, as shown in figure 4(a), the curve graph shows that the three states have obvious difference, the approximate entropy under the fatigue state is mainly concentrated between 0 and 0.2, the approximate entropy is very small, and the disorder degree of the brain thinking is reduced, and the complexity is reduced. The approximate entropy under the waking state and the concentration state is higher and is generally higher than 0.55, which shows that the brain thinking is active, the brain wave fluctuation range is large, the complexity is high, the complexity characteristic of the brain wave signal is better reflected through the calculation of the approximate entropy, the inherent nonlinear characteristic of the brain wave signal is embodied, and the identification precision and the accuracy of the state of the human body are improved to become the indexes for measuring the fatigue state.
(2) Complexity of KC
In the algorithm of Kasper design, the time signal sequence is first subjected to coarse grain processing, that is, each original sequence point is replaced by one bit, so that the whole signal sequence actually becomes a (0,1) sequence, and the specific coarse grain rule is as follows: calculating the average value of a time sequence, setting the value to 1 if the value in the sequence is larger than the average value, and otherwise, using 0 to represent the value. A new time series can be obtained by processing in this way.
Through the algorithm design, the number of different modes contained in the coarse grained sequence can be analyzed, and the specific algorithm process is as follows:
1) adding a character B after a coarse grained (0,1) sequence A (A usually starts with the first character, namely A takes s 1);
2) judging whether B belongs to a character string ABC (ABC is a result of subtracting the last character from AB), if B has appeared in the front, B is called a substring of ABC, and no new mode appears, and the process is called a copying process;
3) adding B to the back of the whole character string, continuously increasing the length of B, judging, if B does not appear in the character string ABC, then considering that a new mode appears, performing an inserting operation at the moment, inserting by using a special symbol such as 'x' and attaching B to the back of the symbol, and at the moment, dividing the two character strings into a front section and a rear section;
4) b is reconstructed, starting again with a character, repeating the above algorithm steps until all strings have been traversed, the appearance of a "x" in a string reflecting the sum of the patterns of the sequence.
The complexity c (n) can be represented by the number of segments into which all strings are divided. Experiments have shown that the complexity c (n) of most (0,1) sequences tends towards a constant value:
Figure BDA0001717396450000161
by normalizing c (n) with b (n) as described above, the complexity formula can be defined as follows:
Ckcas shown in fig. 4(b), the KC value in the fatigue state is approximately 0.2 to 0.3, the KC value in the awake state is approximately 0.3 to 0.5, and the KC value in the concentrated state is approximately 05 to 0.7. When the brain works, the nerve cell activity is relatively in a more orderly state, and with the increase of mental fatigue degree, the control ability of the nerve on the cell activity is weakened, and the nerve cell activityThe excitation degree is increased, the disorder of the nerve cell activity is gradually increased, and the KC value of the nerve cell activity is increased, which shows that the KC value has certain feasibility for grading the fatigue state of the human body.
(3) C0 complexity
C0Complexity defines the proportion of irregular components in a sequence, and the algorithm mainly decomposes the sequence into regular components and irregular components, and performs discrete FFT on a time sequence { x (N) } with a given length of N, wherein N is 0,1,2, …, N-1.
Figure BDA0001717396450000162
Wherein K is 0,1,2, …, N-1, and where { x (N) }, N is 0,1,2, …, N-1, the mean square value is:
Figure BDA0001717396450000163
introducing a parameter r, reserving a frequency spectrum which exceeds the mean square value by r times, and setting the rest part to be zero:
Figure BDA0001717396450000164
the inverse fourier transform of the above equation is:
Figure BDA0001717396450000171
wherein N is 0,1,2, …, N-1, and is defined as C0The complexity is:
Figure BDA0001717396450000172
with increasing parameter r, C0The measurement values gradually increase, which means that as r increases, the less the regular part is removed, the measurement values will increase accordingly, and when r is greater than 2, the measurement values will increase accordinglyThe value is stable, so the value range of the suggested r is 5-10. In view of C0The algorithm has high calculation speed, and the length N of the suggested sequence is greater than 2000; the calculation results are shown in fig. 4 (c). Use of C0The method comprises the steps of calculating the complexity of an electroencephalogram sequence by extracting electroencephalogram characteristic values, analyzing regular signals and irregular signals in the sequence, acquiring C according to the content of the irregular signals calculated, wherein the irregular signals contained in the electroencephalogram are more frequent the brain is, and the C is more frequent the irregular signals are0Value, the more active the brain, C0The larger the value.
S4, optimizing the BP neural network by the artificial fish swarm algorithm to identify fatigue characteristics, wherein the process is as follows:
s41, obtaining the four sub-band energies obtained in S3, ApEn approximate entropy, KC complexity and C0The complexity is used as the input layer data of the neural network, and the fatigue, the wakefulness and the concentration states are represented by (1, 0, 0), (0,1, 0) and (0, 0,1) respectively as the output layer of the neural network;
