CN109033450A - Lift facility failure prediction method based on deep learning - Google Patents

Lift facility failure prediction method based on deep learning Download PDF

Info

Publication number
CN109033450A
CN109033450A CN201810962887.0A CN201810962887A CN109033450A CN 109033450 A CN109033450 A CN 109033450A CN 201810962887 A CN201810962887 A CN 201810962887A CN 109033450 A CN109033450 A CN 109033450A
Authority
CN
China
Prior art keywords
elevator
time
prediction
event
type
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.)
Granted
Application number
CN201810962887.0A
Other languages
Chinese (zh)
Other versions
CN109033450B (en
Inventor
王莉
江海洋
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.)
Taiyuan University of Technology
Original Assignee
Taiyuan University of 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 Taiyuan University of Technology filed Critical Taiyuan University of Technology
Priority to CN201810962887.0A priority Critical patent/CN109033450B/en
Publication of CN109033450A publication Critical patent/CN109033450A/en
Application granted granted Critical
Publication of CN109033450B publication Critical patent/CN109033450B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
    • G06F2216/03—Data mining

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Elevator Control (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The present invention is based on the lift facility failure prediction methods of deep learning, belong to elevator faults electric powder prediction；Technical problem to be solved is to provide a kind of method in time, accurately predicting elevator faults type and elevator faults time；In order to solve the above-mentioned technical problem, specific steps of the present invention are summarized are as follows: first acquire elevator faults and record information, establish real-time elevator faults information bank；It then is sequence of events and time series by elevator faults information processing；Again using sequence of events and time series as the input data of double LSTM, the output embedding of two sequences is obtained by the repetitive exercise of Recognition with Recurrent Neural Network；Combine two output embedding using joint layer, training obtains the background knowledge of intensity function and the non-linear expression of historical influence；Finally according to the characterization result of intensity function, elevator faults type and time are predicted；The present invention can supplemental lift maintenance personal take related precautionary measures early, avoid the generation of event of failure.

