CN109344195B - HMM model-based pipeline security event recognition and knowledge mining method - Google Patents
HMM model-based pipeline security event recognition and knowledge mining method Download PDFInfo
- Publication number
- CN109344195B CN109344195B CN201811251182.4A CN201811251182A CN109344195B CN 109344195 B CN109344195 B CN 109344195B CN 201811251182 A CN201811251182 A CN 201811251182A CN 109344195 B CN109344195 B CN 109344195B
- Authority
- CN
- China
- Prior art keywords
- event
- feature vector
- vector sequence
- hmm model
- sequence
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Abstract
The invention discloses a method for identifying and mining knowledge of a pipeline safety event based on an HMM (hidden Markov model), belonging to the field of monitoring of pipeline safety events; the method comprises the following steps of 1: extracting multi-domain features of signals acquired by each space point to obtain a feature vector sequence of the signals; step 2: inputting the feature vector sequence into an HMM model for off-line training to complete the establishment of a typical event HMM model library; and step 3: after a signal to be recognized is subjected to current feature vector sequence acquisition in the step 1, inputting the current feature vector sequence into a typical event HMM model base for recognition and outputting an event judgment type, and meanwhile, calculating an optimal hidden state sequence as event state sequence evolution process information for outputting to complete knowledge mining; the HMM model is analyzed and recognized based on the characteristic time sequence, so that the event recognition rate is effectively improved; meanwhile, knowledge mining is realized in the evolution process of the event state sequence, so that short-term prediction can be carried out.
Description
Technical Field
The invention belongs to the field of pipeline safety event monitoring, and particularly relates to a HMM model-based pipeline safety event identification and knowledge mining method.
Background
The urban underground pipe network such as water pipes, air pipes, thermal pipelines, oil pipes and the like are urban blood vessels and life lines, the safe operation of the urban underground pipe network is an important guarantee for the safety of lives and properties of people, and the leakage accidents of the pipelines are caused by frequent artificial theft, mechanical construction and other external force damages, so that the safety of the pipelines is threatened, and the immeasurable economic loss and the life and property safety of people are caused. The real-time online monitoring is carried out on the underground pipe network by utilizing the internet of things technology, potential safety hazards such as external force damage and the like are detected and alarmed in time, and the prevention is in the bud, so that the method is an important problem to be solved urgently in the current pipe network safety management.
The phase-sensitive optical time domain reflectometry (phi-OTDR) technology is taken as a representative of a distributed optical fiber sensing technology, and signals such as vibration signals caused by leaked sound waves or other events in the environment along the pipeline are sensed by optical fibers, so that safety events along the pipeline are analyzed in real time, safety early warning is carried out, and the phi-OTDR technology plays an important role in pipeline safety monitoring. The technology has the characteristics of being passive, high in sensing sensitivity, strong in multipoint positioning capability and the like, and has no functional device in the optical fiber and long service life; the single-end detection is adopted, the engineering construction and the maintenance are simple and convenient, and the large-scale safety monitoring of the long-distance pipeline is convenient to realize.
Hidden Markov model HMM (hidden Markov model) is a statistical model used for describing a Markov process containing hidden unknown parameters, can carry out statistical modeling and classification on a state or information evolution process, and is widely applied to the fields of speech recognition, mechanical equipment state monitoring and fault diagnosis.
However, in the practical application of the phase-sensitive optical time domain reflectometry (phi-OTDR), the pipeline laying geographic environment is complex, and the identification of real threatening events is difficult due to the fact that the influence of various time-varying interferences is monitored in a large range; at present, the related feature extraction and event identification methods in the field all process some section of signal extraction static attribute features, ignore the time sequence relation of the event dynamic evolution process, i.e. ignore the context relation of time and space, therefore, only the feature attributes of the signal in the current time period can be summarized and classified on the basis of the summary, and the dynamic change and evolution process of the event state cannot be revealed or the judgment and identification can be performed according to the development process of the physical event, so that there is an urgent need for an event identification method which can solve the problems that the static feature analysis method cannot obtain the event dynamic time sequence change rule, so that the pipeline security event identification rate is low, and the occurring event cannot be predicted.
Disclosure of Invention
The invention aims to: the invention provides a method for identifying and mining knowledge of a pipeline safety event based on an HMM (hidden Markov model), which solves the problems that the existing static feature analysis-based method cannot acquire the dynamic time sequence change rule of the event, so that the identification rate of the pipeline safety event is low, and the happening event cannot be predicted.
The technical scheme adopted by the invention is as follows:
the HMM model-based pipeline security event recognition and knowledge mining method comprises the following steps:
step 1: extracting multi-domain features of signals acquired by each space point to obtain a feature vector sequence of the signals;
step 2: inputting the feature vector sequence into an HMM model for off-line training to complete the establishment of a typical event HMM model library;
and step 3: after the current feature vector and the feature vector sequence of the signal to be recognized are obtained in the step 1, the current feature vector and the feature vector sequence are input into a typical event HMM model base for recognition and output of an event judgment type, and meanwhile, an optimal hidden state sequence is calculated and output as event state sequence evolution process information, so that knowledge mining is completed.
Preferably, the step 1 comprises the following steps:
step 1.1: respectively constructing a typical event database of each space point based on typical physical event samples collected by each space point, and specifically comprising the following steps:
step 1.1.1: event initialization: defining the event type of each space point and the corresponding event label;
step 1.1.2: constructing sample data: the method comprises the steps that a set short-time signal unit SU accumulates signals collected in L short-time signal units SU as a long-time signal, and the long-time signal is used as sample data of each space point event;
step 1.1.3: constructing a typical event database: repeating the step 1.1.2 according to the duration to obtain all sample data of the events of each space point, and completing the construction of a typical event database of each space point;
step 1.2: performing multi-domain feature extraction on each short-time signal unit in a signal sample of a typical event database to obtain a feature vector of each short-time signal unit, combining feature vector sequences of various events according to a time sequence relation, and using the feature vector sequences as a feature vector sequence set for training an HMM model, wherein the feature vector sequence set of the HMM model is as follows:
O'＝[O(1),O(2),Λ,O(k),Λ,O(K)] (2)
the dimension of O' is D multiplied by L multiplied by K, D represents the dimension of the feature vector extracted by each short-time signal unit, L represents the length of the feature vector sequence and corresponds to the number of short-time signal units contained in the original-time signal, and K represents the number of samples of the feature vector sequence;
the k-th set of feature vector sequences is as follows:
O(k)＝[o1 (k),o2 (k),Λ,ot (k),Λ,oL (k)] (3)
wherein the content of the first and second substances,
and the characteristic vector of the t-th short-time signal unit in the characteristic vector sequence set is represented.