s42, most neural network training is usually slow, because the weights of the network are updated based on error information, the selection of an initial value is important for network training, and a good initial value can reduce the training time and obtain the parameters of the neural network model more quickly. Here, we choose the BP neural network optimized based on the artificial fish swarm algorithm to train, which is different from the traditional BP algorithm in that the initial value of the network is not randomly selected or 0 is assigned to the initial value, but the algorithm of the artificial fish swarm is used to find out a superior initial value first. The BP neural network prediction uses an artificial fish swarm algorithm to obtain an optimal network initial weight and a threshold, the network predicts function output after training, and a specific flow chart is shown in FIG. 6.
FIG. 5 shows a BP neural network structure diagram for constructing artificial fish swarm algorithm optimization, wherein M neurons are arranged in an input layer, J neurons are arranged in a hidden layer, K neurons are arranged in an output layer, and weight values from the input layer to the hidden layer are represented as wijThe threshold of hidden layer neurons is denoted bjThe weight value from the hidden layer to the output layer is represented as vjkThe threshold of the output layer neurons is denoted as ak. The parameter to be adjusted in the BP neural network is the weight wij、vjkAnd a threshold value bj、akIf the parameters to be adjusted are set as the state of the artificial fish, the artificial fish x can be expressed as a vector with dimensions M × J + J × K + K:
x＝(w11,…,wM1,b1,…,wiJ,…,wMJ,bJ,v11,…,vJ1,a1…,v1k,…,vJK,aK) Wherein w is11,…,wM1Representing the weight of the input layer neuron to the first hidden layer neuron, b1 representing the threshold of the first neuron in the hidden layer, wiJ,…wMJRepresenting the weights of input layer neurons to the J-th hidden layer neuron, bJ representing the threshold of the J-th hidden layer neuron, v11,…,vJ1Representing the weight of the hidden layer neuron to the first output layer neuron, a1 representing the threshold of the first neuron in the output layer, v1k,…,vJKRepresenting weights from hidden layer neuron to Kth output layer neuron, aKThe valve value of the Kth neuron of the output layer is represented, the food concentration FC of the artificial fish is set to be the reciprocal of the total error E of the BP neural network, namely FC is 1/E, the point with the largest food concentration searched by the artificial fish is the point with the smallest error of the BP neural network, and any two artificial fish x are selected1、x2The euclidean distance d between them is expressed as:
Figure BDA0001717396450000181
in the formula x1、x2The elements in the BP neural network are subtracted in a one-to-one correspondence mode according to the dimensionality, the state of the artificial fish x is changed after foraging, clustering and rear-end collision are carried out on the artificial fish x, and meanwhile initial weight and threshold value are adjusted once for the BP neural network;
s43, the number of nodes of the input layer of the neural network is determined by the index number of the learning sample, the number of nodes of the output layer network is determined by the number of the sample results, and the number of the nodes of the two layers is determined according to the specific problem. Selecting 20 neurons of a middle hidden layer, inputting 17-dimensional data of a sample and 1-dimensional network output according to the characteristics of a BP neural network error function and the design experience of predecessors in combination with experiments, wherein the structure of the neural network is set to be 17-20-1; the parameters for the artificial fish optimization algorithm and BP neural network were chosen as shown in tables 2 and 3.
TABLE 2 Artificial Fish optimization Algorithm parameter selection
Figure BDA0001717396450000191
TABLE 3BP neural network parameter selection
Figure BDA0001717396450000192
S44, selecting the number of artificial fishes as 10 in the artificial fish optimization algorithm, the visible area as 0.5, the crowding factor as 0.618, the step length as 0.1, the iteration frequency as 50 and the repeated exploration frequency as 50; selecting a tansig function by the BP neural network activation function 1, selecting a logsig function by the activation function 2, adopting a trandx function by the training function, setting the maximum training times to be 20000, setting the learning rate to be 0.05, and setting the error requirement precision to be 10-4Momentum factor 0.9;
s45, pair of binary functions
Figure BDA0001717396450000193
Fitting is carried out to test the performance of the artificial fish school optimized BP network (x is more than or equal to 10 and y is less than or equal to 10), the performance is shown as a three-dimensional graph of a binary function in FIG. 7, the step length is taken to be 0.1, 40000 data are obtained in total, 35000 training samples are taken, 5000 training samples are taken, the data are trained for 100 times, and the test result is shown as FIG. 8; from the experimental result, the artificial fish swarm-BP algorithm is obviously superior to the BP neural network algorithm no matter the iteration number of the training or the actual test error.
S46, state recognition of electroencephalogram signals, selecting 2100 groups of data from an electroencephalogram database, 600 groups of centralized state data, 600 groups of awake states, 600 groups of fatigue states for network training, and the rest 300 groups for network testing, grouping in a crossed mode, respectively training by using a BP neural network and an artificial fish swarm-BP (AF-BP) neural network, wherein the training results are shown in fig. 9, and the training comparison results are shown in the following table 4.
TABLE 4 network training results
Figure BDA0001717396450000201
The test samples were divided into 5 groups of 60 signals to be tested, which were verified using BP and modified AF-BP networks, respectively. The test results are shown in tables 5 and 6 below.
TABLE 5BP neural network test results
Figure BDA0001717396450000202
TABLE 6 test results of artificial fish-swarm-BP neural network
Figure BDA0001717396450000203