Description

Lift facility failure prediction method based on deep learning
Technical field
The present invention is based on the lift facility failure prediction methods of deep learning, belong to elevator faults electric powder prediction.
Background technique
With being increasing for skyscraper, the concern of the quality of elevator by people.Failure stop ladder, operation it is unsmooth even The elevator of generation accident has influenced daily life.The failure rate of elevator is reduced, in time, accurately detects debugging Method needs further research.Current Elevator Fault Diagnosis and detection method is mostly to combine elevator structure and principle, proposes needle To the detection method of the mechanical system of elevator, electric control system and safety system, but this needs to consume a large amount of point Analyse the time.
Summary of the invention
The present invention overcomes the shortcomings of the prior art, and technical problem to be solved is to provide one kind in time, accurately Lift facility failure prediction method based on deep learning achievees the purpose that prediction elevator faults type and elevator faults time, Allow maintenance master worker periodically to go to detect according to prediction result, reduces the incidence of elevator faults.
In order to solve the above-mentioned technical problem, the technical solution adopted by the present invention are as follows: the lift facility event based on deep learning Hinder prediction technique, includes the following steps:
Step 1 establishes real-time elevator faults information bank, according to fault message is related, information is complete, information is non-duplicate, event Four principle of operation for hindering the non-artificial mistake of information, screens pretreatment information sequence of effective elevator faults information as network, electricity Terraced fault message includes failure logging information and elevator essential information, and wherein failure logging information includes: elevator faults type, event Hinder reason and fault time, elevator essential information include: elevator date of manufacture, elevator present position, elevator model and elevator longevity Life；
Step 2, build time sequence, including two kinds of features: 1) statistics various types of elevator faults numbers, 2) elevator base This information, both characteristic bindings get up to constitute time series；
Step 3, tectonic event sequence, including two kinds of features: 1) being successively tactic electricity according to time of failure The data record sequence of terraced fault type, 2) time interval between adjacent two event of failure, both characteristic bindings get up to constitute Sequence of events；
Step 4, building LSTM neural network；
Step 5, with double LSTM neural metwork training time serieses and sequence of events, obtain the background knowledge of intensity function Characterization and historical influence characterization；
Step 6 merges background knowledge characterization and historical influence characterization by stratum conjunctum joint layer；
Step 7, using the intensity function learnt by double LSTM, elevator faults are predicted by fault type prediction interval Type utilizes the penalty values of Classification Loss layer quantization class prediction；
Step 8, using the intensity function learnt by double LSTM, elevator faults are predicted by fault time prediction interval Time utilizes the penalty values for returning the prediction of loss layer quantization time；
Step 9 is based on step 7 and the continuous repetitive exercise neural network model of step 8, obtains optimal network model, so Trained optimal models are used afterwards, and elevator faults are predicted by fault type prediction interval and fault time prediction interval respectively Type and elevator faults time；
The data of real-time update are input to model by step 10, improvement and optimization model, elevator faults information bank real-time update The accuracy of middle test model, and sophisticated model is corrected according to actual feedback situation.
Specifically, in the step 2 " build time sequence " way are as follows: it is assumed that tested elevator shares M platform, effective failure Type class shares N kind, and elevator faults time window number is n, then can be by the kth kind failure of i-th elevator, j-th of time window The effective number of faults of type is denoted as1) with one week for time window, Count various fault type numbers2) elevator essential information, both characteristic bindings, which get up, constitutes time series data, tool Body surface shows as follows:
Wherein m indicates elevator model, and d indicates the elevator date of manufacture, and l indicates elevator present position, when n indicates elevator faults Between window number, indicated by taking double horizontal lines as an example in above-mentioned expression M platform elevator in first time window, N kind fault type The expression unit of statistics number and elevator essential information.
Specifically, in the step 3 " tectonic event sequence " way are as follows: 1) by all elevator informations according to elevator id not It is time stamp data by elevator faults time conversion with being stored separately, electricity is arranged according to time of failure for every elevator Terraced fault type, 2) calculate the interval time of the adjacent event of failure of every elevator, 3) fault type T and interval of timestamps I composition Sequence of events minimum unit is specifically expressed as follows shown:
The elevator id indicates unique designation --- the elevator number of elevator.
Specifically, the way for " constructing LSTM neural network " in the step 4 are as follows:
The specific formula of Recognition with Recurrent Neural Network variant LSTM used in the present invention is defined as follows:
it=g (Wi xt+Uiyt-1+Vict-1+bi),
ft=g (Wfxt+Ufyt-1+Vfct-1+bf),
ct=ftct-1+it⊙tanh(Wcxt+Ucyt-1+bc),
ot=g (Woxt+Uoyt-1+Voct+bo),
yt=ot⊙tanh(ct)
It is input gate in LSTM above, forgets door, the calculation formula of location mode and out gate, wherein itIndicate input The calculation formula of door, ftIndicate the calculation formula of forgetting door, ctIndicate location mode calculation formula, otAnd ytIt is common to indicate output The calculation formula of door, otIt indicates that partially output is gone out using which of g function determination unit state, last ytIt indicates unit State using tanh processing after and otBeing multiplied is the determination part to be exported；Wherein U, V are respectively that neural metwork training needs The parameter matrix to be learnt, ⊙ represent array element and are successively multiplied, and function g uses sigmoid activation primitive,Indicate defeated Enter sequence,It indicates to generate implicit layer state, above formula can simplify as formula: (yt,ct)=LSTM (xtyt-1+ ct-1)。
Wherein, LSTM building process specific practice is as follows:
(1) input layer number i, node in hidden layer j, output layer number of nodes k and the list of network netinit: are determined First state dimension, the connection weight between initialization input layer, hidden layer and output layer neuron, input gate forget door, out gate With the connection weight W of cell factoryi,Wf,WO,Wc, initial threshold value bf,bc,bo,bi, give learning rate and neuron motivate letter Number；
(2) it calculates the output valve of each neuron: being f for LSTMt,it,ct,ot,ht；
(3) error calculation: according to the error term of prediction output and each neuron of anticipated output matrix retrospectively calculate；
(4) weight W right value update: is connected to the network according to neural network forecast error updatei,Wf,WO,Wc；
(5) threshold value updates: according to neural network forecast error update network node threshold value bf,bc,bo,bi；
(6) judge whether to terminate, if being not over return step (2)；
(7) after, 5 are entered step using trained double LSTM neural networks；
Specifically, the way in the step 5 " with double LSTM neural metwork training time serieses and sequence of events " are as follows:
Specific formula is as follows:
In above formulaIndicate time series,Indicate sequence of events, wherein ziIndicate the thing in sequence of events Part type, tiThe timestamp that expression event occurs, two sequences are respectively used to study background knowledge and historical influence.
Specifically, " merging background knowledge characterization and historical influence table by stratum conjunctum joint layer in the step 6 The way of sign " are as follows: using tanh function as joint layer Copula, construct the non-linear of point process intensity function and reflect It penetrates,
By the following institute of formula of the Nonlinear Mapping of the stratum conjunctum joint layer point process intensity function constructed Show:
Specifically, " using the intensity function learnt by double LSTM, passing through fault type prediction interval in the step 7 Predict elevator faults type, utilize Classification Loss layer quantization class prediction penalty values " way are as follows:
Wherein, the calculation formula of failure predication main classes and subclass is as follows:
Ut=softMax (Wuet+bU)
ut=softMax (Wu[et,Ut]+bu)
In above formula, U and u respectively represent main classes and subclass, WuAnd buIndicate the model parameter square to be learnt in the training process Battle array；First with softMax function by etIt is out of order primary categories U as input predictiont；Then softMax letter is utilized again Number, by etWith main classes predicted value UtThe predicted value of subclass is calculated as input.
Specifically, " using the intensity function learnt by double LSTM, passing through fault time prediction interval in the step 8 Predict the elevator faults time.Utilize the penalty values for returning the prediction of loss layer quantization time.", way specifically:
st=Wset+bs
Wherein, stIt is the corresponding timestamp of each event, WsAnd bsIndicate the model parameter square to be learnt in the training process Battle array.
Specifically, " using trained optimal models in the step 9, passing through fault type prediction interval and fault time Prediction interval predicts type and the elevator faults time of elevator faults respectively ", way specifically:
Wherein the loss of entire model is the sum of time prediction loss and event type prediction loss.Utilize intersection entropy loss Function does event type prediction, does timestamp prediction with quadratic loss function.Specific loss function formula is as follows:
N represents node total number in above formula, and case point is indexed by l, i.e. l indicates which event, WuIndicate nerve net The parameter matrix for the event category that network model learns,It is the timestamp of next case point,Represent the history letter at time point Breath uses Gauss penalty for loss of time function part:
Wherein σ indicates unity variance, takes σ2=10；
By minimize loss and, the entire model of circuit training finally obtains optimal model.
Specifically, in the step 10 " improvement and optimization model " way are as follows: pass through training set obtain elevator faults prediction Then model obtains test set using the fault message storehouse of real-time update, based on the accuracy of test set examination model, and in real time Improvement and optimization model.
Through the above steps, the elevator device fault prediction model of the point process intensity function based on double LSTM can be completed Building form time series and event sequence data by pre-processing to elevator faults data, double LSTM minds can be passed through Fault signature is obtained through network model, further two kinds of embedding of joint obtain the non-linear representative function of intensity function, lead to The entire model of circuit training is crossed, finally the intensity function Jing Guo point process may finally predict fault time and the failure classes of elevator Type.Elevator faults type and time prediction problem of the present invention suitable for solving practical problems, can help elevator reparing teacher The type and time that elevator faults information prediction elevator will break down known to Fu Liyong can carry out prevention ahead of time and arrange It applies, avoids the generation of elevator faults, reduce the loss of elevator faults bring and danger, be of very high actual application value.
To sum up, the present invention discloses a kind of lift facility failure prediction method based on deep learning, and this method is directed to elevator The on-line operation measurement data of system middle-high density sampling proposes a kind of sequence of events and time sequence to be respectively trained with couple LSTM Column, then in conjunction with the data mining algorithm of point process intensity function.The present invention is from the angle of data, based on the event of elevator history Hinder information sequence, real-time remote monitoring failure elevator is excavated from the elevator warning information of magnanimity using neural network model The inherent law of elevator faults promptly and accurately predicts elevator faults type and time, so that supplemental lift maintenance personal is early Related precautionary measures are taken, the generation of event of failure is avoided.
LSTM (Long Short-TermMemory) is shot and long term memory network, is a kind of time recurrent neural network, is fitted Together in processing and predicted time sequence in be spaced and postpone relatively long critical event.LSTM is different from the place of RNN, mainly Being it in the algorithm joined " processor " judged whether information is useful, and the structure of this processor effect is referred to as cell.Three fan doors are placed in one cell, is called input gate respectively, forgets door and out gate.One information enters LSTM Network in, can be according to rule to determine whether useful.The information for only meeting algorithm certification can just leave, the letter not being inconsistent Breath is then passed into silence by forgeing door.At present it has been proved that LSTM is the effective technology for solving long sequence Dependence Problem, and this skill The universality of art is very high.Each researcher proposes the variable version of oneself according to LSTM one after another, this just allows LSTM can handle Ever-changing Perpendicular Problems.
Application of the RNN on sequence data: from application scenarios, RNN can be applied to two as constructing module of the invention Kind of sequence data: time series and sequence of events, the two can collaborative modeling, be specifically described as follows:
(1) time series: refer to a kind of synchronizing sequence that the Information Statistics of related data are recorded in same time interval.? Time series can capture the feature changed over time in time in nearest time window, therefore RNN often makees time series Correlated series forecasting problem is done for input.For example, video frame can regard a kind of time series data as, we can be according to history Video frame analyzes its inner link, predicts the content of next frame.RNN has been widely used for the neck such as video analysis and speech recognition Domain.
(2) sequence of events: refer to a kind of asynchronous sequence by time stab copyist's part random character.Sequence of events will be random Input of the timestamp of generation as RNN enables the long-term dependence between the more efficient capturing events of sequence of events.Therefore The present invention carries out collaborative modeling to two kinds of sequence datas using double LSTM, can not only capture the update rule of synchronizing information, but also can It is avoided with capturing the asynchronous randomness of burst information using the Nonlinear Mapping of Recognition with Recurrent Neural Network formation condition intensity function The limitation of parametric assumption bring.
Point process refers to the stochastic model of a time series or time series.Point process is that a kind of modeling sequence pushes away The mathematical framework of algorithm is recommended, it measures the dynamic changeability of point process using intensity function.Point process be divided into event point process and Object point process.Event point process: earthquake or other disaster events；The Access Events of server；It is plant-manufactured unqualified Product event；The traffic accident event etc. that road junction occurs.The point process of tape label is initially exactly to be used to predict earthquake and aftershock Pests occurrence rule problem；Object point process: the position that highway is got on the car；The gene etc. of DNA.
The development course introduction of intensity function: mathematical framework of the point process as modeling series model, using conditional intensity The dynamic changeability of function measurement point process.The definition of intensity function: time window [t+dt) in, λ (t) represent in history thing Part Ht={ zi,ti|ti< t } occur under the premise of new events probability of happening, specific formula is as follows:
Wherein E (dN (t) | Ht) indicate in history HtOn the basis of, time interval [t+dt) phase of number occurs for interior event Prestige value.Conditional intensity function plays critical effect in point process sequence prediction algorithm, because intensity function parameterizes shape The different point process of formula will obtain different effects.The change procedure of the intensity function of point process is as follows: Poisson process；Strengthening version Poisson process；Huo Kesi process；Active point process；Self-correction point process；RMTPP model；TRPP model.
The parameterized form of intensity function consists of two parts: background knowledge and history feature.Above method is summarized As shown in table 1.First five kind intensity function is the parameter model artificially assumed according to priori knowledge out as seen from Table 1, is existed certain Limitation, model cannot comply fully with the complicated dynamic change of real sequence problem.And RMTPP model using LSTM learn come From the parameter of the history feature of sequence of events.Although ignoring the background characteristics of time series, the intensity as half parametric Function, RMTPP model also achieve good results；TRPP model builds time series and sequence of events using double LSTM respectively Mould, background knowledge and historical information as intensity function.Therefore the thought for mainly using for reference latter two model constructs our mould Type.
1 point process intensity function of table
The present invention has the advantages that compared with prior art.
1, neural network LSTM and point process are combined and are used for elevator faults prediction by the present invention, are proposed a kind of with double LSTM Sequence of events and time series is respectively trained, then in conjunction with the data mining algorithm of point process intensity function, from the angle of data It sets out, is based on elevator historical failure information sequence, the inherent law for including using neural network model training data passes through nerve The changing rule of the non-linear representative learning elevator faults type of network, to reach prediction elevator faults type and elevator faults The purpose of time allows maintenance master worker periodically to go to detect according to prediction result, reduces the incidence of elevator faults.
2, the present invention utilizes the mass data of elevator faults, and using neural network learning, inherent law occurs in it, avoids The false judgment artificially assumed, so that model more precise and high efficiency.
3, model can verify the accuracy of model, in time using the fault message storehouse of real-time update in time in the present invention Optimized model, so that model has timeliness.
Detailed description of the invention
Fig. 1 is general steps flow chart of the invention.
Fig. 2 is the collaborative modeling figure of time series and time series in the present invention.
Fig. 3 is fault type prediction and fault time prediction framework figure in the present invention.
Specific embodiment
The present invention is described in further details with reference to the accompanying drawing.
As shown in Figure 1, including the following steps: the present invention is based on the lift facility failure prediction method of deep learning