Preferably, the step 2 comprises the following steps:
step 2.1: initializing parameters of the HMM model based on the feature vector sequence set;
step 2.2: based on the step 2.1, utilizing a feature vector sequence set, and based on a Baum-Welch algorithm, iteratively updating each parameter of the HMM model;
step 2.3: judging whether the joint probability error of the iteration process is smaller than the set convergence error or not, if so, stopping the iteration, and finishing HMM model training and building a typical event HMM model library; if not, jumping to step 2.2 to continue iteration.
Preferably, the step 3 includes the following steps:
step 3.1: after the current feature vector sequence of the signal to be recognized is obtained in the step 1, inputting the current feature vector sequence into a typical event HMM model library, calculating the Bayesian posterior probability of the current feature vector sequence under each model in the HMM model library through a Viterbi algorithm, outputting an event label corresponding to the model with the maximum probability as the event type of the signal to be recognized, and completing the recognition and classification of the signal;
step 3.2: and inputting the current feature vector sequence into a typical event HMM model base, calculating a hidden state sequence corresponding to the current feature vector sequence through a Viterbi algorithm, outputting the hidden state sequence serving as event state sequence evolution process information, and finishing knowledge mining.
Preferably, the step 3.1 specifically comprises the following steps:
step 3.1.1: initialization:
step 3.1.2: recursion:
wherein, deltat(i)＝maxP(q1,Λ,qt,qt＝θi,o1,Λ,otλ) represents a path q along time t1,q2,Λ,qtAnd q ist＝θiI.e. the state is theta at time tiIs generated o1,o2,Λ,otThe maximum probability of (d);
indicates a state of theta at time tiAll paths q of1,q2,Λ,qtThe t-1 node of the path with the highest probability;
step 3.1.3: and (4) terminating:
wherein, P*Representing the maximum output probability of the current feature vector sequence under the given model parameters;
step 3.1.4: according to each HMM model parameter lambda in the model libraryiAnd current feature vector sequence OiCalculating the prediction probability P of the feature vector sequence under each modeli *：
Pi *＝P(Oi|λi)
Selecting the model event label corresponding to the maximum prediction probability as the current feature vector sequence OiThe identification and classification of the signals is realized.
Preferably, the multi-domain features employ 44-dimensional features of analysis domains such as time domain, frequency domain, inverse frequency domain, and model coefficients.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. the HMM model of the invention models the time sequence of the event evolution process, and the event recognition rate is greatly improved; the method comprises the steps of extracting and mining a time sequence rule corresponding to the state evolution of an event, realizing short-term prediction of the development process of the occurring dynamic event, and solving the problems that the existing static feature analysis-based method cannot obtain the dynamic time sequence change rule of the event, so that the recognition rate of the pipeline safety event is low, and the occurring event cannot be predicted;
2. the HMM model utilizes the dual information of the evolution relation between the signal feature vector and the feature vector time sequence, and has higher event recognition rate compared with the traditional recognition algorithm which only classifies the static features or feature vectors extracted according to the signal in a certain time period; the problem that the pipeline safety event identification rate is low due to the fact that an existing static characteristic analysis-based method cannot obtain an event dynamic time sequence change rule is solved, and accurate classification of various physical events of pipeline safety monitoring in a large-range dynamic environment is achieved;
3. the method comprises the steps that when typical physical events are identified and classified, development processes of various physical events near a pipeline are described by using hidden state sequences of an HMM model, namely after an HMM model base of various physical events near the pipeline is built, when certain events occur, the collected long-term signals can obtain the hidden state sequences under the HMM model of the events in the HMM model base and serve as event state sequence evolution process information to be output, and state change rules of the events are represented; according to the optimal hidden state sequence of different physical event HMM models, namely the time sequence rule information of event state evolution, the time sequence rule corresponding to the event state evolution can be extracted and mined, and the ongoing dynamic event development process can be predicted in a short term;
4. the HMM model has higher training convergence speed than other models, and when the HMM models of different physical events are trained, the iteration times of model convergence are far less than those of other models, so that the event recognition efficiency is improved;
5. the invention extracts 44-dimensional characteristics of the signal from analysis domains such as time domain, frequency domain, inverse frequency domain, model coefficient and the like in an all-around way, is beneficial to analyzing events comprehensively, and further improves the event recognition rate.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a block diagram of an application system of the present invention;
FIG. 2 is a block flow diagram of the method of the present invention;
FIG. 3 is a table of training/testing set information for HMM models and non-timing models in accordance with the present invention;
FIG. 4 is a table of analysis domain feature extraction according to the present invention;
FIG. 5 is a graph illustrating the evaluation of the clustering effect of each event type according to the present invention;
FIG. 6 is a flowchart of model parameter iterative update based on Baum-Welch algorithm of the present invention;
FIG. 7 is an iterative graph of various event HMM model training processes of the present invention;
FIG. 8 is a graph illustrating the effect of the HMM event recognition rate;
FIG. 9 is a graph comparing the recognition rates of different models with different feature sets according to the present invention;
FIG. 10 is a state diagram of the average evolution process of five typical events according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
It is noted that relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The technical problem is as follows: the method solves the problems that the existing static characteristic analysis-based method cannot acquire the dynamic time sequence change rule of the event, so that the pipeline safety event recognition rate is low and the happening event cannot be predicted;
the technical means is as follows: the HMM model-based pipeline security event recognition and knowledge mining method comprises the following steps:
step 1: extracting multi-domain features of signals acquired by each space point to obtain a feature vector sequence of the signals;
step 2: inputting the feature vector sequence into an HMM model for off-line training to complete the establishment of a typical event HMM model library;
and step 3: after the current feature vector and the feature vector sequence of the signal to be recognized are obtained in the step 1, the current feature vector and the feature vector sequence are input into a typical event HMM model base for recognition and output of an event judgment type, and meanwhile, an optimal hidden state sequence is calculated and output as event state sequence evolution process information, so that knowledge mining is completed.