From the above analysis results: the average accuracy of the BP neural network is 84%, the average accuracy of the artificial fish swarm-BP neural network model is 86.3%, the identification of brain wave signals by adopting the artificial fish swarm-BP neural network model is superior to that of the traditional BP neural network, and the accuracy is improved by 2.3%.
S47, recognizing the state of the brain wave signal by using the neural network trained and tested in the steps, and performing auxiliary recognition by using a support vector machine model to monitor the fatigue of underground personnel;
s5, the main controller processes the concentration data of various sensors to judge the overrun state of the gas, fuses the current safety state and the danger level, sends the overrun state and the danger level to an interface displayed by a monitoring host through the existing underground communication node, and carries out triple alarm in three modes of sound, light and vibration on the safety states with different danger levels through an intelligent bracelet worn on the wrist of the miner. Specifically, the method comprises the following steps:
the coal mine safety situation is divided into 4 situation grades of safety, general safety, less safety, unsafe and the like. Similarly, the gas explosion disaster early warning level is also divided into 4 level intervals, which are respectively: safe, generally safe, less safe, and unsafe, corresponding to the fourth-level warning (green), third-level warning (white), second-level warning (yellow), and first-level warning (red), respectively, as shown in fig. 10.
The coal mine disaster situation assessment and early warning result is the overall description of the risk state of the coal mine underground system, and the risk classification is the classification description of the risk level of the safety production system on the basis of a risk acceptance criterion. The assessment set of the risk level has a plurality of expression modes, and the assessment set K { "safe" for evaluating the risk level of the disaster situation of the coal mine is defined according to the characteristics of a coal mine safety production system and a common risk classification standard; "safer"; "less secure"; "unsafe" }, corresponding rank vector 4, 3, 2, 1, see table 7.
TABLE 7 mine safety situation division table
Figure BDA0001717396450000211
Figure BDA0001717396450000221
The wearable intelligent bracelet gives an alarm by sound, light and vibration, wherein the sound of the alarm module is controlled by a buzzer and an I/O port; the light alarm is represented by a green light, a white light, a yellow light and a red light, and four I/O ports are used for controlling; the vibration alarm is realized by adopting a vibration motor, is controlled by an I/O port and receives data of the processor through a Bluetooth module.
This bracelet system arranges alarm module in the left side of control panel wholly, communicates with the treater through bluetooth module, and bracelet box right side charges for miniature removal is precious. The bracelet box is that independently design 3D prints and makes, and bee calling organ and vibrating motor seal at bracelet box inside admittedly, and the display screen exposes in bracelet box openly, in order to ensure the stability of communication, fixes the bluetooth in bracelet box outside, and bracelet box externally mounted watch chain is convenient for personnel and dresses simultaneously.
The following table 8 shows the level alarm states corresponding to the safety situations of various levels, and the four states of the safety situations respectively correspond to the danger flag bits of the four level vectors of the sensor information of the safe, safer, less safe, unsafe environments and the like. The system adopts different alarm operations by judging the state of the danger zone bit.
TABLE 8 safety status & alarms
Figure BDA0001717396450000222
When the safety situation is in a safety state, the grade vector is 4, the corresponding mental state is centralized, at the moment, the light alarm is green, the buzzer does not alarm, and the vibration motor does not alarm; when the safety situation is in a safer state, the grade vector is 3, the state is clear corresponding to the mental state, the light alarm is a white light, the buzzer does not alarm, and the vibration motor does not alarm; when the safety situation is in a less safe state, the grade vector is 2, the light alarm is red light flashing, the buzzer sounds intermittently and long, and the vibrating motor vibrates intermittently and long; when the safety situation is in an unsafe state, the grade vector is 1, the corresponding mental state is fatigue, at the moment, the light alarm is a yellow light flashing, the buzzer sounds suddenly, and the vibration motor vibrates suddenly and shortly.
Because the acquisition and display of the wearable intelligent bracelet are closely related to the environment acquisition parameters, the intelligent bracelet data stream is integrally designed. The whole data processing design flow of the bracelet data stream is shown in fig. 11.
The environmental parameters include temperature, humidity, methane, carbon dioxide, carbon monoxide, etc. These environmental parameters all define a risk level flag, which, together with the mental state, is normalized to four risk levels: safe, safer, less safe and unsafe, and the dangerous zone bits of the safe, safer, less safe and unsafe are defined as 4, 3, 2 and 1. Taking methane as an example, a flow of collecting the hazard zone bits of methane is described, and a specific flow chart is shown in fig. 12.
Where CCH4 is the methane concentration and DCH4Flag is the methane risk Flag. The sensor collects the gas concentration. If the DCH Flag is less than 0.4 percent, the DCH4Flag is equal to 4, otherwise, the judgment is continued; if the value is less than 0.5%, the DCH4Flag is equal to 3, otherwise, the judgment is continued; if the value is less than 0.9%, the DCH4Flag is equal to 2, otherwise the DCH4Flag is equal to 1; and then outputs a methane hazard Flag bit DCH4 Flag.
After the control panel collects the dangerous zone bits of each environmental parameter data and environmental parameters, data are packaged and sent to the intelligent bracelet through the Bluetooth. The intelligent bracelet carries out bubbling sequencing to the danger level, judges the highest danger zone bit, reports to the police according to the order of highest danger level, and one-level warning > second grade warning > tertiary early warning > level four early warning promptly. The hazard flag bit comparison is shown in fig. 13.