Step 1 establishes real-time elevator faults information bank, according to fault message is related, information is complete, information is non-duplicate, event Four principle of operation for hindering the non-artificial mistake of information, screens pretreatment information sequence of effective elevator faults information as network, electricity Terraced fault message includes failure logging information and elevator essential information, and wherein failure logging information includes: elevator faults type, event Hinder reason and fault time, elevator essential information include: elevator date of manufacture, elevator present position, elevator model and elevator longevity Life.
Step 2, build time sequence, including two kinds of features: 1) statistics various types of elevator faults numbers, 2) elevator base This information, both characteristic bindings get up to constitute time series；
Specifically, the way of " build time sequence " are as follows: it is assumed that tested elevator shares M platform, effective fault type type is total There is N kind, elevator faults time window number is n, then can be by the effectively event of the kth kind fault type of i-th elevator, j-th of time window Barrier quantity is denoted as1) with one week for time window, various failures are counted Type number2) elevator essential information, both characteristic bindings, which get up, constitutes time series data, is specifically expressed as follows institute Show:
Wherein m indicates elevator model, and d indicates the elevator date of manufacture, and l indicates elevator present position, when n indicates elevator faults Between window number, indicated by taking double horizontal lines as an example in above-mentioned expression M platform elevator in first time window, N kind fault type The expression unit of statistics number and elevator essential information.
Step 3, tectonic event sequence, including two kinds of features: 1) being successively tactic electricity according to time of failure The data record sequence of terraced fault type, 2) time interval between adjacent two event of failure, both characteristic bindings get up to constitute Sequence of events；
Specifically, the way of " tectonic event sequence " are as follows: 1) all elevator informations according to elevator id difference are stored separately, It is time stamp data by elevator faults time conversion, elevator faults type is arranged according to time of failure for every elevator, 2) calculate the interval time of the adjacent event of failure of every elevator, 3) fault type T and interval of timestamps I composition sequence of events are minimum Unit is specifically expressed as follows shown:
The elevator id indicates unique designation --- the elevator number of elevator.
Step 4, building LSTM neural network；
Specific way are as follows:
The specific formula of Recognition with Recurrent Neural Network variant LSTM used in the present invention is defined as follows:
it=g (Wi xt+Uiyt-1+Vict-1+bi),
ft=g (Wfxt+Ufyt-1+Vfct-1+bf),
ct=ftct-1+it⊙tanh(Wcxt+Ucyt-1+bc),
ot=g (Woxt+Uoyt-1+Voct+bo),
yt=ot⊙tanh(ct)
It is input gate in LSTM above, forgets door, the calculation formula of location mode and out gate, wherein itIndicate input The calculation formula of door, ftIndicate the calculation formula of forgetting door, ctIndicate location mode calculation formula, otAnd ytIt is common to indicate output The calculation formula of door, otIt indicates that partially output is gone out using which of g function determination unit state, last ytIt indicates unit State using tanh processing after and otBeing multiplied is the determination part to be exported；Wherein U, V are respectively that neural metwork training needs The parameter matrix to be learnt, ⊙ represent array element and are successively multiplied, and function g uses sigmoid activation primitive,Indicate defeated Enter sequence,It indicates to generate implicit layer state, above formula can simplify as formula: (yt,ct)=LSTM (xtyt-1+ ct-1)。
Wherein, LSTM building process specific practice is as follows:
(1) input layer number i, node in hidden layer j, output layer number of nodes k and the list of network netinit: are determined First state dimension, the connection weight between initialization input layer, hidden layer and output layer neuron, input gate forget door, out gate With the connection weight W of cell factoryi,Wf,WO,Wc, initial threshold value bf,bc,bo,bi, give learning rate and neuron motivate letter Number；
(2) it calculates the output valve of each neuron: being f for LSTMt,it,ct,ot,ht；
(3) error calculation: according to the error term of prediction output and each neuron of anticipated output matrix retrospectively calculate；
(4) weight W right value update: is connected to the network according to neural network forecast error updatei,Wf,WO,Wc；
(5) threshold value updates: according to neural network forecast error update network node threshold value bf,bc,bo,bi；
(6) judge whether to terminate, if being not over return step (2)；
(7) after, 5 are entered step using trained double LSTM neural networks；
Step 5, with double LSTM neural metwork training time serieses and sequence of events, obtain the background knowledge of intensity function Characterization and historical influence characterization；
Specifically, the way of " with double LSTM neural metwork training time serieses and sequence of events " are as follows:
Specific formula is as follows:
In above formulaIndicate time series,Indicate sequence of events, wherein ziIndicate the thing in sequence of events Part type, tiThe timestamp that expression event occurs, two sequences are respectively used to study background knowledge and historical influence.
Step 6 merges background knowledge characterization and historical influence characterization by stratum conjunctum joint layer；
Specific way are as follows: using tanh function as joint layer Copula, construct point process intensity function Nonlinear Mapping,
By the following institute of formula of the Nonlinear Mapping of the stratum conjunctum joint layer point process intensity function constructed Show:
Step 7, using the intensity function learnt by double LSTM, elevator faults are predicted by fault type prediction interval Type utilizes the penalty values of Classification Loss layer quantization class prediction；
Specific way are as follows:
The calculation formula of failure predication main classes and subclass is as follows:
Ut=softMax (Wuet+bU)
ut=softMax (Wu[et,Ut]+bu)
In above formula, U and u respectively represent main classes and subclass, WuAnd buIndicate the model parameter square to be learnt in the training process Battle array；First with softMax function by etIt is out of order primary categories U as input predictiont；Then softMax letter is utilized again Number, by etWith main classes predicted value UtThe predicted value of subclass is calculated as input.
Step 8, using the intensity function learnt by double LSTM, elevator faults are predicted by fault time prediction interval Time utilizes the penalty values for returning the prediction of loss layer quantization time；
Specific way are as follows:
st=Wset+bs
Wherein, stIt is the corresponding timestamp of each event, WsAnd bsIndicate the model parameter square to be learnt in the training process Battle array.
Step 9 is based on step 7 and the continuous repetitive exercise neural network model of step 8, obtains optimal network model, so Trained optimal models are used afterwards, and elevator faults are predicted by fault type prediction interval and fault time prediction interval respectively Type and elevator faults time；
Specifically, " using trained optimal models, being distinguished by fault type prediction interval and fault time prediction interval pre- Measure type and the elevator faults time of elevator faults " way are as follows:
Wherein the loss of entire model is the sum of time prediction loss and event type prediction loss.Utilize intersection entropy loss Function does event type prediction, does timestamp prediction with quadratic loss function.Specific loss function formula is as follows:
N represents node total number in above formula, and case point is indexed by l, i.e. l indicates which event, WuIndicate nerve net The parameter matrix for the event category that network model learns,It is the timestamp of next case point,Represent the history letter at time point Breath uses Gauss penalty for loss of time function part:
Wherein σ indicates unity variance, takes σ2=10；
By minimize loss and, the entire model of circuit training finally obtains optimal model.
The data of real-time update are input to model by step 10, improvement and optimization model, elevator faults information bank real-time update The accuracy of middle test model, and sophisticated model is corrected according to actual feedback situation.
Specifically, the way of " improvement and optimization model " are as follows: obtain elevator faults prediction model by training set, then utilize The fault message storehouse of real-time update obtains test set, based on the accuracy of test set examination model, and improvement and optimization model in real time.
As shown in Fig. 2, illustrating the generating mode of time series and sequence of events, two sequences are for above-mentioned steps 2 and step 3 Collaborative modeling.Since time series is made of basic information and statistical information, it is known that the background that time series carries data is known Know information, the asynchronous timestamp sequence generated at random of sequence of events carries sudden information, only cooperates with two kinds of sequences Modeling could capture more complete information intension.
Fig. 3 is elevator faults type prediction and the major architectural figure that fault time is predicted, specific introduction is such as step 5-9 institute Show.
The embodiment of the present invention is explained in detail above in conjunction with attached drawing, but the present invention is not limited to above-mentioned implementations Example, within the knowledge of a person skilled in the art, can also make without departing from the purpose of the present invention Various change out.