The step 1 comprises the following steps:
step 1.1: respectively constructing a typical event database of each space point based on typical physical event samples collected by each space point, and specifically comprising the following steps:
step 1.1.1: event initialization: defining the event type of each space point and the corresponding event label;
step 1.1.2: constructing sample data: the method comprises the steps that a set short-time signal unit SU accumulates signals collected in L short-time signal units SU as a long-time signal, and the long-time signal is used as sample data of each space point event;
step 1.1.3: constructing a typical event database: repeating the step 1.1.2 according to the duration to obtain all sample data of the events of each space point, and completing the construction of a typical event database of each space point;
step 1.2: performing multi-domain feature extraction on each short-time signal unit in a signal sample of a typical event database to obtain a feature vector of each short-time signal unit, combining feature vector sequences of various events according to a time sequence relation, and using the feature vector sequences as a feature vector sequence set for training an HMM model, wherein the feature vector sequence set of the HMM model is as follows:
O'＝[O(1),O(2),Λ,O(k),Λ,O(K)] (2)
the dimension of O' is D multiplied by L multiplied by K, D represents the dimension of the feature vector extracted by each short-time signal unit, L represents the length of the feature vector sequence and corresponds to the number of short-time signal units contained in the original-time signal, and K represents the number of samples of the feature vector sequence;
the k-th set of feature vector sequences is as follows:
O(k)＝[o1 (k),o2 (k),Λ,ot (k),Λ,oL (k)] (3)
wherein the content of the first and second substances,
and the characteristic vector of the t-th short-time signal unit in the characteristic vector sequence set is represented.
The step 2 comprises the following steps:
step 2.1: initializing parameters of the HMM model based on the feature vector sequence set;
step 2.2: based on the step 2.1, utilizing a feature vector sequence set, and based on a Baum-Welch algorithm, iteratively updating each parameter of the HMM model;
step 2.3: judging whether the joint probability error of the iteration process is smaller than the set convergence error or not, if so, stopping the iteration, and finishing HMM model training and building a typical event HMM model library; if not, jumping to step 2.2 to continue iteration.
The step 3 comprises the following steps:
step 3.1: after the current feature vector sequence of the signal to be recognized is obtained in the step 1, inputting the current feature vector sequence into a typical event HMM model library, calculating the Bayesian posterior probability of the current feature vector sequence under each model in the HMM model library through a Viterbi algorithm, outputting an event label corresponding to the model with the maximum probability as the event type of the signal to be recognized, and completing the recognition and classification of the signal;
step 3.2: and inputting the current feature vector sequence into a typical event HMM model base, calculating a hidden state sequence corresponding to the current feature vector sequence through a Viterbi algorithm, outputting the hidden state sequence serving as event state sequence evolution process information, and finishing knowledge mining.
Step 3.1 specifically comprises the following steps:
step 3.1.1: initialization:
step 3.1.2: recursion:
wherein, deltat(i)＝maxP(q1,Λ,qt,qt＝θi,o1,Λ,otλ) represents a path q along time t1,q2,Λ,qtAnd q ist＝θiI.e. the state is theta at time tiIs generated o1,o2,Λ,otThe maximum probability of (d);
indicates a state of theta at time tiAll paths q of1,q2,Λ,qtThe t-1 node of the path with the highest probability;
step 3.1.3: and (4) terminating:
wherein, P*Representing the maximum output probability of the current feature vector sequence under the given model parameters;
step 3.1.4: according to each HMM model parameter lambda in the model libraryiAnd current feature vector sequence OiCalculating the prediction probability P of the feature vector sequence under each modeli *：
Pi *＝P(Oi|λi)
Selecting the model event label corresponding to the maximum prediction probability as the current feature vector sequence OiThe identification and classification of the signals is realized.
The multi-domain features adopt 44-dimensional features of analysis domains such as time domain, frequency domain, inverse frequency domain, model coefficients and the like.
The technical effects are as follows: the invention can realize the accurate classification of various physical events of the pipeline safety monitoring in a large-scale dynamic environment; the HMM model is used for modeling a time sequence of an event evolution process, and the model utilizes dual information of a signal feature vector and a feature vector time sequence evolution relation, so that compared with a traditional recognition algorithm for classifying only according to static features or feature vectors extracted from signals in a certain time period, the HMM model has higher event recognition rate; while identifying and classifying typical physical events, the time sequence rule corresponding to the evolution of the event state can be extracted and mined: describing the development process of various physical events near a pipeline by using a hidden state sequence of a hidden Markov model, and outputting an optimal hidden state sequence as a time sequence rule of event state evolution when the HMM model of a certain type of event is trained, wherein the optimal hidden state sequence represents the state change rule of the type of event; according to the optimal hidden state sequence of different physical event HMM models, namely the time sequence rule information of event state evolution, the short-term prediction of the developing process of the occurring dynamic event can be further realized.
The features and properties of the present invention are described in further detail below with reference to examples.
Example 1
As shown in fig. 1-10, the distributed optical fiber acoustic wave/vibration sensing system (DVS/DAS) based on linear phase demodulation of the present invention is used to implement long-distance pipeline safety monitoring, and the structure of the application system and its working principle are shown in fig. 1. The system hardware comprises a detection optical cable, an optical signal demodulation device and a signal processing host; the detection optical cable is usually laid along underground pipelines, transmission cables and urban roads by adopting common single-mode communication optical fibers, and can also directly utilize the spare fiber cores of the communication optical cables laid along the pipelines or the roads. The optical signal demodulation equipment comprises two types of optical devices and electrical devices, wherein a continuous coherent optical signal is generated by an ultra-narrow line width laser and is modulated into an optical pulse signal by an acousto-optic or electro-optic modulator, the optical pulse signal is intensively amplified by an erbium-doped fiber amplifier EDFA, and the amplified optical pulse signal is injected into a detection optical cable through a port 1 and a port 2 of an isolator and a circulator in sequence; the optical pulse signal generates Rayleigh scattering in the transmission process of the optical cable, then returns to Rayleigh scattering optical signals along the optical cable, is received by a port 2 and a port 3 of the circulator, is coupled by a first coupler after being filtered by an optical filter, and then is injected into an unbalanced Mach-Zehnder or Michelson interferometer and the like, and is determined by a demodulation method, three paths of phase change information introduced by external disturbance with the phase difference of 120 degrees are output by a second coupler of 3 x 3, so that an action signal of sound waves/vibration on the optical fiber can be obtained, the demodulated optical signal is converted into an electric signal by a photoelectric detector, and then is synchronously acquired by a synchronous trigger analog-to-digital converter controlled by a waveform generation card, and finally, the digital electric signal is transmitted to a signal processing host machine in real time through interfaces such as a network and the like. The signal processing host is a common computer host PC or an FPGA/DSP embedded mainboard and is used for analyzing and processing optical fiber detection signals, intelligently analyzing, processing and identifying and classifying sensed sound waves and vibration signals through a specific signal processing algorithm, and determining the position of the signals according to an optical time domain reflection principle.