The flow chart of the specific sorting is shown in fig. 14.
The first step is as follows: the danger flag bit information transmitted one second is input, i is defined as 0, j is defined as 0, and the temporary intermediate variable temp is defined as 0.
The second step is that: and judging whether i < the total number of elements-1, if so, entering a third part, and if not, storing the final result into a highest risk zone bit and finishing.
The third step: and continuously judging whether j < the total number of elements is-1, if so, entering the fourth step, otherwise, i adds 1 by itself and returns to the second step.
The fourth step: judging whether the jth element is larger than the jth +1 element, if so, assigning the jth +1 element value to temp, assigning the jth element value to the jth +1 element, assigning the temp value to the jth element, then adding 1 to j and returning to the third step. Otherwise, j is added with 1 and returns to the third step.
The highest risk flag D can thus be determined.
After the highest zone bit D is judged on the bracelet, the alarm operation is started. When D is 4 and D is 3, the bracelet display and warning algorithm is as shown in fig. 15.
The first step is as follows: firstly, initializing to obtain a highest risk zone bit, then judging the highest risk zone bit D, if D is not less than 4 and not more than 1, continuously judging D is not less than 2 and finishing the danger alarm operation, and if true, entering the next step.
The second step is that: and judging whether D is equal to 4, if false, entering the next step, if true, turning on a green light, turning off a white yellow red light, displaying a 4-level alarm and environment parameter page1, displaying for 500ms in a delayed mode, displaying an alarm level and a mental parameter page2, and then entering the next step.
The third step: and D is 3, the white light is on, the green-yellow-red light is off, a 3-level alarm and environment parameter page1 is displayed, the time delay is 500ms, an alarm level and spirit parameter page2 is displayed, the time delay is 500ms, and the process is finished.
When D is 2 and D is 1, the bracelet display and alarm algorithm is as shown in fig. 16.
The first step is as follows: firstly, entering initialization, receiving a highest risk zone bit, judging the highest risk zone bit D, if D is more than or equal to 3 or D is less than or equal to 4, judging that the system belongs to a safe working state, entering safety early warning operation, if the D is not less than 3 or D is less than or equal to 4, continuously judging that D is more than or equal to 1 or D is less than or equal to 2, and carrying out danger warning operation.
The second step is that: and (4) entering a danger alarm operation, further judging the highest danger zone bit, if D is not 2, directly entering the next step, and if D is true, entering a second-level alarm operation. Show 2 grades of warning on the bracelet interface, the green white red light goes out, and the yellow light dodges slowly with certain frequency, and the vibrating motor shakes for a long time, and the buzzer sounds for a long time, then further judges which parameter danger is reported to the police. Taking gas alarm as an example, if the gas danger Flag bit DCH4Flag is true 2, the bracelet then displays the gas alarm page, displays the page in a delayed manner for 500ms, and further determines after the display is completed or the condition is false; if the carbon monoxide dangerous flag bit DCOFlag is true, refreshing and displaying a carbon monoxide alarm page, delaying and displaying by a bracelet for 500ms, and further judging after the display is finished or the condition is false; if the fatigue danger Flag bit Dco2Flag is true, refreshing and displaying a fatigue alarm page, delaying to display for 500ms, and further judging after the display is finished or the condition is false; if the temperature danger flag bit Dtempflag is true, refreshing and displaying a temperature alarm page, delaying to display for 500ms, and further judging after the display is finished or the condition is false; and if the carbon dioxide dangerous Flag bit DCO2Flag is true, refreshing and displaying a carbon dioxide alarm page, delaying to display for 500ms, and further judging after the display is finished or the condition is false. Because the interface is limited, only methane, carbon monoxide and fatigue are displayed on the flow chart, and other judgments are finished in the same way after the judgments are finished.
The third step: and D is equal to 1, displaying a level 1 alarm, wherein a green white yellow lamp is turned off, a red lamp flashes, a vibration motor vibrates for a short time, a buzzer sounds for a short time, and then further judging which parameter is to be alarmed. Since the alarm is issued when the danger flag is 2 or 1, it is not only determined whether the danger flag is 1, but also 2, for example, gas. If the gas danger Flag bit DCH4Flag is less than or equal to 2, displaying a gas alarm page, delaying to display for 500ms, and further judging after the display is finished or the condition is false; if the carbon monoxide dangerous flag bit DCOFlag is not more than 2, displaying a carbon monoxide alarm page, delaying to display for 500ms, and further judging after the display is finished or the condition is false; if the fatigue danger Flag Dco2Flag is less than or equal to 2, displaying a fatigue alarm page, delaying to display for 500ms, and further judging after the display is finished or the condition is false; and sequentially judging the temperature danger flag bits DtempFlag and the like. Similarly, only the judgment processes of methane, carbon monoxide and fatigue are shown on the graph, and the other judgment processes of the danger zone bit are the same and are finished.
The technical solution of the present invention is not limited to the limitations of the above specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims (1)