Claims (10)

1. the lift facility failure prediction method based on deep learning, which comprises the steps of:
Step 1 establishes real-time elevator faults information bank, according to fault message is related, information is complete, information is non-duplicate, failure letter Four principle of operation for ceasing non-artificial mistake screens pretreatment information sequence of effective elevator faults information as network, elevator event Barrier information includes failure logging information and elevator essential information, and wherein failure logging information includes: elevator faults type, failure original Cause and fault time, elevator essential information include: elevator date of manufacture, elevator present position, elevator model and elevator service life；
Step 2, build time sequence, including two kinds of features: 1) counting various types of elevator faults numbers, 2) elevator is believed substantially Breath, both characteristic bindings get up to constitute time series；
Step 3, tectonic event sequence, including two kinds of features: 1) being successively tactic elevator event according to time of failure Hinder the data record sequence of type, 2) time interval between adjacent two event of failure, both characteristic bindings get up composition event Sequence；
Step 4, building LSTM neural network；
Step 5, with double LSTM neural metwork training time serieses and sequence of events, obtain the background knowledge characterization of intensity function It is characterized with historical influence；
Step 6 merges background knowledge characterization and historical influence characterization by stratum conjunctum joint layer；
Step 7, using the intensity function learnt by double LSTM, elevator faults class is predicted by fault type prediction interval Type utilizes the penalty values of Classification Loss layer quantization class prediction；
Step 8, using the intensity function learnt by double LSTM, when predicting elevator faults by fault time prediction interval Between, utilize the penalty values for returning the prediction of loss layer quantization time；
Step 9 is based on step 7 and the continuous repetitive exercise neural network model of step 8, obtains optimal network model, then makes With trained optimal models, the type of elevator faults is predicted respectively by fault type prediction interval and fault time prediction interval With the elevator faults time；
Step 10, improvement and optimization model, elevator faults information bank real-time update, the data of real-time update are input in model and are surveyed The accuracy of die trial type, and sophisticated model is corrected according to actual feedback situation.
2. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
The way of " build time sequence " in the step 2 are as follows: it is assumed that tested elevator shares M platform, effective fault type type is total There is N kind, elevator faults time window number is n, then can be by the effectively event of the kth kind fault type of i-th elevator, j-th of time window Barrier quantity is denoted as1) with one week for time window, various failures are counted Type number2) elevator essential information, both characteristic bindings, which get up, constitutes time series data, is specifically expressed as follows institute Show:
Wherein m indicates elevator model, and d indicates the elevator date of manufacture, and l indicates elevator present position, and n indicates elevator faults time window Mouthful number, indicated by taking double horizontal lines as an example in above-mentioned expression M platform elevator in first time window, the statistics of N kind fault type The expression unit of number and elevator essential information.
3. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
The way of " tectonic event sequence " in the step 3 are as follows: 1) all elevator informations according to elevator id difference are stored separately, It is time stamp data by elevator faults time conversion, elevator faults type is arranged according to time of failure for every elevator, 2) calculate the interval time of the adjacent event of failure of every elevator, 3) fault type T and interval of timestamps I composition sequence of events are minimum Unit is specifically expressed as follows shown:
The elevator id indicates unique designation --- the elevator number of elevator.
4. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
The way of " building LSTM neural network " in the step 4 are as follows:
The specific formula of Recognition with Recurrent Neural Network variant LSTM used in the present invention is defined as follows:
it=g (Wi xt+Uiyt-1+Vict-1+bi),
ft=g (Wfxt+Ufyt-1+Vfct-1+bf),
ot=g (Wo xt+Uoyt-1+Voct+bo),
yt=ot⊙tanh(ct)
It is input gate in LSTM above, forgets door, the calculation formula of location mode and out gate, wherein itIndicate the meter of input gate Calculate formula, ftIndicate the calculation formula of forgetting door, ctIndicate location mode calculation formula, otAnd ytThe common meter for indicating out gate Calculate formula, otIt indicates that partially output is gone out using which of g function determination unit state, last ytIt indicates location mode benefit After being handled with tanh and otBeing multiplied is the determination part to be exported；Wherein U, V are respectively that neural metwork training needs to learn Parameter matrix,Array element is represented successively to be multiplied, function g uses sigmoid activation primitive,Indicate list entries,It indicates to generate implicit layer state, above formula can simplify as formula: (yt,ct)=LSTM (xtyt-1+ct-1)。
5. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
Way in the step 5 " with double LSTM neural metwork training time serieses and sequence of events " are as follows:
Specific formula is as follows:
In above formulaIndicate time series,Indicate sequence of events, wherein ziIndicate the event class in sequence of events Type, tiThe timestamp that expression event occurs, two sequences are respectively used to study background knowledge and historical influence.
6. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
The way for " background knowledge characterization being merged in the step 6 by stratum conjunctum joint layer and historical influence characterizes " are as follows: Using tanh function as joint layer Copula, the Nonlinear Mapping of point process intensity function is constructed,
Formula by the Nonlinear Mapping of the stratum conjunctum joint layer point process intensity function constructed is as follows:
7. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
" using the intensity function learnt by double LSTM, elevator event is predicted by fault type prediction interval in the step 7 Hinder type, utilize Classification Loss layer quantization class prediction penalty values " way are as follows:
Wherein, the calculation formula of failure predication main classes and subclass is as follows:
Ut=softMax (Wuet+bU)
ut=softMax (Wu[et,Ut]+bu)
In above formula, U and u respectively represent main classes and subclass, WuAnd buIndicate the model parameter matrix to be learnt in the training process； First with softMax function by etIt is out of order primary categories U as input predictiont；Then softMax function is utilized again, it will etWith main classes predicted value UtThe predicted value of subclass is calculated as input.
8. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
" using the intensity function learnt by double LSTM, elevator event is predicted by fault time prediction interval in the step 8 Downtime.Utilize the penalty values for returning the prediction of loss layer quantization time.", way specifically:
st=Wset+bs
Wherein, stIt is the corresponding timestamp of each event, WsAnd bsIndicate the model parameter matrix to be learnt in the training process.
9. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
In the step 9 " trained optimal models is used, are distinguished by fault type prediction interval and fault time prediction interval pre- Measure type and the elevator faults time of elevator faults ", way specifically:
Wherein the loss of entire model is the sum of time prediction loss and event type prediction loss.Utilize cross entropy loss function Event type prediction is done, does timestamp prediction with quadratic loss function.Specific loss function formula is as follows:
N represents node total number in above formula, and case point is indexed by l, i.e. l indicates which event, WuIndicate neural network model The parameter matrix of the event category to learn,It is the timestamp of next case point,The historical information for representing time point, for Loss of time function part uses Gauss penalty:
Wherein σ indicates unity variance, takes σ2=10；
By minimize loss and, the entire model of circuit training finally obtains optimal model.
10. the lift facility failure prediction method according to claim 1 based on deep learning, it is characterised in that:
The way of " improvement and optimization model " in the step 10 are as follows: elevator faults prediction model is obtained by training set, it is then sharp Test set is obtained with the fault message storehouse of real-time update, based on the accuracy of test set examination model, and improvement and optimization mould in real time Type.
CN201810962887.0A 2018-08-22 2018-08-22 Elevator equipment fault prediction method based on deep learning Active CN109033450B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810962887.0A CN109033450B (en) 2018-08-22 2018-08-22 Elevator equipment fault prediction method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810962887.0A CN109033450B (en) 2018-08-22 2018-08-22 Elevator equipment fault prediction method based on deep learning

Publications (2)

Publication Number Publication Date
CN109033450A true CN109033450A (en) 2018-12-18
CN109033450B CN109033450B (en) 2021-11-05

Family

ID=64627980

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810962887.0A Active CN109033450B (en) 2018-08-22 2018-08-22 Elevator equipment fault prediction method based on deep learning

Country Status (1)

Country Link
CN (1) CN109033450B (en)

Cited By (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109782603A (en) * 2019-02-03 2019-05-21 中国石油大学(华东) The detection method and monitoring system of rotating machinery coupling fault
CN109978026A (en) * 2019-03-11 2019-07-05 浙江新再灵科技股份有限公司 A kind of elevator position detection method and system based on LSTM network
CN110008928A (en) * 2019-04-16 2019-07-12 杭州再灵云梯信息科技有限公司 The analysis method to elevator accident based on deep learning method
CN110018425A (en) * 2019-04-10 2019-07-16 北京理工大学 A kind of power battery fault diagnosis method and system
CN110135241A (en) * 2019-03-27 2019-08-16 浙江新再灵科技股份有限公司 A kind of statistical analysis system of travel of elevator (lift)
CN110264270A (en) * 2019-06-20 2019-09-20 京东数字科技控股有限公司 A kind of behavior prediction method, apparatus, equipment and storage medium
CN110348513A (en) * 2019-07-10 2019-10-18 北京华电天仁电力控制技术有限公司 A kind of Wind turbines failure prediction method based on deep learning
CN110580213A (en) * 2019-09-16 2019-12-17 浪潮软件股份有限公司 Database anomaly detection method based on cyclic marking time point process
CN110996066A (en) * 2019-12-19 2020-04-10 浙江中控技术股份有限公司 Accident backtracking method and device
CN111027591A (en) * 2019-11-13 2020-04-17 西安交通大学 Node fault prediction method for large-scale cluster system
WO2020143409A1 (en) * 2019-01-07 2020-07-16 阿里巴巴集团控股有限公司 Method and device for predicting business indicators
CN111931819A (en) * 2020-07-13 2020-11-13 江苏大学 Machine fault prediction and classification method based on deep learning
CN112083244A (en) * 2020-08-30 2020-12-15 西南电子技术研究所（中国电子科技集团公司第十研究所） Integrated avionics equipment fault intelligent diagnosis system
CN112101431A (en) * 2020-08-30 2020-12-18 西南电子技术研究所（中国电子科技集团公司第十研究所） Electronic equipment fault diagnosis system
CN112232370A (en) * 2020-09-16 2021-01-15 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） Fault analysis and prediction method for engine
CN113112004A (en) * 2021-04-21 2021-07-13 上海电力大学 Time-space prediction method based on Hox process and graph neural network
CN113111585A (en) * 2021-04-15 2021-07-13 德州欧瑞电子通信设备制造有限公司 Intelligent cabinet fault prediction method and system and intelligent cabinet
CN113489744A (en) * 2021-07-27 2021-10-08 哈尔滨工业大学 Internet of things attack pattern recognition method based on hoxon multivariate process modeling
CN113793227A (en) * 2021-09-16 2021-12-14 中国电子科技集团公司第二十八研究所 Human-like intelligent perception and prediction method for social network events
CN113834184A (en) * 2021-08-18 2021-12-24 青岛海尔空调器有限总公司 Control method and device for air conditioner and server
CN113884809A (en) * 2021-09-28 2022-01-04 国网黑龙江省电力有限公司 Secondary equipment fault positioning method and system based on neural network
CN114266201A (en) * 2022-03-01 2022-04-01 杭州市特种设备检测研究院(杭州市特种设备应急处置中心) Self-attention elevator trapping prediction method based on deep learning
CN114626467A (en) * 2022-03-17 2022-06-14 湖南优湖科技有限公司 Feature cross elevator trapping time series prediction model construction method based on deep learning, obtained model and prediction method
WO2022121932A1 (en) * 2020-12-10 2022-06-16 东北大学 Adaptive deep learning-based intelligent forecasting method, apparatus and device for complex industrial system, and storage medium
CN115385209A (en) * 2021-05-25 2022-11-25 太原理工大学 Alarm and password removing method for abnormal opening of elevator door
CN117034157A (en) * 2023-10-08 2023-11-10 广州健新科技有限责任公司 Hydropower equipment fault identification method and system combining multimodal operation data
CN117131457A (en) * 2023-10-26 2023-11-28 杭州海兴泽科信息技术有限公司 AI model-based electric power big data acquisition and processing method and system

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2003097506A1 (en) * 2002-05-14 2003-11-27 Otis Elevator Company Neural network detection of obstructions within and motion toward elevator doors
CN101833596A (en) * 2010-03-31 2010-09-15 桂林电子科技大学 Two-damage quantitative identification method of rectangular and round section beam structure
CN102491140A (en) * 2011-12-27 2012-06-13 阳西县电梯配件有限公司 Characteristic-signal-based elevator safety checking device and elevator safety checking method
CN105844050A (en) * 2016-04-12 2016-08-10 吉林大学 Time correlation based replacement time calculation method for numerical control machine tool components
CN106125713A (en) * 2016-06-14 2016-11-16 西安交通大学 The assessment of reliability growth and Forecasting Methodology in the case of a kind of Interval censoring
CN106516923A (en) * 2016-08-31 2017-03-22 江苏鸿信系统集成有限公司 Elevator running failure prediction method based on technology of Internet of Things
CN107992693A (en) * 2017-12-08 2018-05-04 中国北方发动机研究所(天津) A kind of cylinder head reliable life evaluation method

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2003097506A1 (en) * 2002-05-14 2003-11-27 Otis Elevator Company Neural network detection of obstructions within and motion toward elevator doors
CN101833596A (en) * 2010-03-31 2010-09-15 桂林电子科技大学 Two-damage quantitative identification method of rectangular and round section beam structure
CN102491140A (en) * 2011-12-27 2012-06-13 阳西县电梯配件有限公司 Characteristic-signal-based elevator safety checking device and elevator safety checking method
CN105844050A (en) * 2016-04-12 2016-08-10 吉林大学 Time correlation based replacement time calculation method for numerical control machine tool components
CN106125713A (en) * 2016-06-14 2016-11-16 西安交通大学 The assessment of reliability growth and Forecasting Methodology in the case of a kind of Interval censoring
CN106516923A (en) * 2016-08-31 2017-03-22 江苏鸿信系统集成有限公司 Elevator running failure prediction method based on technology of Internet of Things
CN107992693A (en) * 2017-12-08 2018-05-04 中国北方发动机研究所(天津) A kind of cylinder head reliable life evaluation method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
P PANITSRISIT: "Sensor System for Fault Detection Identification and Accommodation of Elevator of UAV", 《SICE ANNUAL CONFERENCE 2011》 *
王鑫 等: "基于LSTM循环神经网络的故障时间序列预测", 《北京航空航天大学学报》 *

Cited By (38)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020143409A1 (en) * 2019-01-07 2020-07-16 阿里巴巴集团控股有限公司 Method and device for predicting business indicators
CN109782603A (en) * 2019-02-03 2019-05-21 中国石油大学(华东) The detection method and monitoring system of rotating machinery coupling fault
CN109978026B (en) * 2019-03-11 2021-03-09 浙江新再灵科技股份有限公司 Elevator position detection method and system based on LSTM network
CN109978026A (en) * 2019-03-11 2019-07-05 浙江新再灵科技股份有限公司 A kind of elevator position detection method and system based on LSTM network
CN110135241A (en) * 2019-03-27 2019-08-16 浙江新再灵科技股份有限公司 A kind of statistical analysis system of travel of elevator (lift)
CN110135241B (en) * 2019-03-27 2021-08-31 浙江新再灵科技股份有限公司 Statistical analysis system of elevator stroke
CN110018425A (en) * 2019-04-10 2019-07-16 北京理工大学 A kind of power battery fault diagnosis method and system
CN110008928A (en) * 2019-04-16 2019-07-12 杭州再灵云梯信息科技有限公司 The analysis method to elevator accident based on deep learning method
CN110264270A (en) * 2019-06-20 2019-09-20 京东数字科技控股有限公司 A kind of behavior prediction method, apparatus, equipment and storage medium
CN110348513A (en) * 2019-07-10 2019-10-18 北京华电天仁电力控制技术有限公司 A kind of Wind turbines failure prediction method based on deep learning
CN110580213A (en) * 2019-09-16 2019-12-17 浪潮软件股份有限公司 Database anomaly detection method based on cyclic marking time point process
CN111027591A (en) * 2019-11-13 2020-04-17 西安交通大学 Node fault prediction method for large-scale cluster system
CN111027591B (en) * 2019-11-13 2022-07-12 西安交通大学 Node fault prediction method for large-scale cluster system
CN110996066A (en) * 2019-12-19 2020-04-10 浙江中控技术股份有限公司 Accident backtracking method and device
CN110996066B (en) * 2019-12-19 2021-12-24 浙江中控技术股份有限公司 Accident backtracking method and device
CN111931819A (en) * 2020-07-13 2020-11-13 江苏大学 Machine fault prediction and classification method based on deep learning
CN112083244B (en) * 2020-08-30 2022-10-28 西南电子技术研究所（中国电子科技集团公司第十研究所） Integrated intelligent diagnosis system for faults of avionic equipment
CN112083244A (en) * 2020-08-30 2020-12-15 西南电子技术研究所（中国电子科技集团公司第十研究所） Integrated avionics equipment fault intelligent diagnosis system
CN112101431A (en) * 2020-08-30 2020-12-18 西南电子技术研究所（中国电子科技集团公司第十研究所） Electronic equipment fault diagnosis system
CN112232370A (en) * 2020-09-16 2021-01-15 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） Fault analysis and prediction method for engine
WO2022121932A1 (en) * 2020-12-10 2022-06-16 东北大学 Adaptive deep learning-based intelligent forecasting method, apparatus and device for complex industrial system, and storage medium
CN113111585A (en) * 2021-04-15 2021-07-13 德州欧瑞电子通信设备制造有限公司 Intelligent cabinet fault prediction method and system and intelligent cabinet
CN113112004A (en) * 2021-04-21 2021-07-13 上海电力大学 Time-space prediction method based on Hox process and graph neural network
CN115385209B (en) * 2021-05-25 2023-08-29 太原理工大学 Alarm and password release method for abnormal opening of elevator door
CN115385209A (en) * 2021-05-25 2022-11-25 太原理工大学 Alarm and password removing method for abnormal opening of elevator door
CN113489744B (en) * 2021-07-27 2022-05-10 哈尔滨工业大学 Internet of things attack pattern recognition method based on hoxophone multivariate process modeling
CN113489744A (en) * 2021-07-27 2021-10-08 哈尔滨工业大学 Internet of things attack pattern recognition method based on hoxon multivariate process modeling
CN113834184A (en) * 2021-08-18 2021-12-24 青岛海尔空调器有限总公司 Control method and device for air conditioner and server
CN113793227B (en) * 2021-09-16 2023-10-31 中国电子科技集团公司第二十八研究所 Intelligent human-like perception and prediction method for social network event
CN113793227A (en) * 2021-09-16 2021-12-14 中国电子科技集团公司第二十八研究所 Human-like intelligent perception and prediction method for social network events
CN113884809A (en) * 2021-09-28 2022-01-04 国网黑龙江省电力有限公司 Secondary equipment fault positioning method and system based on neural network
CN114266201B (en) * 2022-03-01 2022-07-22 杭州市特种设备检测研究院(杭州市特种设备应急处置中心) Self-attention elevator trapping prediction method based on deep learning
CN114266201A (en) * 2022-03-01 2022-04-01 杭州市特种设备检测研究院(杭州市特种设备应急处置中心) Self-attention elevator trapping prediction method based on deep learning
CN114626467A (en) * 2022-03-17 2022-06-14 湖南优湖科技有限公司 Feature cross elevator trapping time series prediction model construction method based on deep learning, obtained model and prediction method
CN117034157A (en) * 2023-10-08 2023-11-10 广州健新科技有限责任公司 Hydropower equipment fault identification method and system combining multimodal operation data
CN117034157B (en) * 2023-10-08 2024-01-12 广州健新科技有限责任公司 Hydropower equipment fault identification method and system combining multimodal operation data
CN117131457A (en) * 2023-10-26 2023-11-28 杭州海兴泽科信息技术有限公司 AI model-based electric power big data acquisition and processing method and system
CN117131457B (en) * 2023-10-26 2024-01-26 杭州海兴泽科信息技术有限公司 AI model-based electric power big data acquisition and processing method and system

Also Published As

Publication number Publication date
CN109033450B (en) 2021-11-05

Similar Documents

Publication Publication Date Title
CN109033450A (en) Lift facility failure prediction method based on deep learning
US11270579B2 (en) Transportation network speed foreeasting method using deep capsule networks with nested LSTM models
CN109492830B (en) Mobile pollution source emission concentration prediction method based on time-space deep learning
Mangalathu et al. Deep learning-based classification of earthquake-impacted buildings using textual damage descriptions
Toledo et al. Statistical validation of traffic simulation models
Zeng et al. Development of recurrent neural network considering temporal‐spatial input dynamics for freeway travel time modeling
CN104376716B (en) Method for dynamically generating bus timetables on basis of Bayesian network models
CN107786369A (en) Based on the perception of IRT step analyses and LSTM powerline network security postures and Forecasting Methodology
CN110009046B (en) Urban community safety prediction method based on big data
CN114266200B (en) Nitrogen dioxide concentration prediction method and system
CN110210169A (en) A kind of shield machine failure prediction method based on LSTM
CN112508265A (en) Time and activity multi-task prediction method and system for business process management
CN112508305A (en) Public place entrance pedestrian flow prediction method based on LSTM
CN113516304A (en) Space-time joint prediction method and device for regional pollutants based on space-time graph network
CN114662791A (en) Long time sequence pm2.5 prediction method and system based on space-time attention
Liu et al. Comprehensive functional resilience assessment methodology for bridge networks using data-driven fragility models
Kowalski et al. PM10 forecasting through applying convolution neural network techniques
Liu et al. Post-earthquake assessment model for highway bridge networks considering traffic congestion due to earthquake-induced bridge damage
CN111524349B (en) Context feature injected multi-scale traffic flow prediction model establishing method and using method
KR102589525B1 (en) Method and apparatus for providing earthquake fragility function, learning method for providing earthquake fragility function
CN117010683A (en) Operation safety risk prediction method based on hybrid neural network and multiple agents
CN117494034A (en) Air quality prediction method based on traffic congestion index and multi-source data fusion
CN116894364A (en) Post-earthquake damage rapid prediction method for high-speed railway track bridge system
CN114401135B (en) Internal threat detection method based on LSTM-Attention user and entity behavior analysis technology
CN116523187A (en) Engineering progress monitoring method and system based on BIM

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