The system returns an original signal track distributed along the space at each moment, continuously accumulates T original signal tracks on a time axis, and constructs a time-space signal matrix with a time dimension of T and a space dimension of S:
{XX＝xts(t＝1，2，Λ，T；s＝1，2，Λ，S)} (1)
wherein T is the time sampling length, S is the space sampling length, the unit is the sampling point, xtsThe amplitude of the acquired signal at a certain spatial point at a certain time. Acquiring a time-space response signal of the distributed optical fiber sound/vibration sensing system through time accumulation, wherein the abscissa is a space axis and represents a data acquisition space point along the pipeline, and the distance between the two space points is delta S; the ordinate is a time axis, and the sampling interval delta T of two sampling points is 1/fs，fsThe pulse trigger frequency on the time axis, i.e. the temporal sampling frequency. The signal processing flow, i.e. event recognition and knowledge mining, is shown in fig. 2:
step S1: respectively constructing a typical event database based on signals acquired by each space point, and converting the typical event database into a feature vector sequence required by training an HMM (hidden Markov model) through multi-domain feature extraction;
step S2: training a typical event HMM model by using the feature vector sequence off-line to establish a typical event HMM model library;
step S3: during online identification, calculating output probabilities obtained by inputting test signals into various typical event models according to a model matching thought, and taking event labels corresponding to the models with the highest probabilities as event judgment types of the test signals; and outputting the optimal hidden state sequence of the event model as the evolution process and the time sequence rule information of the event to complete knowledge mining.
The HMM model of the invention models the time sequence of the event evolution process, and has higher event recognition rate; the method comprises the steps of extracting and mining a time sequence rule corresponding to the state evolution of an event, realizing short-term prediction of the development process of the happening dynamic event, and solving the problems that the identification rate of the pipeline safety event is low and the happening event cannot be predicted due to the fact that the dynamic time sequence change rule of the event cannot be obtained based on a static characteristic analysis method.
Example 2
Based on embodiment 1, the step S1 includes the following steps:
respectively constructing a typical event database of each space point based on typical physical event samples collected by each space point, and specifically comprising the following steps:
in this embodiment, typical events in the pipeline safety monitoring include: background noise, manual mining, mechanical mining, traffic interference and factory interference which are easy to misjudge, and the event labels are set to be 1, 2, 3, 4 and 5 in sequence. Dividing time signals of different types of events continuously acquired based on each space point into a short-time signal unit SU according to a fixed time length, wherein the length of the short-time signal is set according to practical application, and the embodiment is set to be 1 s; taking the short-time signal as a basic unit, continuously accumulating L short-time signal units from the initial time of the occurrence of a certain event to form a long-time signal, and taking the long-time signal as sample data of the event; based on signal samples collected by each spatial point in pipeline safety monitoring, a typical event database of each spatial point is constructed according to the method, and specific data is shown in fig. 3, wherein fig. 3 contains training/testing set information of an HMM model and training/testing set information of a non-time sequence model.
Performing multi-domain feature extraction on all short-term signal units of each signal sample in a typical event database to obtain feature vectors of each short-term signal unit, and sorting the feature vectors into feature vector sequences of various events according to a time sequence relation, wherein the feature vector sequences are used as a feature vector sequence set for training an HMM model, and the method specifically comprises the following steps:
extracting all-dimensional characteristics of signals from a plurality of analysis domains such as a time domain, a frequency domain, an inverse frequency domain, a model coefficient and the like by each short-time signal unit in the signal sample, comprehensively analyzing events and being beneficial to improving the event recognition rate; as shown in fig. 4, based on the above four analysis domains, 44 features are extracted in total, the feature dimension D is 44, and one long-term signal with a length of L seconds includes L groups of short-term signal units, so that the event sample data set X of K groups of signals is converted into an HMM model feature vector sequence set for input after feature extraction, and the HMM model feature vector sequence set is as follows:
O'＝[O(1),O(2),Λ,O(k),Λ,O(K)] (2)
the dimension of O' is DxLxK, D is the dimension of a feature vector extracted by each short-time signal unit, L is the length of a feature vector sequence and corresponds to the number of short-time signal units contained in an original-time signal, and K is the number of samples of the feature vector sequence;
the k-th set of feature vector sequences is as follows:
O(k)＝[o1 (k),o2 (k),Λ,ot (k),Λ,oL (k)] (3)
wherein the content of the first and second substances,
and the characteristic vector of the t-th short-time signal unit in the characteristic vector sequence set is represented.
During off-line training, a training signal sample set is converted into a training set of a feature vector sequence through the processes and used as an HMM model observation sequence input for off-line training of model parameters; during on-line testing, the test signal sample set is converted into a test set of the characteristic vector sequence through the processes. The HMM model utilizes the dual information of the evolution relation of the signal feature vector and the feature vector time sequence, and has higher event recognition rate compared with the traditional recognition algorithm which only classifies the static features or feature vectors extracted according to the signal in a certain time period.