1. The intelligent mining helmet control method based on human body information and environment information fusion is characterized by comprising the following steps:
s1, acquisition of original brain wave signals: acquiring an original brain wave signal of a miner by an electroencephalogram signal acquisition device arranged on the mining helmet, preprocessing the original brain wave signal, obtaining a processed brain wave signal, and sending the processed brain wave signal to an electroencephalogram signal processor arranged on the rear side of a mining helmet body; meanwhile, an infrared composite five-in-one gas sensor and a DHT11 temperature and humidity sensor are adopted to measure the concentrations of carbon dioxide, methane, carbon monoxide and formaldehyde, and gas parameters are transmitted to a main controller;
the electroencephalogram signal acquisition device comprises a first electroencephalogram electrode, a second electroencephalogram electrode, a third electroencephalogram electrode and an electroencephalogram signal processor, wherein the first electroencephalogram electrode is arranged on the inner surface of the mining helmet body and used for acquiring mental states of the forehead part on the right side of the brain, the second electroencephalogram electrode is used for acquiring electric potentials at earlobes and shielding reference signals, the third electroencephalogram electrode is used for shielding artifact signals below the brain, the electroencephalogram signal processor is arranged on the rear side of the mining helmet body and used for preprocessing signals acquired by the first electroencephalogram electrode, the second electroencephalogram electrode and the third electroencephalogram electrode, and the signal output end of the electroencephalogram signal processor is connected with the signal input;
s2, denoising the original brain wave signal, wherein the process is as follows:
s21, selecting a wavelet basis function most similar to the original signal in the electroencephalogram signal processor, determining the decomposition scale of wavelet transformation, and performing wavelet transformation on the signal by using a Mallat tower method to obtain high-frequency coefficient components and low-frequency coefficient components with different decomposition scales;
s22, calculating a wavelet threshold value on the j scale, comparing the wavelet entropies of n subband signals, selecting the wavelet coefficient of the subband with the largest wavelet entropy value, considering that the wavelet coefficient of the subband is caused by noise, calculating the median value of the wavelet coefficient of the subband as the noise variance of the j scale, and thus calculating the wavelet threshold value of the j scale;
s23, because the wavelet coefficient values of the noise on different scales are different, the wavelet coefficient of the noise is smaller and smaller with the increase of the decomposition scale, the wavelet threshold values of different scales are respectively calculated according to S22, and thresholding processing is carried out on the high-frequency coefficient component of each scale to obtain an approximate high-frequency wavelet coefficient;
s24, utilizing the low-frequency coefficient component of the highest wavelet decomposition layer and the approximate high-frequency wavelet coefficient components of different scales after threshold processing to form the coefficient component required by signal reconstruction, and reconstructing according to the reconstruction mode of multi-resolution analysis to obtain a pure brain wave signal;
s3, extracting brain wave features, wherein the process is as follows:
s31, selecting a wavelet basis to perform wavelet decomposition on the brain wave signal: the electroencephalogram signal processor performs wavelet decomposition and reconstruction on electroencephalogram signals acquired by the electroencephalogram signal acquisition device by using a sym5 wavelet basis function;
s32, decomposing brain wave signals by six layers by adopting a sym5 wavelet function to obtain six layers of low-frequency a6, six layers of high-frequency d6, five layers of high-frequency d5 and four layers of high-frequency d4 of the signals respectively; through frequency detection of FFT, six layers of low-frequency a6, six layers of high-frequency d6, five layers of high-frequency d5 and four layers of high-frequency d4 are found to be respectively approximate to the frequencies of delta waves, theta waves, alpha waves and beta waves, so that signals of wavelet decomposition are adopted to represent the delta waves, the theta waves, the alpha waves and the beta waves; obtaining F (k) by fast Fourier transform of brain waves f (n) in frontal lobe area;
Figure FDA0002848911230000021
wherein, f (N) is the frontal lobe area brain wave discrete signal collected by the brain wave collecting device, N is the serial number of the sampling points, N is the total number of the sampling points, and k is an integer;
s33, carrying out wavelet decomposition on the original data to obtain delta (x), theta (x), alpha (x) and beta (x) sub-band signals of different frequency bands, wherein the sub-band energy is as follows:
Figure FDA0002848911230000031
Figure FDA0002848911230000032
calculate the energy ratio of each signal: eall＝E(δ)+E(θ)+E(α)+E(β)，En(δ)＝E(δ)/Eall，En(θ)＝E(θ)/Eall，En(α)＝E(α)/Eall，En(β)＝E(β)/EallTherefore, the ratio of sub-signals of EEG of miners in a fatigue state, a waking state and a concentration state can be quantitatively analyzed;
s34, calculating ApEn approximate entropy, KC complexity and C0The complexity is compared with the change conditions of the three characteristic values of the brain wave signals in a fatigue state, a waking state and a concentration state, and a plurality of groups of fatigue characteristic values, a plurality of groups of waking characteristic values and a plurality of groups of concentration characteristic values are respectively listed and stored and are subsequently used as input data of a neural network;
s4, optimizing the BP neural network by the artificial fish swarm algorithm to identify fatigue characteristics, wherein the process is as follows:
s41, obtaining the four sub-band energies obtained in S3, ApEn approximate entropy, KC complexity and C0The complexity is used as the input layer data of the neural network, and the fatigue, the wakefulness and the concentration states are represented by (1, 0, 0), (0,1, 0) and (0, 0,1) respectively as the output layer of the neural network;
s42, setting M neurons in the input layer, J neurons in the hidden layer and K neurons in the output layer in the BP neural network, and expressing the weight value from the input layer to the hidden layer as wijThe threshold of hidden layer neurons is denoted bjThe weight value from the hidden layer to the output layer is represented as vjkThe threshold of the output layer neurons is denoted as ak(ii) a The parameter to be adjusted in the BP neural network is the weight wij、vjkAnd a threshold value bj、akIf the parameters to be adjusted are set as the state of the artificial fish, the artificial fish x can be expressed as a vector with dimensions M × J + J × K + K:
x＝(w11,…,wM1,b1,…,wiJ,…,wMJ,bJ,v11,…,vJ1,a1…,v1k,…,vJK,aK) Wherein w is11,…,wM1Representing the weight of the input layer neuron to the first hidden layer neuron, b1 representing the threshold of the first neuron in the hidden layer, wiJ,…wMJRepresenting the weights of input layer neurons to the J-th hidden layer neuron, bJ representing the threshold of the J-th hidden layer neuron, v11,…,vJ1Representing the weight of the hidden layer neuron to the first output layer neuron, a1 representing the threshold of the first neuron in the output layer, v1k,…,vJKRepresenting weights from hidden layer neuron to Kth output layer neuron, aKThe valve value of the Kth neuron of the output layer is represented, the food concentration FC of the artificial fish is set to be the reciprocal of the total error E of the BP neural network, namely FC is 1/E, the point with the largest food concentration searched by the artificial fish is the point with the smallest error of the BP neural network, and any two artificial fish x are selected1、x2The euclidean distance d between them is expressed as:
Figure FDA0002848911230000041
in the formula x1、x2The elements in the artificial fish x are subtracted in a one-to-one correspondence manner strictly according to the dimension, the state of the artificial fish x is changed after foraging, clustering and rear-end collision are carried out on the artificial fish x, and meanwhile, the initial weight and the threshold value of the BP neural network are adjusted once;
s43, selecting 20 neurons of the middle hidden layer, 17-dimensional data of sample input and 1-dimensional network output according to the characteristics of the BP neural network error function and the design experience of predecessors by combining experiments, wherein the network output is 1-dimensional, namely the structure of the neural network is set to be 17-20-1;
s44, selecting the number of artificial fishes as 10 in the artificial fish optimization algorithm, the visible area as 0.5, the crowding factor as 0.618, the step length as 0.1, the iteration frequency as 50 and the repeated exploration frequency as 50; selecting a tansig function by the BP neural network activation function 1, selecting a logsig function by the activation function 2, adopting a trandx function by the training function, setting the maximum training times to be 20000, setting the learning rate to be 0.05, and setting the error requirement precision to be 10-4Momentum factor 0.9;
s45, pair of binary functions
Figure FDA0002848911230000051
Fitting to test the performance of the artificial fish swarm optimized BP network: taking 40000 data with the step length of 0.1, wherein 35000 data are taken as training samples, 5000 data are taken as testing samples, and the data are trained for 100 times;
s46, selecting 2100 groups of data, 600 groups of centralized state data, 600 groups of awake states and 600 groups of fatigue states from the electroencephalogram database, wherein 1800 groups are used for network training, the other 300 groups are used for network testing, and grouping is performed in a crossed mode; recognizing the state of the brain wave signal by utilizing the neural network trained and tested in the steps to realize the monitoring of the fatigue of underground personnel;
s5, the main controller processes the concentration data of various sensors to judge the overrun state of the gas, fuses the current safety state and the danger level, sends the data to the interface displayed by the monitoring host through the existing underground communication node, and carries out triple alarm in three modes of sound, light and vibration on the safety state of different danger levels through the intelligent bracelet which is connected with the main controller through the Bluetooth module and is worn on the wrist of the miner.
CN201810715885.1A 2018-07-03 2018-07-03 Intelligent mining helmet control method based on human body information and environment information fusion Active CN108903936B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810715885.1A CN108903936B (en) 2018-07-03 2018-07-03 Intelligent mining helmet control method based on human body information and environment information fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810715885.1A CN108903936B (en) 2018-07-03 2018-07-03 Intelligent mining helmet control method based on human body information and environment information fusion