Example 3
Based on embodiment 1, the step S2 includes the following steps:
training a typical event HMM model by using the feature vector sequence off-line to establish a typical event HMM model library; the method comprises the following steps of training parameters of various event HMM models in an off-line mode based on a feature vector sequence set of typical events, and establishing an off-line typical event HMM model library for pipeline safety monitoring, wherein the off-line training comprises two steps of initialization and iterative updating of model parameters:
(1) initializing model parameters
One HMM model is generally denoted as:
λ＝(π,A,B) (4)
wherein, pi is an initial probability distribution vector, pi ═ pi (pi)1,π2,Λ,πN)，
πi＝P(qt＝θi),1≤i≤N (5)
A is a state transition probability matrix, and A ═ aij)N×N，
aij＝P(qt+1＝θj|qt＝θi),1≤i,j≤N (6)
B is an observed value probability matrix, and B is (B)j(o))，
Where N represents the number of states of the Markov chain in the model, e.g., where N states are θ1,θ2,Λ,θN，qtRepresenting the state of the Markov chain at time t, qt∈(θ1,θ2,Λ,θN)；
The observation value probability matrix B is described by a Gaussian mixture model GMM, and the parameters of the GMM include the number M of Gaussian mixture elements and a mean value matrix muN×M×DCovariance matrix sigmaN×M×D×DAnd Gaussian component weight coefficient matrix W3×3；
M in observation value probability matrix B, i.e. in equations (7) and (8), is the number of Gaussian elements, BjlIs the first Gaussian density function in state j, wjl、μjl、|σjlL respectively represents the determinant values of the weight coefficient, the mean vector and the covariance matrix of the l mixed Gaussian element in the j state; bj(o) represents the probability of generating the feature vector o in the j-th state, and D is the dimension of the feature vector.
In this embodiment, Kmeans clustering is performed on a feature matrix O' formed by a plurality of groups of feature vector sequences of each type of event, and the optimal cluster number is used as the number N of hidden states experienced by long-term signals of each type of event; the contour Coefficient (Silhouette coeffient) is used as a clustering effect evaluation index, as shown in fig. 5, the Coefficient value range is [ -1, 1], the larger the value is, the better the clustering effect is, and finally, according to the clustering number corresponding to the maximum value of the contour Coefficient of each type of event in fig. 5, the hidden state numbers of five types of events, namely manual mining, mechanical mining, traffic interference, factory interference and background noise, are respectively determined as follows: 2.2, 3 and 3.
After the number N of hidden states of each event is determined, each parameter of the HMM model is initialized, the initialization of pi and A has little influence on the result of model convergence, and pi and A are initialized randomly under the constraint conditions that the sum of all elements in pi is 1 and the sum of the probability of each row in the A matrix is 1.
The initialization of the parameter B has a large influence on the convergence of the model, and the probability that each state in the HMM generates a certain feature vector is estimated by adopting a Gaussian mixture model GMM. GMM parameter estimation adopts a Kmeans clustering algorithm, each feature vector in a plurality of groups of feature vector sequences of each type of event is distributed to N hidden states through Kmeans clustering, the feature vector in each state is subjected to Kmeans clustering again, the number of clustering centers is set as the number M of mixed Gaussian elements of a GMM model, and each element w of a weight coefficient, a mean vector and a covariance matrix is subjected to weight coefficient, mean vector and covariance matrix according to the following three formulasjl、μjl、σjlRespectively initialized.
μjlMean vector of feature vectors in Gauss Yuan l (10)
σjlCovariance matrix of feature vector in Gauss element l (11)
By this, the parameter initialization process of the HMM model ends.
(2) Model parameter updating based on Baum-Welch algorithm
After the initialization of the model parameters, the parameters of the model are iteratively updated based on the Baum-Welch algorithm by using the training set of feature vector sequences until convergence.
For the feature vector sequence set O', assuming that the feature vector sequences are independent of each other, the joint probability distribution of all feature vector sequences is as follows:
the parameter estimation of the HMM is based on the frequency of occurrence of each event, and is estimated using the Baum-Welch algorithm with K mutually independent feature vector sequences.
And calculating a weight coefficient, a mean vector and a covariance matrix by using a re-estimation formula, wherein the re-estimation formula for updating parameters is as follows:
wherein the transition probability
Indicating that the k-th group of feature vector sequences is in the state i at the time t and in the state i at the time t +1Probability of state j;
forward variable
The state of the k group observation vector sequence at the time t is represented as thetaiThe observed value in the first t seconds is o1,o2,Λ,otThe probability of (d);
backward variable
The state of the k group observation vector sequence at the time t is represented as thetaiAnd an observed value of o from t +1 second to L secondst+1,ot+2,Λ,oLThe probability of (d);
And (3) updating the HMM parameters once by using all the characteristic vector sequences of the training set according to a re-estimation formula respectively to serve as one iteration of the training process, and stopping the iteration when the combined probability P (O' | lambda) error of the current iteration process and the later iteration process is smaller than the set convergence error. So far, model training is completed.
Taking model parameter training of an artificial mining event as an example, after iteration of parameter updating is converged, the data structure of the obtained HMM model is represented as follows:
hmm＝{N:2,numM:[3,3],π:[1×2double],A:[2×2double],gmm:{1×2struct}}
the HMM model includes a plurality of HMM states, each HMM state includes a number of gaussian elements in the GMM model, each HMM state includes an initial probability distribution, each HMM state includes a number of gaussian elements in the GMM model, each HMM state includes a number of gaussian element in the GMM model, each HMM state includes a number of each HMM state, each HMM state includes a number of states, each HMM state, each state includes a number of each state, each state corresponds to a number of each state, each HMM is a number of each state, each state corresponds to a number of each state, each state corresponds to a number of each state, each state corresponds to a structure, each state corresponds to a state, each state corresponds to a state, eachThe parameters of the GMM model in each state. The maximum number of iteration steps in the iterative update of the parameter is set to 8, and the convergence error e is 5 × 10-6Fig. 7 is an iteration curve of the training process of the HMM model for each type of event, and the HMM model can be converged after 2 iterations, which illustrates that the learning speed of the HMM model is relatively high.
And training the models in sequence based on the feature vector sequence training sets of the five events to respectively obtain five corresponding HMM models, storing the parameters of each model, and completing the construction of a typical event HMM model library for pipeline safety monitoring.
Example 4
Based on embodiment 1, the step S3 includes the following steps:
step S3.1: during online identification, calculating output probabilities obtained by inputting test signals into various typical event models according to a model matching thought, and taking a model with the highest probability as an event judgment type of the test signals; the event test signal is converted into a feature vector sequence based on the feature extraction method in the step 1, the feature vector sequence is input into a typical event HMM model library, the Bayesian posterior probability of the current test feature vector sequence under each model in the model library is calculated through a Viterbi algorithm, and an event label corresponding to the model with the maximum probability is output as the event type of the test signal, so that the identification and classification of the signal are realized.
The process of calculating the Bayesian posterior probability based on the Viterbi algorithm:
a) initialization:
b) recursion:
c) and (4) terminating:
wherein, deltat(i)＝maxP(q1,Λ,qt,qt＝θi,o1,Λ,otλ) represents a path q along time t1,q2,Λ,qtAnd q ist＝θiI.e. the state is theta at time tiIs generated o1,o2,Λ,otThe maximum probability of (d);
indicates a state of theta at time tiAll paths q of1,q2,Λ,qtThe t-1 node of the path with the highest probability;
when the recursion is complete, when t is L, P*Corresponding to the maximum output probability of the current feature vector sequence under the given model parameters, and so on, according to the model parameters lambda of each HMM in the model libraryiAnd current feature vector sequence OiCalculating the prediction probability P of the feature vector sequence under each modeli *＝P(Oi|λi) Selecting the model event label corresponding to the maximum prediction probability as OiThe identification and classification of the signals is realized.
In this embodiment, 10 sets of test signal samples of the above five types of events are tested, and the event test recognition rate is obtained as shown in fig. 8, where the overall average recognition rate of the HMM model reaches 98%, where the test recognition rates for manual mining, mechanical mining, traffic interference and background noise can reach 100%, and the test recognition rate for only confusable factory interference is 90%, which may be caused by that the factory interference event includes various types of factory data such as forging plants, processing plants, and the like.
The method provided by the invention is compared with the pipeline safety monitoring event recognition result based on the traditional method, namely a non-time sequence classification model, such as five types of Support Vector Machine (SVM), random forest Radio Frequency (RF), Xgboost model XGB, decision tree DT and Bayesian Network (BN). For the HMM model, 10 groups of test cases are randomly selected from various event data sets to serve as the HMM model, and the rest test cases are used for training the HMM model; the other classification models are divided into training sets and test sets in the conventional ratio of 7: 3. The feature vector sequence of the traditional method is constructed by two methods respectively: 1) directly splicing the feature vectors of each short-time signal unit to form a feature vector of a long-time signal with the length of DxT, and setting a feature vector set 1; 2) and directly carrying out feature extraction similar to the short-time signal on the long-time signal to obtain a one-dimensional feature vector with the length of D, and setting the one-dimensional feature vector as a feature vector set 2. Respectively inputting the constructed feature set 1 and feature set 2 into a non-time sequence classification model for classification and recognition, and comparing the classification result with the recognition result of an HMM model, as shown in FIG. 9: the recognition effect of the five models based on the feature set 2 is better than that based on the feature set 1, which shows that the feature set embodying the time sequence relationship is beneficial to improving the recognition accuracy, but the best recognition rate of the five non-time sequence models aiming at the two feature sets is only 94%, and the whole recognition rate is 98% lower than the average recognition rate of the HMM model provided by the invention.
Step S3.2: calculating the maximum posterior probability P of the feature vector sequence of the unknown event type based on Viterbi algorithm formulas (18) - (21)*Meanwhile, outputting a hidden state sequence corresponding to the characteristic vector sequence of the obtained test signal, wherein a calculation formula of the hidden state sequence is as follows:
wherein arg represents δL(i) Maximum time thetaiThe value of (a) is selected,
represents the optimal implicit shape of time t-L,
represents the optimal hidden state of the Markov chain at the time t,
hidden state sequence
Can be used as the evolution process of the event state sequence to output, and can carry out short-time prediction according to the evolution process of the event state sequence. The evolution process of the average state sequence of each event is obtained based on all samples in the database of this embodiment, as shown in fig. 10, where L is 30. The state sequence of the factory interference and the background noise is stable all the time in the duration, which shows that the states of the two events are relatively stable and are continuous, the factory interference is a stable continuous state with the event, and the background noise is a stable continuous state without the event; the traffic interference and the manual mining events have discontinuity, so the state sequences of the two events are switched randomly or periodically in the two states, but the interval time periods for switching the two events are different; the state change of the mechanical digging is complex and long in duration, and the occurrence process of the event is also illustrated. When the signal analysis time is long enough, for example 24 hours a day, the state analysis of the all-weather time signal can know the specific time and position of the event, the event duration and the event occurrence rule, and by observing the event occurrence process, the time point and the occurrence frequency of the event which often occur can be counted, and the current stage to which the event is developed can be known, so that the prevention can be realized in the future, and the pipeline safety monitoring can be effectively carried out.
In conclusion, the HMM model of the invention models the time sequence of the event evolution process, and the event recognition rate is as high as 98%; the time sequence rule corresponding to the event state evolution is extracted and mined, the short-term prediction of the developing process of the occurring dynamic event is realized, and the problems that the pipeline safety event identification rate is low and the event prediction cannot be carried out due to the fact that the dynamic change of the event cannot be obtained through static analysis are solved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (5)
1. The HMM model-based pipeline security event recognition and knowledge mining method is characterized by comprising the following steps: the method comprises the following steps:
step 1: extracting multi-domain features of signals acquired by each space point to obtain a feature vector sequence of the signals;
step 2: inputting the feature vector sequence into an HMM model for off-line training to complete the establishment of a typical event HMM model library;
and step 3: after acquiring a current feature vector and a feature vector sequence of a signal to be recognized in the step 1, inputting the current feature vector and the feature vector sequence into a typical event HMM model library for recognition and outputting an event judgment type, and meanwhile, calculating an optimal hidden state sequence as event state sequence evolution process information for outputting to finish knowledge mining;
the step 1 comprises the following steps:
step 1.1: respectively constructing a typical event database of each space point based on typical physical event samples collected by each space point, and specifically comprising the following steps:
step 1.1.1: event initialization: defining the event type of each space point and the corresponding event label;
step 1.1.2: constructing sample data: the method comprises the steps that a set short-time signal unit SU accumulates signals collected in L short-time signal units SU as a long-time signal, and the long-time signal is used as sample data of each space point event;
step 1.1.3: constructing a typical event database: repeating the step 1.1.2 according to the duration to obtain all sample data of the events of each space point, and completing the construction of a typical event database of each space point;
step 1.2: performing multi-domain feature extraction on each short-time signal unit in a signal sample of a typical event database to obtain a feature vector of each short-time signal unit, combining feature vector sequences of various events according to a time sequence relation, and using the feature vector sequences as a feature vector sequence set for training an HMM model, wherein the feature vector sequence set of the HMM model is as follows:
O'＝[O(1),O(2),…,O(k),…,O(K)] (2)
the dimension of O' is D multiplied by L multiplied by K, D represents the dimension of the feature vector extracted by each short-time signal unit, L represents the length of the feature vector sequence and corresponds to the number of short-time signal units contained in the original-time signal, and K represents the number of samples of the feature vector sequence;
the k-th set of feature vector sequences is as follows:
O(k)＝[o1 (k),o2 (k),…,ot (k),…,oL (k)] (3)
wherein the content of the first and second substances,
and the characteristic vector of the t-th short-time signal unit in the characteristic vector sequence set is represented.
2. The HMM model-based pipeline security event recognition and knowledge mining method of claim 1, wherein: the step 2 comprises the following steps:
step 2.1: initializing parameters of the HMM model based on the feature vector sequence set;
step 2.2: based on the step 2.1, utilizing a feature vector sequence set, and based on a Baum-Welch algorithm, iteratively updating each parameter of the HMM model;
step 2.3: judging whether the joint probability error of the iteration process is smaller than the set convergence error or not, if so, stopping the iteration, and finishing HMM model training and building a typical event HMM model library; if not, jumping to step 2.2 to continue iteration.
3. The HMM model-based pipeline security event recognition and knowledge mining method according to claim 1 or 2, wherein: the step 3 comprises the following steps:
step 3.1: after the current feature vector sequence of the signal to be recognized is obtained in the step 1, inputting the current feature vector sequence into a typical event HMM model library, calculating the Bayesian posterior probability of the current feature vector sequence under each model in the HMM model library through a Viterbi algorithm, outputting an event label corresponding to the model with the maximum probability as the event type of the signal to be recognized, and completing the recognition and classification of the signal;
step 3.2: and inputting the current feature vector sequence into a typical event HMM model base, calculating a hidden state sequence corresponding to the current feature vector sequence through a Viterbi algorithm, outputting the hidden state sequence serving as event state sequence evolution process information, and finishing knowledge mining.
4. The HMM model-based pipeline security event recognition and knowledge mining method of claim 3, wherein: the step 3.1 specifically comprises the following steps:
step 3.1.1: initialization:
step 3.1.2: recursion:
wherein, deltat(i)＝maxP(q1,…,qt,qt＝θi,o1,…,otλ) represents a path q along time t1,q2,…,qtAnd q ist＝θiI.e. the state is theta at time tiIs generated o1,o2,…,otThe maximum probability of (d);
indicates a state of theta at time tiAll paths q of1,q2,…,qtThe t-1 node of the path with the highest probability;
step 3.1.3: and (4) terminating:
wherein, P*Representing the maximum output probability of the current feature vector sequence under the given model parameters;
step 3.1.4: according to each HMM model parameter lambda in the model libraryiAnd current feature vector sequence OiCalculating the prediction probability P of the feature vector sequence under each modeli *：
Pi *＝P(Oi|λi)
Selecting the model event label corresponding to the maximum prediction probability as the current feature vector sequence OiThe identification and classification of the signals is realized.
5. The HMM model-based pipeline security event recognition and knowledge mining method of claim 2, wherein: the multi-domain features employ 44-dimensional features of a time domain, a frequency domain, an inverse frequency domain, and a model coefficient analysis domain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811251182.4A CN109344195B (en) | 2018-10-25 | 2018-10-25 | HMM model-based pipeline security event recognition and knowledge mining method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811251182.4A CN109344195B (en) | 2018-10-25 | 2018-10-25 | HMM model-based pipeline security event recognition and knowledge mining method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109344195A CN109344195A (en) | 2019-02-15 |
| CN109344195B true CN109344195B (en) | 2021-09-21 |
Family
ID=65312089
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811251182.4A Active CN109344195B (en) | 2018-10-25 | 2018-10-25 | HMM model-based pipeline security event recognition and knowledge mining method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109344195B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110362065B (en) * | 2019-07-17 | 2022-07-19 | 东北大学 | State diagnosis method of anti-surge control system of aircraft engine |
| CN110673199B (en) * | 2019-08-30 | 2022-05-13 | 昆明理工大学 | U-shaped tube blockage state assessment method based on low-frequency sound pressure signal analysis |
| CN110946556B (en) * | 2019-12-27 | 2022-07-15 | 南京信息工程大学 | Parkinson resting state tremor evaluation method based on wearable somatosensory network |
| CN111242021B (en) * | 2020-01-10 | 2022-07-29 | 电子科技大学 | Distributed optical fiber vibration signal feature extraction and identification method |
| CN111639711B (en) * | 2020-05-29 | 2023-04-18 | 中国地质大学（武汉） | Oil pipeline leakage monitoring method based on pressure monitoring time sequence data |
| CN111539394B (en) * | 2020-07-08 | 2020-10-30 | 浙江浙能天然气运行有限公司 | Pipeline along-line third-party construction early warning method based on time domain characteristics and space-time information |
| CN112488371B (en) * | 2020-11-20 | 2022-09-13 | 国网福建省电力有限公司厦门供电公司 | Pipeline intelligent early warning method and system based on big data |
| CN112487631A (en) * | 2020-11-25 | 2021-03-12 | 中国科学院力学研究所 | Intelligent identification method for working condition parameters of transverse landslide buried pipeline |
| CN113407185B (en) * | 2021-03-10 | 2023-01-06 | 天津大学 | Compiler optimization option recommendation method based on Bayesian optimization |
| CN114254716B (en) * | 2022-03-02 | 2022-05-27 | 浙江鹏信信息科技股份有限公司 | High-risk operation identification method and system based on user behavior analysis |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201326909Y (en) * | 2008-09-27 | 2009-10-14 | 中国石油天然气股份有限公司 | Pipeline safety early warning system of combining sound vibration and electromagnetic wave |
| CN102543074A (en) * | 2011-12-30 | 2012-07-04 | 中国农业科学院农业信息研究所 | Agricultural product data acquisition system as well as voice recognition system and method of mobile equipment |
| CN103413443A (en) * | 2013-07-03 | 2013-11-27 | 太原理工大学 | Short-term traffic flow forecasting method based on hidden Markov model |
| CN106599920A (en) * | 2016-12-14 | 2017-04-26 | 中国航空工业集团公司上海航空测控技术研究所 | Aircraft bearing fault diagnosis method based on coupled hidden semi-Markov model |
| CN106683661A (en) * | 2015-11-05 | 2017-05-17 | 阿里巴巴集团控股有限公司 | Role separation method and device based on voice |
| CN107274086A (en) * | 2017-06-09 | 2017-10-20 | 东莞市毅豪电子科技有限公司 | A kind of gridding governance information approach based on hidden Markov model |
| CN107862108A (en) * | 2017-10-12 | 2018-03-30 | 成都阜特科技股份有限公司 | A kind of industrial machinery health status analysis and Forecasting Methodology and its system |
| CN108038111A (en) * | 2017-12-11 | 2018-05-15 | 中译语通科技股份有限公司 | A kind of machine translation pipeline method for building up and system, computer program, computer |
| US10068054B2 (en) * | 2013-01-17 | 2018-09-04 | Edico Genome, Corp. | Bioinformatics systems, apparatuses, and methods executed on an integrated circuit processing platform |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7702507B2 (en) * | 2005-11-10 | 2010-04-20 | Symbol Technologies, Inc. | Voice activated rapid deployment for mobile computers |
-
2018
- 2018-10-25 CN CN201811251182.4A patent/CN109344195B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201326909Y (en) * | 2008-09-27 | 2009-10-14 | 中国石油天然气股份有限公司 | Pipeline safety early warning system of combining sound vibration and electromagnetic wave |
| CN102543074A (en) * | 2011-12-30 | 2012-07-04 | 中国农业科学院农业信息研究所 | Agricultural product data acquisition system as well as voice recognition system and method of mobile equipment |
| US10068054B2 (en) * | 2013-01-17 | 2018-09-04 | Edico Genome, Corp. | Bioinformatics systems, apparatuses, and methods executed on an integrated circuit processing platform |
| CN103413443A (en) * | 2013-07-03 | 2013-11-27 | 太原理工大学 | Short-term traffic flow forecasting method based on hidden Markov model |
| CN106683661A (en) * | 2015-11-05 | 2017-05-17 | 阿里巴巴集团控股有限公司 | Role separation method and device based on voice |
| CN106599920A (en) * | 2016-12-14 | 2017-04-26 | 中国航空工业集团公司上海航空测控技术研究所 | Aircraft bearing fault diagnosis method based on coupled hidden semi-Markov model |
| CN107274086A (en) * | 2017-06-09 | 2017-10-20 | 东莞市毅豪电子科技有限公司 | A kind of gridding governance information approach based on hidden Markov model |
| CN107862108A (en) * | 2017-10-12 | 2018-03-30 | 成都阜特科技股份有限公司 | A kind of industrial machinery health status analysis and Forecasting Methodology and its system |
| CN108038111A (en) * | 2017-12-11 | 2018-05-15 | 中译语通科技股份有限公司 | A kind of machine translation pipeline method for building up and system, computer program, computer |
Non-Patent Citations (4)
| Title |
|---|
| Pipeline damage and leak sound recognition based on HMM;Changsheng Ai 等;《2008 7th the World Congress on Intelligent Control and Automation》;20080808;1940-1944 * |
| The early-warning model of equipment chain in gas pipeline based on DNN-HMM;Jingwei Qiu 等;《Journal of Natural Gas Science and Engineering》;20151115;第27卷;1710-1722 * |
| 基于DOFVS的新型压力输水管道泄露在线监测方法;吴慧娟 等;《仪器仪表学报》;20170115;第38卷(第01期);159-165 * |
| 基于最优阶次HMM的电机故障诊断方法研究;胡为 等;《仪器仪表学报》;20130315;第34卷(第03期);524-530 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109344195A (en) | 2019-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109344195B (en) | HMM model-based pipeline security event recognition and knowledge mining method | |
| CN108932480B (en) | Distributed optical fiber sensing signal feature learning and classifying method based on 1D-CNN | |
| Wu et al. | A dynamic time sequence recognition and knowledge mining method based on the hidden Markov models (HMMs) for pipeline safety monitoring with Φ-OTDR | |
| CN111242021B (en) | Distributed optical fiber vibration signal feature extraction and identification method | |
| CN107425906B (en) | Distributing optical fiber sensing signal processing method towards underground pipe network safety monitoring | |
| Wu et al. | One-dimensional CNN-based intelligent recognition of vibrations in pipeline monitoring with DAS | |
| CN113049084B (en) | Attention mechanism-based Resnet distributed optical fiber sensing signal identification method | |
| CN110995339B (en) | Method for extracting and identifying time-space information of distributed optical fiber sensing signal | |
| CN107590516B (en) | Gas transmission pipeline leakage detection and identification method based on optical fiber sensing data mining | |
| CN111222743B (en) | Method for judging vertical offset distance and threat level of optical fiber sensing event | |
| CN111649817B (en) | Distributed optical fiber vibration sensor system and mode identification method thereof | |
| Yang et al. | Pipeline safety early warning by multifeature-fusion CNN and LightGBM analysis of signals from distributed optical fiber sensors | |
| CN114692681B (en) | SCNN-based distributed optical fiber vibration and acoustic wave sensing signal identification method | |
| CN111520615B (en) | Pipe network leakage identification and positioning method based on line spectrum pair and cubic interpolation search | |
| CN112749637B (en) | SNN-based distributed optical fiber sensing signal identification method | |
| Fouda et al. | Pattern recognition of optical fiber vibration signal of the submarine cable for its safety | |
| CN113822201B (en) | Deep learning method for underwater object shape recognition based on flow field velocity component time course | |
| US20220329068A1 (en) | Utility Pole Hazardous Event Localization | |
| Wu et al. | Simultaneous extraction of multi-scale structural features and the sequential information with an end-to-end mCNN-HMM combined model for fiber distributed acoustic sensor | |
| Moallemi et al. | Model-based vs. data-driven approaches for anomaly detection in structural health monitoring: A case study | |
| Marie et al. | A hybrid model integrating MPSE and IGNN for events recognition along submarine cables | |
| Romano et al. | Burst detection and location in water distribution systems | |
| CN110361164A (en) | A kind of submarine optical fiber cable breakpoint location and disturbance location analysis method and system | |
| CN110135645B (en) | All-fiber dynamic and static monitoring and trend prediction system and method for overhead transmission line | |
| US20230366725A1 (en) | Utility pole integrity assessment by das and machine learning using environmental noise |
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 |