Publications (2)

Publication Number Publication Date
CN108903936A CN108903936A (en) 2018-11-30
CN108903936B true CN108903936B (en) 2021-01-26

Family

ID=64425296

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810715885.1A Active CN108903936B (en) 2018-07-03 2018-07-03 Intelligent mining helmet control method based on human body information and environment information fusion

Country Status (1)

Country Link
CN (1) CN108903936B (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109784361A (en) * 2018-12-05 2019-05-21 鲁东大学 Seashell products classifying identification method and device
CN109614932B (en) * 2018-12-12 2022-03-25 西安科技大学 Environment identification method based on electroencephalogram field change mechanism, mining helmet and cloud platform
CN109884465B (en) * 2019-03-01 2023-09-29 辽宁工业大学 Unidirectional ground fault positioning method based on signal injection method
FR3103373A1 (en) * 2019-11-27 2021-05-28 Thales ENVIRONMENTAL MONITORING ASSISTANCE SYSTEM AND ASSOCIATED PROCESS
CN110840452B (en) * 2019-12-10 2024-08-27 广西师范大学 Brain wave signal filtering device and method
CN112336355A (en) * 2020-11-06 2021-02-09 广东电网有限责任公司电力科学研究院 Safety supervision system, device and equipment based on electroencephalogram signal operating personnel
CN112842359B (en) * 2021-01-25 2022-09-16 国网江苏省电力有限公司电力科学研究院 Pressure and fatigue information monitoring method for intelligent safety helmet
CN113017184A (en) * 2021-03-29 2021-06-25 广东电网有限责任公司 Intelligent safety helmet and intelligent monitoring system
CN113506246B (en) * 2021-06-15 2022-11-25 西安建筑科技大学 Concrete 3D printing component fine detection method based on machine vision
CN114983080A (en) * 2022-05-23 2022-09-02 北京市政建设集团有限责任公司 Intelligent safety helmet
CN117649197B (en) * 2023-11-29 2024-08-20 北京希安科电子系统工程有限责任公司 Visual intelligent monitoring data security control substation system and process thereof

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030055355A1 (en) * 2000-10-16 2003-03-20 Viertio-Oja Hanna Elina Method and apparatus for determining the cerebral state of a patient with fast response
CN103989485A (en) * 2014-05-07 2014-08-20 朱晓斐 Human body fatigue evaluation method based on brain waves
WO2014167460A1 (en) * 2013-04-12 2014-10-16 Commissariat A L'energie Atomique Et Aux Energies Alternatives Methods to monitor consciousness
US20160361041A1 (en) * 2015-06-15 2016-12-15 The Research Foundation For The State University Of New York System and method for infrasonic cardiac monitoring
CN106388780A (en) * 2016-09-21 2017-02-15 广州视源电子科技股份有限公司 Sleep state detection method and system based on fusion of two classifiers and detector
CN108229401A (en) * 2018-01-05 2018-06-29 杭州电子科技大学 A kind of multi-modal Modulation recognition method based on AFSA-SVM

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030055355A1 (en) * 2000-10-16 2003-03-20 Viertio-Oja Hanna Elina Method and apparatus for determining the cerebral state of a patient with fast response
WO2014167460A1 (en) * 2013-04-12 2014-10-16 Commissariat A L'energie Atomique Et Aux Energies Alternatives Methods to monitor consciousness
CN103989485A (en) * 2014-05-07 2014-08-20 朱晓斐 Human body fatigue evaluation method based on brain waves
US20160361041A1 (en) * 2015-06-15 2016-12-15 The Research Foundation For The State University Of New York System and method for infrasonic cardiac monitoring
CN106388780A (en) * 2016-09-21 2017-02-15 广州视源电子科技股份有限公司 Sleep state detection method and system based on fusion of two classifiers and detector
CN108229401A (en) * 2018-01-05 2018-06-29 杭州电子科技大学 A kind of multi-modal Modulation recognition method based on AFSA-SVM

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Anxiety Level Detection Using BCI of Miner’s Smart Helmet;Mei Wang等;《Mobile Netw Appl》;20170922;全文 *
Mobile Healthcare System for Driver Based on Drowsy Detection Using EEG Signal Analysis;Mu Zhendong 等;《Metallurgical and Mining Industry》;20151231(第7期);第266-273页 *
一种基于智能手机与脑－ 机接口技术的环境控制系统;王湃 等;《西安科技大学学报》;20150930;第35卷(第5期);第656-661页 *
脑电近似熵分析的思维分类识别;江朝晖 等;《应用科学学报》;20060731;第24卷(第4期);第359-362页 *

Also Published As

Publication number Publication date
CN108903936A (en) 2018-11-30

Similar Documents

Publication Publication Date Title
CN108903936B (en) Intelligent mining helmet control method based on human body information and environment information fusion
Al Nahian et al. Towards an accelerometer-based elderly fall detection system using cross-disciplinary time series features
CN110133610B (en) Ultra-wideband radar action identification method based on time-varying distance-Doppler diagram
CN110367967B (en) Portable lightweight human brain state detection method based on data fusion
CN109124625B (en) Driver fatigue state level grading method
CN109498041A (en) Driver road anger state identification method based on brain electricity and pulse information
Peng et al. 3D-STCNN: Spatiotemporal Convolutional Neural Network based on EEG 3D features for detecting driving fatigue
CN113743471B (en) Driving evaluation method and system
CN109568760A (en) Sleep environment adjusting method and system
CN104586387A (en) Method for extracting and fusing time, frequency and space domain multi-parameter electroencephalogram characters
CN105411565A (en) Heart rate variability feature classification method based on generalized scale wavelet entropy
CN105320969A (en) A heart rate variability feature classification method based on multi-scale Renyi entropy
CN112800928B (en) Epileptic seizure prediction method of global self-attention residual error network integrating channel and spectrum characteristics
CN104887222A (en) Reversible electroencephalogram analysis method
CN102133100A (en) Sparse representation-based electroencephalogram signal detection method
Ding et al. Cascaded convolutional neural network with attention mechanism for mobile EEG-based driver drowsiness detection system
CN109624984A (en) A kind of mobile Internet intelligent vehicle-mounted system
CN109875584A (en) The detection method and its warning system of physiological driver&#39;s fatigue
CN113076878A (en) Physique identification method based on attention mechanism convolution network structure
CN110916672A (en) Old people daily activity monitoring method based on one-dimensional convolutional neural network
CN113397482B (en) Human behavior analysis method and system
CN114343638A (en) Fatigue degree evaluation method and system based on multi-modal physiological parameter signals
Zhang et al. A novel multidimensional feature extraction method based on vmd and wpd for emotion recognition
CN112617761A (en) Sleep stage staging method for self-adaptive multipoint generation
Chen et al. A deep learning approach for sleep-wake detection from hrv and accelerometer data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant