WO2022141213A1 - 一种智慧城市智轨车辆故障基因预测方法及系统 - Google Patents

一种智慧城市智轨车辆故障基因预测方法及系统 Download PDF

Info

Publication number
WO2022141213A1
WO2022141213A1 PCT/CN2020/141464 CN2020141464W WO2022141213A1 WO 2022141213 A1 WO2022141213 A1 WO 2022141213A1 CN 2020141464 W CN2020141464 W CN 2020141464W WO 2022141213 A1 WO2022141213 A1 WO 2022141213A1
Authority
WO
WIPO (PCT)
Prior art keywords
fault
dna sequence
prediction
data
esns
Prior art date
Application number
PCT/CN2020/141464
Other languages
English (en)
French (fr)
Inventor
刘辉
杨睿
李燕飞
夏雨
Original Assignee
中南大学
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 中南大学 filed Critical 中南大学
Priority to PCT/CN2020/141464 priority Critical patent/WO2022141213A1/zh
Priority to AU2020471210A priority patent/AU2020471210B2/en
Publication of WO2022141213A1 publication Critical patent/WO2022141213A1/zh

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
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B25/00—ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
    • G16B25/10—Gene or protein expression profiling; Expression-ratio estimation or normalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]

Definitions

  • the invention relates to the field of vehicle fault detection, in particular to a method and system for predicting fault genes of smart city smart rail vehicles.
  • ART Autonomous Rail Rapid Transit
  • the patent with publication number CN203732247U adopts the method of physical pressure spring switch to diagnose the unit fault signal.
  • This method has certain application limitations, and the fault detection method cannot be adaptively adjusted according to the train conditions.
  • the research gap in fault early warning is also an urgent problem to be solved.
  • the technical problem to be solved by the present invention is to provide a method and system for predicting fault genes of smart city smart rail vehicles in view of the deficiencies of the prior art, so as to improve the accuracy of fault early warning.
  • the technical solution adopted in the present invention is: a method for predicting fault genes of intelligent rail vehicles in a smart city, comprising the following steps:
  • Predictive modeling based on DNA coding can deeply develop the potential information in the vibration data of train components, resulting in more accurate fault prediction.
  • it also includes:
  • the resulting predictive model can help industry managers predict the failure of urban smart rail train equipment, so that it can be repaired in advance before the failure occurs.
  • step 2) the specific implementation process of encoding the vibration data into a DNA sequence includes:
  • A) select the g-th column sample of the original vibration signal X collected, assign the g-th column sample to the initial DNA helical sequence data matrix X h(0) , and the matrix after the assignment is represented as X g ;
  • the vibration data encoding based on the continuous projection method can convert the original vibration signal into a U-dimensional feature vector expressed by the four base elements of A, T, C, and G, so as to avoid losing effective information.
  • Feature extraction for base pairs of coded gene sequences can find out the most representative features, and use low-dimensional data to express as much high-dimensional information as possible, which can also avoid model overfitting in the modeling process
  • step 3 includes:
  • E) Increase the number of iterations by 1, take the new pool layer number ⁇ m+1 and pool matrix spectral radius ⁇ m+1 as the input of the objective optimization function of the multi-objective gray wolf optimization algorithm, and return to step C) until the multi-objective gray
  • the target optimization function value of the wolf optimization algorithm reaches the expected accuracy or completes the set number of iterations, completes the training of the ESNs deep echo state network, and obtains the optimal parameters ⁇ optimal and ⁇ optimal .
  • the optimal parameters ⁇ optimal and ⁇ optimal correspond to the ESNs
  • the deep echo state network model is the prediction model.
  • the ESNs deep echo state network model has excellent data fitting ability.
  • the ESNs deep echo state network model with parameters optimized by the multi-objective gray wolf optimization algorithm has a smaller prediction error and can more accurately predict vehicle failures.
  • ⁇ is the number of reservoir layers
  • ⁇ is the spectral radius of the reservoir matrix
  • V t is the true value of the DNA sequence, is the average of all true values
  • N is the length of the DNA sequence, and 1 ⁇ t ⁇ N
  • CT indicates the failure of the car body
  • ZXJ indicates the failure of the bogie
  • QY indicates the failure of the traction drive control system
  • ZD indicates the failure of the braking system
  • LJ indicates the failure of the vehicle end connection device
  • SL indicates the failure of the current receiving device
  • SB indicates the failure of the vehicle interior equipment and cab equipment
  • NSE and KGE are indicators used to measure the stability of the model. Setting the objective function for optimization based on these two indicators can make the prediction model more robust.
  • It also includes: using the pre-determined candidate vehicle component fault gene V s as the input of the clustering model to build a template library. Building a template library can help relevant industry personnel compare the similarities and differences between current faults and historical faults, so as to take more accurate maintenance operations.
  • the specific implementation process of building a template library includes:
  • Step 1 Use the pre-determined candidate vehicle component fault gene V s obtained by the continuous projection method as the input of the random adjacency embedding algorithm to obtain the conditional probability p j
  • i of points vi and v j is minimized to obtain the minimized conditional probability p j
  • Step 2 Calculate the minimum value p ij of the difference between high and low dimensional conditional probability according to the conditional probability minimization result, Minimize the cost function L by gradient descent: get the optimal solution the optimal solution Output as the clustering result of the tSNE clustering algorithm; the clustering result corresponds to the template library template of the ART urban smart rail vehicle:
  • CT, ZXJ, QY, ZD, LJ, SL, SB are the fault categories in the DNA sequence template library; CT: vehicle body fault; ZXJ: bogie fault; QY: traction drive control system fault; ZD: braking system fault; LJ: vehicle end connection device fault; SL: current receiving device fault; SB: vehicle interior equipment and cab equipment fault; n represents the number of data samples, and KL represents the divergence.
  • the combination of continuous projection method and t-SNE clustering method avoids the disadvantageous situation that a large amount of effective information about vehicle failures is lost, and soft clustering can obtain more reliable template library information.
  • step 4 it also includes:
  • the method of the present invention also includes:
  • the specific implementation process includes: performing binary inverse coding conversion on the prediction results output by the prediction model, wherein the binary inverse The combined base pair of adenine and thymine in the prediction result after code conversion is decoded and corresponds to the number 0, that is, the degree of equipment failure does not reach the warning line threshold, and the combined base pair of guanine and Ccytosine is decoded and corresponds to Number 1, that is, the degree of equipment failure has reached the warning line threshold, and it must be repaired.
  • the present invention also provides a smart city smart rail vehicle fault gene prediction system, which includes computer equipment; the computer equipment is configured or programmed to perform the steps of the above method.
  • the present invention has the following beneficial effects:
  • the present invention provides an accurate fault prediction method based on a DNA sequence template library on the basis of the existing fault diagnosis technology of intelligent rail vehicles.
  • the data acquisition module combining wireless sensor network and high and low frequency vibration measuring instruments can collect a large number of historical fault signals, and the data encoding module of multi-source vibration signals converts vibration signals into coded gene sequences, which can encode DNA sequences of base pairs
  • the feature extraction module can screen out the pre-determined candidate vehicle component fault genes, and the construction of the DNA sequence template library module can help the relevant industry personnel compare the similarities and differences between newly detected faults and historical faults, and can encode the DNA helix sequence deep learning fault early warning modeling
  • the module can predict potential failures of train components.
  • the multi-objective optimization-based DNA helix sequence prediction strategy module can improve the accuracy of fault warning.
  • the DNA sequence helix decoding and the fault visualization module of the virtual template library help maintenance personnel to quickly identify fault types.
  • the present invention builds a DNA sequence template library that can encode fault modules, corresponding to the seven major components of urban smart rail vehicles (car body, bogie, traction drive control system, braking system, vehicle end connection device, current receiving device, Vehicle interior equipment and cab equipment).
  • the fault template library is used as a matching template for virtual faults, which provides an accurate direction for training a reliable fault early warning model.
  • An accurate and complete fault information database is more conducive to the staff to compare the similarities and differences of the faults of the new and old equipment of the autonomous rail train to carry out fault maintenance.
  • the present invention proposes a modeling method for fault diagnosis and multi-fault prediction matching of autonomous rail trains.
  • Vibration sensors are installed on the major components of the smart rail train to collect real-time vibration data signals, and through the wireless sensor network (WSN). ) for transmission, and establish deep echo state networks (ESNs) for multi-objective optimal prediction of equipment failures, which greatly improves the accuracy of failure prediction.
  • WSN wireless sensor network
  • ESNs deep echo state networks
  • a complete system framework is built around data collection, original signal helical coding and decoding, gene signal conversion, gene sequence feature extraction, establishment of a DNA sequence template library for failure modules, failure prediction, etc.
  • the model can be embedded into the Hadoop big data platform for training to improve the training speed.
  • FIG. 1 is a schematic diagram of a method according to an embodiment of the present invention.
  • Step 1 Collection of historical failure data of new smart rail train components
  • the invention uses high and low frequency sensors and electric sensors to collect historical vibration information data of various types of urban smart rail vehicle components, and the technology change greatly reduces the cost of popularization and application of sensors.
  • the wireless sensor network plays an important role.
  • the information acquisition modules involved in step 1 include the vibration amplitude acquisition module of vehicle components, Vibration frequency acquisition module and vibration period acquisition module.
  • the collected information includes the vibration amplitude A, frequency f, period T and other signals of vehicle components, which need to be filtered through a filter, and finally the original vibration signal X is obtained.
  • low frequency vibration measurement relative type moving coil type electric sensor
  • high frequency vibration measurement inertia type moving coil type electric sensor.
  • Electric sensors can perform vibration tests on important components such as some civilian industrial vehicles.
  • Step 2 Code conversion of DNA helix sequence data from multi-source vibrational signals
  • the collected multi-source vibration data needs to be encoded into DNA sequences.
  • the encoded vibration signals have more obvious and distinguishable characteristics, which facilitates subsequent prediction work.
  • the base data of DNA sequence is a high-dimensional or ultra-high-dimensional matrix in mathematical expression after being arranged.
  • robust and robust dimensionality reduction processing is required.
  • the Continuous Projection Algorithm (SPA) processing of DNA helical sequences can achieve rapid dimensionality reduction to solve the collinearity problem, with few parameters to be adjusted and the idea is simple (see Soares SF C, Gomes A A, Araujo M C U, et al.
  • the sample of the gth column of the collected original vibration signal X is selected and assigned to the initial DNA helix sequence data matrix X h(0) , and the assigned matrix is expressed as X g .
  • the next step is to calculate the orthogonal projection of the assigned DNA helix sequence data matrix X g and the maximum projection value matrix X h(z-1) in the subspace:
  • F is the projection operator, that is, the projection of the initial helical sequence orthogonal to other helical sequences; h(z-1) is the maximum projection value; h(0) is the minimum projection value, and so on, for a total of Z Projection; where the maximum projection value is normalized to be G, that is, vertical projection, and the minimum projection is 0, that is, parallel projection. Starting from the minimum, the projection angle changes by a value ⁇ , and the projection value increases.
  • z is the ordinal number of the projected value.
  • the data matrix set Y with dimension U obtained after the continuous projection algorithm (SPA) dimension reduction process can be expressed as follows:
  • the selection of the initial helical sequence X h(0) is very critical, which directly affects the accuracy of the algorithm.
  • the algorithm can be regarded as a matrix projection in essence. In the present invention, it corresponds to the transformation of the dimension of the data type, and the vibration signal is mapped into a set of low-dimensional gene expression, and these genes represent the expression of the faulty component.
  • step 2 the equipment vibration signal X collected in step 1 needs to be transformed into a gene sequence.
  • the dimension reduction dimension U is defined according to the degree to which information needs to be preserved.
  • the continuous projection method (SPA) reduces the dimension of the original vibration data to obtain Y, and Y is used as a set of data, and the amplitude of the data in the matrix varies from high to low. , according to the empirical threshold, the data is normalized.
  • Y is roughly divided into 4 categories according to the proportion of 25% of one base, namely A, T, C, G, after normalization
  • the data whose amplitude is [0.75, 1] is defined as class A, and then the final proportion of A, T, C, and G is adjusted according to the ratio of the number of sensors in each part of the device.
  • the vibration sample data after the continuous projection algorithm (SPA) dimensionality reduction process is defined as the U-dimensional feature vector expressed by the four base elements of A, T, C, and G, that is, the transformed encoded gene required for the subsequent steps.
  • Sequence signal the purpose here is to divide the data into 4 categories.
  • B 1 , B 2 , B 3 , and B 4 are used to replace the four bases "A, T, C, G" for expression.
  • the preprocessed vibration signal is converted into an encoded gene sequence.
  • Step 3 Base pair feature extraction of the encoded gene sequence
  • the DNA sequence after coding conversion does not yet have the characteristics for efficient and high-precision prediction. Feature extraction is required to extract the deep expression of equipment failures, and arrange and combine them to form predictable DNA sequences.
  • step 1 selection of historical fault data
  • step 2 coding conversion of DNA helical sequence data
  • step 2 coding conversion of DNA helical sequence data
  • the DNA sequence feature extraction of the faulty parts of the independent autonomous rail train is carried out by calculating the content, position and transition probability of the bases in the transformed gene sequence.
  • n i is defined as the number of occurrences of a single base point B i in the DNA sequence S.
  • n ij is defined as the occurrence of the base pair B i B j in the DNA sequence S. number of times.
  • the specific calculation formula is:
  • W ij can be regarded as the probability of base B i being transferred to base B j , that is, the base transfer probability vector.
  • the content vectors are C 1 , C 2 , C 3 ,...,C U .
  • the base position ratio D i is obtained by conversion, and the mathematical expression is as follows:
  • the position ratio vectors are D 1 , D 2 , D 3 ,...,D U .
  • a usable U-dimensional vector can be obtained.
  • These feature vectors are defined as pre-decided candidate vehicle component failure genes.
  • Step 4 Build the DNA sequence template library of the faulty module
  • the candidate fault gene feature vector extracted in step 3 is input into the (t-distributed random neighborhood embedding) t-SNE clustering model in this link, and the DNA sequence template library of the fault module is established through fine clustering division.
  • the template library corresponds to the seven major sections of the urban smart rail vehicle, which are the car body (CT) library, the bogie (ZXJ) library, the traction drive control system (QY) library, the braking system (ZD) library, and the vehicle end connection device. (LJ) library, current receiving device (SL) library, vehicle interior equipment and cab equipment (SB) library.
  • CT car body
  • ZXJ the bogie
  • QY traction drive control system
  • ZD braking system
  • LJ current receiving device
  • SB vehicle interior equipment
  • the continuous projection algorithm SPA
  • the continuous projection algorithm is first used to reduce the vibration signal to a small and medium-sized space.
  • the multi-dimensional space U is expressed by multi-dimensional base features, and finally the t-SNE clustering method is used to obtain the final clustering result, which can achieve the effect of soft clustering.
  • Each clustering result corresponds to the failure of a component, and the clustered results are sent to the predictor model in step 5 for training, and then the DNA sequence template is used for secondary detailed division.
  • t-SNE is a nonlinear dimensionality reduction algorithm that can explore high-dimensional data.
  • the DNA sequence clustering method of the vehicle fault module t-SNE is as follows:
  • the data is transformed by random adjacency embedding (SNE).
  • SNE random adjacency embedding
  • the high-dimensional Euclidean distance between the data is transformed and expressed as a similar conditional probability.
  • V i of the candidate vehicle component fault gene The mathematical calculation of the conditional probability p j
  • Vi , V j are the data points in the DNA sequence S, and ⁇ i is the Gaussian variance centered on the data points Vi , V j .
  • s is the sequence number of the pre-determined candidate vehicle component fault gene.
  • the optimal solution is several clusters that can be expressed as CT, ZXJ, QY, ZD, LJ, SL and SB.
  • y is the number of iterations in the iterative process
  • y max is the maximum total number of iterations
  • ⁇ is the learning rate
  • ⁇ (y) is the learning momentum
  • the template library corresponds to the fault type, and a gene feature corresponds to the fault of a component. Finally, the system issues a diagnostic warning report.
  • the final optimal solution The clustering results can be expressed as several clusters of CT, ZXJ, QY, ZD, LJ, SL, and SB, and visualized as a clustering template of the DNA sequences of seven large parts of ART urban smart rail vehicles.
  • the template library expression corresponding to the clustering results is as follows:
  • CT car body
  • ZXJ bogie
  • QY traction drive control system
  • ZD braking system
  • LJ vehicle end connection device
  • SL current receiving device
  • SB vehicle interior equipment and cab equipment.
  • Step 5 Multi-objective optimization deep learning fault warning modeling that encodes DNA helical sequences
  • the pre-determined candidate vehicle component fault genes are normalized and input into the model for fault prediction training of urban smart rail train equipment.
  • the specific modeling process is as follows:
  • NSE Nash Sutcliffe Efficiency
  • KGE Klingupta Efficiency
  • the number of layers ⁇ of the deep echo state network reservoir and the radius ⁇ of the matrix spectrum of each layer reservoir are set as variables to be optimized.
  • the reservoir node of the deep echo state network is initially set to 15, and the input and output layers of each layer of the network are relative to each other, and then the deep feature representation of the encoded data is learned.
  • Model training The training set, the initial number of layers ⁇ 0 of the ESNs deep echo state network model reservoir and the initial radius ⁇ 0 of the matrix spectrum of each layer reservoir are used as the input of the ESNs deep echo state network model to have the number of reservoir layers ⁇ 0
  • the ESNs deep echo state network model with m and the reservoir matrix spectral radius ⁇ m is used as the output to train the ESNs deep echo state network model.
  • the multi-objective gray wolf optimization algorithm is embedded in the leader selection mechanism and the archive storage mechanism to improve the convergence ability.
  • the test set, the number of reservoir layers ⁇ m and the spectral radius ⁇ m of the reservoir matrix are used as the input of the objective optimization function of the multi-objective gray wolf optimization algorithm, and the value of the objective optimization function is calculated; where m represents the current number of iterations, 0 ⁇ m ⁇ 200.
  • ⁇ is the number of reservoir layers
  • ⁇ is the spectral radius of the reservoir matrix
  • V t is the true value of the DNA base sequence
  • N is the length of the DNA sequence
  • ZXJ indicates the bogie Fault
  • QY means traction drive control system fault
  • ZD means braking system fault
  • LJ means vehicle end connecting device fault
  • SL means current receiving device fault
  • SB means vehicle interior equipment and cab equipment fault.
  • the number of iterations is increased by 1, and the new pool layer number ⁇ m+1 and pool matrix spectral radius ⁇ m+1 are used as the input of the objective optimization function of the multi-objective gray wolf optimization algorithm, and return to step 4) until the multi-objective gray
  • the target optimization function value of the wolf optimization algorithm reaches the expectation or the set number of iterations is completed, the ESNs deep echo state network training is completed, and the optimal parameters ⁇ optimal and ⁇ optimal are obtained, and the optimal parameters ⁇ optimal and ⁇ optimal correspond to the depth of ESNs
  • the echo state network model is the prediction model.
  • the fault category belongs to a sub-fault in a certain fault category in the template library, the fault category is classified into the template library of the fault and recorded as the old fault If the fault category does not belong to any category in the template library, update the template library, directly add the prediction result to the template library, and mark it as a new fault
  • the DNA sequence template library guides the direction for the subsequent training of the prediction model.
  • Step 6 Decoding of DNA Helix Sequences and Visualization of Failure of Virtual Template Library
  • the original vibration data is subjected to coding transformation and feature extraction steps, the purpose of which is to extract the depth feature expression of the original sequence with inconspicuous original characteristics, and then input the easily distinguishable depth feature sequence into the Go to the prediction model for training.
  • the sequence results predicted by the trained model are still deep feature expressions.
  • the prediction results of each DNA fragment are connected end to end to form a complete base sequence code, and the corresponding faults correspond to the virtual DNA template library. specific type in .
  • the representation of the results is not concrete, so it is necessary to decode the DNA sequence and visualize the failure to restore to the predicted data whose data type corresponds to the original vibration data.
  • the prediction model obtained in step 5 is used to predict the vibration data collected in real time, and then the prediction results output by the prediction model (the code of the ART urban smart rail vehicle failure prediction output result)
  • the mechanism is based on the base coding system in step 3, so the composition of its sequence is based on A, T, G, C bases and is obtained through deep learning prediction modeling) to perform binary inverse coding conversion.
  • the prediction output result is decoded to display the 0/1 state, that is, the visualization of the prediction result is completed, so as to realize timely warning.
  • the binding base pairs of A (adenine) and T (thymine) correspond to the number 0 after decoding, that is, the degree of equipment failure does not reach the warning line threshold, and the binding bases of G (guanine) and C (cytosine) After decoding, the corresponding number 1, that is, the equipment failure degree has reached the warning line threshold, and it must be repaired.
  • the fault early warning model can also provide a reliable guarantee for the safe and stable operation of ART urban smart rail vehicles.
  • the DNA sequence data storage method provides infinite possibilities for information transmission, reception and storage, and the storage time can fully meet the needs of information use in the era of big data.
  • Step 7 Distributed System Infrastructure Embedding
  • the module can be embedded in the distributed system infrastructure to speed up model training and self-learning update speed, so as to meet application requirements to a greater extent .
  • Available distributed system infrastructures include MapReduce, Apache Spark, Hadoop, etc. (see Dittrich J,Quiané-Ruiz J A.Efficient big data processing in Hadoop MapReduce[J].Proceedings of the VLDB Endowment,2012,5(12):2014 -2015.).
  • MapReduce MapReduce
  • Apache Spark Hadoop
  • Hadoop etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Economics (AREA)
  • Genetics & Genomics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Biotechnology (AREA)
  • Development Economics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Game Theory and Decision Science (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

一种智慧城市智轨车辆故障基因预测方法及系统，采集列车部件的振动数据X h (0)＝[e 1，e 2，e 3，...，e n]∈R，其中，e 1，e 2，...，e n代表列车上每一个采样点的振动信息；将所述振动数据编码为DNA序列，提取所述DNA序列的特征，并排列组合以形成可预测的DNA序列，即候选车辆部件故障基因；利用所述候选车辆部件故障基因训练ESNs深度回声状态网络，得到预测模型。可以准确预测车辆故障。

Description

一种智慧城市智轨车辆故障基因预测方法及系统 技术领域
本发明涉及车辆故障检测领域，具体是一种智慧城市智轨车辆故障基因预测方法及系统。
背景技术
21世纪10年代以来，交通拥堵问题、能源危机问题、环境污染问题、土地短缺问题等对城市交通提出了挑战。为了在舒适性、安全性、节能环保等方面做的更好，新一代的城市客运Autonomous Rail Rapid Transit(ART)自主轨道列车出现在人们的视野当中。ART被用于解决大城市郊区和小城市地区的交通问题，它不依赖于现有的铁轨，而是可以通过对地通信和线控专项技术来实现自主无轨道的自动驾驶，大大降低了人力物力的损耗。然而，对于ART智轨列车的故障检测手段尚存在很大的空白，目前大都采用类似于城轨列车的原始检测方法。如公开号为CN203732247U的专利采用物理压力弹簧开关的方法进行单元故障信号的诊断。该方法存在一定的运用局限性，且故障检测手段不能根据列车车况自适应的调整，故障预警方面存在研究空白也是亟需解决的问题。
发明内容
本发明所要解决的技术问题是，针对现有技术不足，提供一种智慧城市智轨车辆故障基因预测方法及系统，提高故障预警准确性。
为解决上述技术问题，本发明所采用的技术方案是：一种智慧城市智轨车辆故障基因预测方法，包括以下步骤：
1)采集列车部件的振动数据X h(0)＝[e 1,e 2,e 3,...,e n]∈R，其中，e 1,e 2,...,e n代表列车上每一个采样点的振动信息；n表示采样点个数；
2)将所述振动数据编码为DNA序列，提取所述DNA序列的特征，并排列组合以形成可预测的DNA序列，即候选车辆部件故障基因；
3)利用所述候选车辆部件故障基因训练ESNs深度回声状态网络，得到预测模型；
基于DNA编码的预测建模可以深度开发列车部件振动数据中的潜在信息，从而得到更精准的故障预测。
优选地，还包括：
4)根据实时采集的振动数据，利用所述预测模型预测车辆故障。所得预测模型可以帮助行业管理人员预测城市智轨列车设备出现的故障，从而在故障发生之前提前维修。
步骤2)中，将所述振动数据编码为DNA序列的具体实现过程包括：
A)选中采集的原始振动信号X的第g列样本，将所述第g列样本赋值到初始的DNA螺旋序列数据矩阵X h(0)，赋值后的矩阵表示为X g；
B)计算赋值后的DNA螺旋序列数据矩阵X g与最大投影值矩阵X h(z-1)在子空间中的正交投影，得到维度为U的数据矩阵集合Y；z是投影值的序号；h(z-1)是最大投影值；最大投影值正规化后为G，即垂直投影，最小投影值h(0)为0，即平行投影，从最小值开始投影角度每变化一个数值Υ，投影值增大
Figure PCTCN2020141464-appb-000001
Z为投影值数量；
C)将所述数据矩阵集合Y划分为由A，T，C，G四种碱基元素表达的U维特征向量；将A，T，C，G整合为DNA序列S＝S 1,S 2,S 3,...,S N；其中，N为DNA序列长度。
基于连续投影法降维的振动数据编码可将原始振动信号转换为由A，T，C，G四种碱基元素表达的U维特征向量，避免丢失有效信息。
步骤2)中，候选车辆部件故障基因V s表示为：V s＝(W 11,W 12,...,W UU,C 1,...,C U,D 1,...,D U)；其中，碱基B i转移到碱基B j的概率
Figure PCTCN2020141464-appb-000002
n i为单个碱基点B i在DNA序列S中出现的次数；B i为DNA序列S中第i个数据点位置的碱基；1≤i≤U；U指碱基元素所代表特征向量的维度；N为DNA序列S的长度；n ij为碱基对B iB j在DNA序列S中出 现的次数；碱基含量
Figure PCTCN2020141464-appb-000003
碱基位置比
Figure PCTCN2020141464-appb-000004
DNA序列S中碱基B i出现的位置标记为S i，s i是S i中的值。对可编码基因序列碱基对进行特征提取可以找出其中最具代表性的特征，用低维度的数据表达尽可能多的高维度信息，这也可以避免建模过程中模型过拟合。
步骤3)的具体实现过程包括：
A)将车辆部件故障基因V s随机划分为训练集和测试集；初始化多目标灰狼优化算法的迭代次数m、预期精度；
B)将所述训练集、ESNs深度回声状态网络模型储蓄池的初始层数θ 0和每一层储蓄池矩阵谱的初始半径κ 0作为ESNs深度回声状态网络模型的输入，以具有储蓄池层数θ m和储蓄池矩阵谱半径κ m的ESNs深度回声状态网络模型作为输出，训练ESNs深度回声状态网络模型；
C)将所述测试集、储蓄池层数θ m和储蓄池矩阵谱半径κ m作为多目标灰狼优化算法两个目标优化函数的输入，计算两个目标优化函数的值；
D)根据两个所述目标优化函数的值的乘积，更新ESNs深度回声状态网络储蓄池层数和每一层储蓄池矩阵谱半径的搜索路径，使得下一次两个目标函数值的乘积大于当前次两个目标函数值的乘积，从而得到新的储蓄池层数θ m+1和储蓄池矩阵谱半径κ m+1；
E)迭代次数加1，将新的储蓄池层数θ m+1和储蓄池矩阵谱半径κ m+1作为多目标灰狼优化算法目标优化函数的输入，返回步骤C)，直至多目标灰狼优化算法目标优化函数值达到预期精度或完成所设定的迭代次数，完成ESNs深度回声状态网络训练，并获取最优参数θ optimal和κ optimal，该最优参数θ optimal和κ optimal对应的ESNs深度回声状态网络模型即预测模型。
ESNs深度回声状态网络模型具有优秀的数据拟合能力，经过多目标灰狼优化算法优化过参数的ESNs深度回声状态网络模型具有更小的预测误差，可以更准确的预测车辆故障。
两个所述目标优化函数表达式为：
Figure PCTCN2020141464-appb-000005
Figure PCTCN2020141464-appb-000006
Figure PCTCN2020141464-appb-000007
Figure PCTCN2020141464-appb-000008
Figure PCTCN2020141464-appb-000009
Figure PCTCN2020141464-appb-000010
其中，θ是储蓄池层数，κ是储蓄池矩阵谱半径，
Figure PCTCN2020141464-appb-000011
是将θ和κ代入ESNs模型输出的的预测值，
Figure PCTCN2020141464-appb-000012
是所有预测值的平均值；V t是DNA序列的真实值，
Figure PCTCN2020141464-appb-000013
是所有真实值的平均值；N是DNA序列的长度，且1≤t≤N，下标CT表示车体故障，ZXJ表示转向架故障，QY表示牵引传动控制系统故障，ZD表示制动系统故障，LJ表示车端连接装置故障，SL表示受流装置故障，SB表示车辆内部设备和驾驶室设备故障；
Figure PCTCN2020141464-appb-000014
Figure PCTCN2020141464-appb-000015
NSE和KGE是用于衡量模型稳定性的指标，以这两个指标为基础设置目标函数进行优化可以使预测模型具有更强的鲁棒性。
还包括：将预判定的候选车辆部件故障基因V s作为聚类模型的输入，搭建 模板库。搭建模板库可以帮助行业有关人员对比当前故障和历史故障的异同，从而采取更精准的维修操作。
搭建模板库的具体实现过程包括：
步骤1：将连续投影法降维得到的预判定的候选车辆部件故障基因V s作为随机邻接嵌入算法的输入，得到高维数据点V i和V j的条件概率p j|i、低维数据点v i和v j的条件概率q j|i，将条件概率最小化，得到最小化的高维数据的条件概率p j|i和最小化的低维数据的条件概率q ij；
步骤2：依据条件概率最小化结果计算出高低维条件概率差异的最小值p ij，
Figure PCTCN2020141464-appb-000016
通过梯度下降法最小化代价函数L：
Figure PCTCN2020141464-appb-000017
得到最优解
Figure PCTCN2020141464-appb-000018
将所述最优解
Figure PCTCN2020141464-appb-000019
作为tSNE聚类算法的聚类结果输出；所述聚类结果对应ART城市智轨车辆的模板库template：
template＝[CT,ZXJ,QY,ZD,LJ,SL,SB]；
其中，CT,ZXJ,QY,ZD,LJ,SL,SB为DNA序列模板库中的故障类别；CT：车体故障；ZXJ：转向架故障；QY：牵引传动控制系统故障；ZD：制动系统故障；LJ：车端连接装置故障；SL：受流装置故障；SB：车辆内部设备和驾驶室设备故障；n代表数据样本数，KL表示散度。
连续投影法降维和t-SNE聚类相结合的方法避免了车辆故障有效信息大量丢失的不利情况，柔和聚类可以得到更可靠的模板库信息。
步骤4)之后，还包括：
5)判断所述预测模型输出的预测序列对应故障类别是否与所述模板库中的故障类别匹配，若所述故障类别从属于模板库中某一故障类别中的子类别，则将所述故障类别划入该故障类别中，记为旧类故障
Figure PCTCN2020141464-appb-000020
若所述故障类别不属于模板库中的任何一类，则将预测序列对应的故障类别添加进模板库，并标记为新的故障类型
Figure PCTCN2020141464-appb-000021
模板库比对机制帮助有关人员快速识别当前故障，模板库更新机制帮助完善模板库的内容，从而容纳更多故障信息。
本发明的方法还包括：
利用预测模型对实时采集的振动数据进行预测，再利用DNA螺旋序列解码及虚拟模板库实现预测结果的可视化；具体实现过程包括：对预测模型输出的预测结果进行二进制逆编码转换，其中，二进制逆编码转换后的预测结果中的腺嘌呤、胸腺嘧啶的结合碱基对被解码后对应数字0，即设备故障程度未达到警戒线阈值，鸟嘌呤、C胞嘧啶的结合碱基对被解码后对应数字1，即设备故障程度达到了警戒线阈值，必须进行检修修复。
本发明还提供了一种智慧城市智轨车辆故障基因预测系统，其包括计算机设备；所述计算机设备被配置或编程为用于执行上述方法的步骤。
与现有技术相比，本发明所具有的有益效果为：
1)本发明在现有智轨车辆故障诊断技术的基础上，提供了一种基于DNA序列模板库的精准故障预测方法。无线传感网络与高低频振动测量仪器相结合的数据采集模块可收集大量历史故障信号，多源振动信号的数据编码模块将振动信号转换为可编码的基因序列，可编码碱基对的DNA序列特征提取模块可筛选出预判定的候选车辆部件故障基因、DNA序列模板库模块的搭建可以帮助行业有关人员比对新检测故障与历史故障的异同、可编码DNA螺旋序列深度学习的故障预警建模模块可预测列车部件潜在的故障、基于多目标优化的DNA螺旋序列预测策略模块可提升故障预警的精度、DNA序列螺旋解码及虚拟模板库的故障可视化模块帮助维修人员快速识别故障类型。
2)本发明搭建了可编码故障模块的DNA序列模板库，对应城市智轨车辆的七大部件(车体、转向架、牵引传动控制系统、制动系统、车端连接装置、受流装置、车辆内部设备和驾驶室设备)。故障模板库作为虚拟故障的匹配模板，为训练可靠的故障预警模型提供了准确方向。精准完善的故障信息库更有利于工作人员比对自主轨道列车新老设备的故障异同从而进行故障检修。
3)本发明提出了一种自主轨道列车的故障诊断多故障预测匹配的建模方法，在智轨列车的各大部件上安装振动传感器，采集实时振动数据信号，并通过无线传感网络(WSN)进行传输，并建立深度回声状态网络(ESNs)以进行设备故障的多目标优化预测，极大地提高了故障预测的准确性。
4)围绕数据采集、原始信号螺旋编码解码、基因信号转化、基因序列特征 提取、建立故障模块的DNA序列模板库、故障预测等环节搭建了一个完整的系统框架，基于时效性的兼顾，本发明的模型可以被嵌入到Hadoop大数据平台中进行训练，提高训练速度。
附图说明
图1为本发明实施例方法原理图。
具体实施方式
如图1，本发明实施例具体实现过程如下：
步骤1：新型智轨列车部件历史故障数据采集
本发明运用高、低频传感器和电动式传感器采集各型号城市智轨车辆部件的历史振动信息数据，技术更迭很大程度上降低了传感器的普及应用成本。另外，无线传感网络(WSN)扮演了重要的角色，运用该方法可以将多辆列车的振动信号及时上传整合到数据集成平台，步骤1涉及到的信息采集模块包括车辆部件振动幅度采集模块、振动频率采集模块和振动周期采集模块。采集到的信息包括车辆部件的振动幅度A、频率f、周期T等信号，需通过滤波器进行数据过滤，最后得到原始振动信号X。
本发明中：低频振动测量：相对式动圈型电动传感器；高频振动测量：惯性式动圈型电动传感器。电动式传感器可对一些民用工业入车辆等的重要部件进行振动测试。
步骤2：多源振动信号的DNA螺旋序列数据编码转换
为了完成对爆炸信息的有效存储，采集到的多源振动数据需要被编码为DNA序列，编码后的振动信号其特征表现更加明显且易区分，从而便于后续的预测工作。DNA序列的碱基数据排列开后在数学表现上是高维或超高维的矩阵，为了有效利用这些数据，需要对其进行稳健鲁棒降维处理。对DNA螺旋序列进行连续投影算法(SPA)处理可实现快速降维以解决共线性问题，需调整的参数少且思想简单(参见Soares S F C,Gomes A A,Araujo M C U,et al.The successive projections algorithm[J].TrAC Trends in Analytical Chemistry,2013,42:84-98.)。它相比传统的降维方法具有高效、可解释性强的特点。能有效的降低数据的维度且保持关键信息不丢失。
首先给定一个初始的原始振动信号序列数据(即步骤1中采集到的新型智轨 列车部件原始振动信号)矩阵X h(0)＝[e 1,e 2,e 3,...,e n]∈R，其中矩阵的每一列代表一个DNA螺旋序列数据样本，e 1,e 2,...,e n代表每一个基础高低频振动传感器采集到的振动信息，采样频率为0.2s。在第一次迭代之前，选中采集的原始振动信号X的第g列样本赋值到初始的DNA螺旋序列数据矩阵X h(0)，赋值后的矩阵表示为X g。接下来一步需要对赋值后的DNA螺旋序列数据矩阵X g与最大投影值矩阵X h(z-1)在子空间中的正交投影进行计算：
Figure PCTCN2020141464-appb-000022
h(z)＝arg(max(||FX g||,g∈E))        (2)
式中，F为投影算子，即初始螺旋序列正交于其他螺旋序列的投影；h(z-1)是最大投影值；h(0)为最小投影值，后续以此类推，共Z个投影；其中最大投影值正规化后为G，即垂直投影，最小投影为0，即平行投影，从最小值开始投影角度每变化一个数值Υ，投影值增大
Figure PCTCN2020141464-appb-000023
z是投影值的序号。
经过连续投影算法(SPA)降维处理后得到的维度为U的数据矩阵集合Y可以被表示如下：
Y＝{x h(z)；z＝0,1,2,3,...,U-1}            (3)
其中初始螺旋序列X h(0)的选择十分关键，直接影响了算法的精度，该算法本质上可以看作是一种矩阵投影。在本发明中对应的则是数据类型维度的转化，振动信号被映射成一组低维度的基因表达，这些基因代表发生故障部件的表达情况。
在步骤2中，需将步骤1中采集到的设备振动信号X进行基因序列的转化。根据需要保留信息的程度来定义降维维度U，连续投影法(SPA)将原始振动数据降维后得到Y，Y作为一整组数据的集合，其矩阵中数据的幅值大小有高有低，根据经验阈值，将数据进行归一化处理，此处粗略的先将Y按照25％一个碱基占比的比例将Y划分为4类，即A，T，C，G，归一化后幅值处于[0.75，1]的数据定义为A类，然后根据各部位设备传感器数量的配比，来调整A，T，C， G的最终占比。经过连续投影算法(SPA)降维处理后的振动样本数据被定义为由A，T，C，G四种碱基元素表达的U维特征向量，即后续步骤所需的经过转化的可编码基因序列信号，此处的目的是将数据划分为4类。为了方便表示，分别用B 1,B 2,B 3,B 4代替“A，T，C，G”四种碱基表达。经过预处理后的振动信号被转化为可编码的基因序列。
步骤3：可编码的基因序列的碱基对特征提取
经过编码转换后的DNA序列还不具备可供高效高精度预测的特性，需要进行特征提取的操作来将设备故障的深度表达提取出来，并排列组合以形成可预测的DNA序列。
将经过步骤1(历史故障数据采集)、步骤2(DNA螺旋序列数据编码转换)处理后的U维可编码基因序列信号输入到故障特征提取模块。该环节通过计算转化后的基因序列中碱基的含量、位置以及转移概率几个特征来进行独立的自主轨道列车故障部件的DNA序列特征提取。
A1.将步骤2中降维得到的可编码基因序列定义为S＝S 1,S 2,S 3,...,S N，其长度为N，如果该DNA序列中第k(1≤k≤N)个数据点位置的碱基为B i(1≤i≤U)，则记为S k＝B i；对于连续两个碱基点的情况，如果第l(1≤l≤N-1)个数据点位置的碱基为B i，第l+1个数据点位置的碱基为B j，记为S lS l+1＝B iB j(1≤i,j≤U)。
A2.定义碱基转移概率W ij。首先将n i定为单个碱基点B i在DNA序列S中出现的次数，此外，对于连续两个点碱基的情况，定义n ij为碱基对B iB j在DNA序列S中出现的次数。具体计算式为：
Figure PCTCN2020141464-appb-000024
对于特殊情况，如果碱基B i没有出现在DNA序列S当中，或者出现但是只出现在最后一次，那么此时可以视W ij的分子分母为0，即W ij＝0。
补充说明，
Figure PCTCN2020141464-appb-000025
这是因为：
Figure PCTCN2020141464-appb-000026
Figure PCTCN2020141464-appb-000027
所以可将W ij视为碱基B i转移到碱基B j的概率，即碱基转移概率向量。
A3.定义碱基含量C i。DNA序列S中碱基B i(1≤i≤U)的含量可以被如下表达式记录：
Figure PCTCN2020141464-appb-000028
对于U维碱基来说，其含量向量为C 1,C 2,C 3,...,C U。
A4.定义碱基位置比D i。将DNA序列S中碱基B i(1≤i≤U)出现的位置标记为S i，叠加表达式如下：
Figure PCTCN2020141464-appb-000029
转化得到碱基位置比D i，数学表达式如下所示：
Figure PCTCN2020141464-appb-000030
对于U维碱基来说，其位置比向量为D 1,D 2,D 3,...,D U。
可编码基因序列经特征提取后，可得到能利用的U维向量。将上述步骤计算得到的基转移概率向量、碱基含量向量、碱基位置比向量进行整合可得到V s＝(W 11,W 12,...,W UU,C 1,...,C U,D 1,...,D U)。这些特征向量被定义为预判定的候选车辆部件故障基因。
步骤4：建立故障模块的DNA序列模板库
步骤3提取出的候选故障基因特征向量在此环节被输入到(t分布随机邻域嵌入)t-SNE聚类模型当中，经过精细的聚类划分来建立故障模块的DNA序列模板库。模板库对应到城市智轨车辆的7大板块，分别是车体(CT)库、转向架(ZXJ)库、牵引传动控制系统(QY)库、制动系统(ZD)库、车端连接装置(LJ)库、受流装置(SL)库、车辆内部设备和驾驶室设备(SB)库。其中括号中的简写代表获取基因序列表达的标签。值得一提的是，如过直接用连续投影算法(SPA)将振动信号降到3维空间则会丢失大量关键信息，所以在本发明中先用连续投影算法(SPA)降到一个中小型的多维空间U，并用多维碱基特征来表达，最后再利用t-SNE聚类方法得到最后的聚类结果，这样可以达到柔和聚类的效果。每个聚类结果对应一个部件的故障，聚类出的结果被输送到步骤5的预测器模型当中进行训练，然后再利用DNA序列模板进行二次的详细划分。t-SNE是一种可以探索高维数据的非线性降维的算法，在本发明中车辆故障模块t-SNE的DNA序列聚类方法如下：
B1.首先通过随机邻接嵌入(SNE)来把数据进行转化，数据间的高维欧几里得距离被转化后表示为相似的条件概率，具体的，候选车辆部件故障基因高维数据点V i、V j的条件概率p j|i的数学计算给出如下：
Figure PCTCN2020141464-appb-000031
式中，V i,V j是DNA序列S中的数据点，σ i是以数据点V i,V j为中心的高斯方差。
B2.高维数据点到低维数据点的转化。同样的，对于低维数据点v i,v j而言，其条件概率q j|i的计算方法也是类似的：
Figure PCTCN2020141464-appb-000032
在这一过程当中，随机邻域嵌入算法尝试讲条件概率的差异最小化。对于t-SNE来说，设v服从t分布，则可以得到：
Figure PCTCN2020141464-appb-000033
其中，s为预判定的候选车辆部件故障基因的序号。
B3.测量高低维度条件概率差异之和的最小值。在这一环节中，SNE使用梯度下降法来最小化Kullback-Leibler差异距离，同时SNE的代价函数将注意力投放到了映射数据的局部结构，更进一步的，运用t-SNE的重尾分布来减轻优化该函数的拥挤问题。为了使P和Q的分布尽可能接近，则需要使KL的散度尽可能的小，并计算p ij：
Figure PCTCN2020141464-appb-000034
KL散度的值越小，两个分布之间的距离越近。当散度KL＝0，时，则说明P和Q的分布是相同的。如果降维后特征空间中的点的概率分布近似于原始特征空间中的点的概率分布，则可以得到定义良好的聚类，此处通过梯度下降法最小化代价函数：
Figure PCTCN2020141464-appb-000035
B4.迭代寻优，对变量目标函数L进行寻优，不断更新低维数据点，直至得到对应所求的最优解
Figure PCTCN2020141464-appb-000036
该最优解为可以表达为CT、ZXJ、QY、ZD、LJ、SL和SB的几簇。
Figure PCTCN2020141464-appb-000037
其中，y为迭代过程中迭代次数，y max为最大迭代总次数，η为学习率，α(y)为学习动量，低维数据的集合
Figure PCTCN2020141464-appb-000038
此环节需要大量历史故障数据作为支撑。模板库对应故障类型，一种基因特征表现对应到一种部件的故障，最后系统发出诊断预警报告。最终得到的最优解
Figure PCTCN2020141464-appb-000039
聚类结果可以表达为CT、ZXJ、QY、ZD、LJ、SL、SB的几簇，可视化表达为7个ART城市智轨车辆大部件的DNA序列的聚类模板。由聚类结果对应得到的模板库表达如下所示：
template＝[CT,ZXJ,QY,ZD,LJ,SL,SB]        (16)
CT：车体；ZXJ：转向架；QY：牵引传动控制系统；ZD：制动系统；LJ：车端连接装置；SL：受流装置；SB：车辆内部设备和驾驶室设备。聚类形成的模板库通过至此，故障模块的DNA序列模板库搭建完成。
具体地，模板库的搭建可以总结为：
a：将经过连续投影法(SPA)降维得到的预判定的候选车辆部件故障基因V s作为随机邻接嵌入(SNE)算法的输入来分别得到高维数据点V i、V j和低维数据点v i,v j的条件概率p j|i和q j|i，并进一步将条件概率最小化得到最小化的高维数据的条件概率p j|i和最小化的低维数据的条件概率q ij。
b：依据条件概率最小化结果计算出高低维条件概率差异的最小值，并计算
Figure PCTCN2020141464-appb-000040
通过梯度下降法最小化代价函数L，其中n为数据样本的数量，最后根据上述结果来计算得到最优解
Figure PCTCN2020141464-appb-000041
也就是说将最优解
Figure PCTCN2020141464-appb-000042
作为tSNE聚类算法的聚类结果输出。这些输出的聚类信息熵簇对应7个ART城市智轨车辆大件的DNA序列的聚类模板。
步骤5：可编码DNA螺旋序列的多目标优化深度学习故障预警建模
将预判定的候选车辆部件故障基因归一化后输入模型进行城市智轨列车设备的故障预测训练。具体的建模过程如下：
C1.设置训练集和测试集。按照训练集、测试集分别60％，40％的比例将输入模型的数据进行划分，此外，预测模型的评价指标设定为纳什萨克利夫效率(NSE)指数，克林古普塔效率(KGE)指数，数值越接近1代表模型的性能越好。
C2.搭建与智轨列车设备部件DNA序列特征模板库形成映射关系的深度学习预测模型，并进行模型参数的优化。深度回声状态网络中储蓄池的层数和每一层储蓄池矩阵谱的半径的设定对预测模型的预测精度起到很大的影响作用。为了使ESNs模型的性能得到再一次的提升，多目标灰狼优化算法(MOGWO)被用 于进行ESNs的储蓄池的层数和每一层储蓄池矩阵谱的半径参数寻优。参数寻优过程与ESNs建模过程同时进行，具体实施细节如下：
1)选定优化算法、初始化参数：选择多目标灰狼优化算法进行ESNs模型的参数优化。优化算法的迭代次数设置为200，预期精度为
Figure PCTCN2020141464-appb-000043
当达到预设迭代次数或满足预期精度时迭代停止。
2)设定优化变量：将深度回声状态网络储蓄池的层数θ和每一层储蓄池矩阵谱的半径κ设定为需要优化的变量。在本环节中，深度回声状态网络的储蓄池节点被初始设定为15，网络各层的输入输出层层相对，继而学习可编码数据的深度特征表示。
3)模型训练。将所述训练集、ESNs深度回声状态网络模型储蓄池的初始层数θ 0和每一层储蓄池矩阵谱的初始半径κ 0作为ESNs深度回声状态网络模型的输入，以具有储蓄池层数θ m和储蓄池矩阵谱半径κ m的ESNs深度回声状态网络模型作为输出，训练ESNs深度回声状态网络模型。
4)进行模型参数的多目标优化(参见MIRJALILI S,SAREMI S,MIRJALILI S M,et al.Multi-objective grey wolf optimizer[J].Expert Systems With Applications,2016,47:106-19.)。为了进一步提升模型性能，将多目标灰狼优化算法嵌入领导者选择机制和存档储存机制来提高收敛能力。将所述测试集、储蓄池层数θ m和储蓄池矩阵谱半径κ m作为多目标灰狼优化算法目标优化函数的输入，计算目标优化函数值；其中m表示当前迭代次数，0≤m≤200。
设置优化目标函数为最大化各类设备的纳什萨克利夫效率(NSE)指数和克林古普塔效率(KGE)指数，当目标函数object1和目标函数object2通过多目标优化取到综合最优时，会形成一组同时包含多个(θ,κ)的帕累托面解集，该解集上每一(θ,κ)都对应两个目标函数值的综合最优，优化函数值可以被计算如下：
Figure PCTCN2020141464-appb-000044
Figure PCTCN2020141464-appb-000045
Figure PCTCN2020141464-appb-000046
Figure PCTCN2020141464-appb-000047
Figure PCTCN2020141464-appb-000048
Figure PCTCN2020141464-appb-000049
其中，θ是储蓄池层数，κ是储蓄池矩阵谱半径，
Figure PCTCN2020141464-appb-000050
是将θ和κ代入ESNs模型输出的预测值，V t是DNA碱基序列的真实值，N是DNA序列的长度，且1≤t≤N，下标CT表示车体故障，ZXJ表示转向架故障，QY表示牵引传动控制系统故障，ZD表示制动系统故障，LJ表示车端连接装置故障，SL表示受流装置故障，SB表示车辆内部设备和驾驶室设备故障。
Figure PCTCN2020141464-appb-000051
Figure PCTCN2020141464-appb-000052
5)根据两个所述目标优化函数的值的乘积来更新ESNs深度回声状态网络储蓄池的层数和每一层储蓄池矩阵谱的半径的搜索路径，使得下一次两个目标函数值的乘积大于当前次两个目标函数值的乘积，得到新的储蓄池层数θ m+1和储蓄 池矩阵谱半径κ m+1。
6)迭代次数加1，将新的储蓄池层数θ m+1和储蓄池矩阵谱半径κ m+1作为多目标灰狼优化算法目标优化函数的输入，返回步骤4)，直至多目标灰狼优化算法目标优化函数值达到预期或完成所设定的迭代次数，完成ESNs深度回声状态网络训练，并获取最优参数θ optimal和κ optimal，该最优参数θ optimal和κ optimal对应的ESNs深度回声状态网络模型即预测模型。
当DNA碱基序列的预测值贴近真实值时，就说明预测模型训练合理，准确的完成了设备的故障预测任务。预测出来的结果可以对应步骤4模板库中的template＝[CT,ZXJ,QY,ZD,LJ,SL,SB]七类故障，判断所述预测模型输出的预测序列对应故障类别是否与模板库中的故障类别匹配，若所述故障类别从属于模板库中某一故障类别中的子故障，则将所述故障类别划入该故障的模板库中，记为旧类故障
Figure PCTCN2020141464-appb-000053
若所述故障类别不属于模板库中的任何一类，则进行模板库的更新，直接将该预测结果添加进模板库，并标记为新故障
Figure PCTCN2020141464-appb-000054
DNA序列模板库为预测模型后续的训练指引了方向。
步骤6：DNA螺旋序列解码及虚拟模板库的故障可视化
在本发明中，原始的振动数据被进行了编码转化和特征提取等步骤，其目的是使原本特性不明显的原始序列的深度特征表达被抽取出来，然后将这一易区分的深度特征序列输入到预测模型当中进行训练，训练好的模型预测出的序列结果依旧是深度特征表达，每个DNA片段的预测结果被首尾相连后形成完整的碱基序列编码，相应的故障对应到虚拟DNA模板库中的具体类型。然而结果的表现形式不是具象的，所以要对其进行DNA序列解码和故障可视化，以还原到数据类型对应于原始振动数据的预测数据。为了实现城市智轨电车设备的故障预测结果可视化，利用步骤5得到的预测模型对实时采集的振动数据进行预测，再对预测模型输出的预测结果(该ART城市智轨车辆故障预测输出结果的编码机制是基于步骤3的碱基编码体系的，所以其序列的构成建立在A，T，G，C碱基的基础上通过深度学习预测建模得到)进行二进制逆编码转换。此处，将预测输出结果进行解码操作，以显示为0/1状态，即完成预测结果的可视化，以实现及时 预警。其中，A(腺嘌呤)、T(胸腺嘧啶)的结合碱基对被解码后对应数字0，即设备故障程度未达到警戒线阈值，G(鸟嘌呤)、C(胞嘧啶)的结合碱基对被解码后对应数字1，即设备故障程度达到了警戒线阈值，必须进行检修修复。而故障的预警模型搭建也可以为ART城市智轨车辆的安全稳定运行提供可靠的保障。
DNA故障螺旋序列解码编码的优势明显，对于爆炸式的信息存储，DNA序列数据存储方法为信息收发和保存提供了无限可能，且存储时间完全能够满足大数据时代的信息使用需求。
步骤7：分布式系统基础架构嵌入
综合本发明提出方法的时耗性以及实际工程中智轨设备维修的实时性需要，可将模块嵌入分布式系统基础架构以加快模型训练和自学习更新速度，从而更大程度上的满足应用需求。可用的分布式系统基础架构包括MapReduce、Apache Spark、Hadoop等(参见DittrichJ,Quiané-Ruiz J A.Efficient big data processing in Hadoop MapReduce[J].Proceedings of the VLDB Endowment,2012,5(12):2014-2015.)。这些大规模数据处理的分析引擎和集群计算系统具有高效性、易用性、通用性和兼容性等特征，可以极大的满足使用需求。

Claims (10)

  1. 一种智慧城市智轨车辆故障基因预测方法，其特征在于，包括以下步骤：
    1)采集列车部件的振动数据X h(0)＝[e 1,e 2,e 3,...,e n]∈R，其中，e 1,e 2,...,e n代表列车上每一个采样点的振动信息；n表示采样点个数；
    2)将所述振动数据编码为DNA序列，提取所述DNA序列的特征，并排列组合所述特征，形成可预测的DNA序列，即候选车辆部件故障基因；
    3)利用所述候选车辆部件故障基因训练ESNs深度回声状态网络，得到预测模型；
    优选地，还包括：
    4)根据实时采集的振动数据，利用所述预测模型预测车辆故障。
  2. 根据权利要求1所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于，步骤2)中，将所述振动数据编码为DNA序列的具体实现过程包括：
    A)选中采集的原始振动信号X的第g列样本，将所述第g列样本赋值到初始的DNA螺旋序列数据矩阵X h(0)，赋值后的矩阵表示为X g；
    B)计算赋值后的DNA螺旋序列数据矩阵X g与最大投影值矩阵X h(z-1)在子空间中的正交投影，得到维度为U的数据矩阵集合Y；z是投影值的序号；h(z-1)是最大投影值；最大投影值正规化后为G，即垂直投影，最小投影值h(0)为0，即平行投影，从最小值开始投影角度每变化一个数值Υ，投影值增大
    Figure PCTCN2020141464-appb-100001
    Z为投影值数量；
    C)将所述数据矩阵集合Y划分为由A，T，C，G四种碱基元素表达的U维特征向量；将A，T，C，G整合为DNA序列S＝S 1,S 2,S 3,...,S N；其中，N为DNA序列长度。
  3. 根据权利要求2所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于，步骤2)中，候选车辆部件故障基因V s表示为：V s＝(W 11,W 12,...,W UU,C 1,...,C U,D 1,...,D U)；其中，碱基B i转移到碱基B j的概率
    Figure PCTCN2020141464-appb-100002
    n i为单个碱基点B i在DNA序列S中出现的次数；B i为DNA序列S中第i个数据点位置的碱基；1≤i≤U；U指碱基元素所代表特征向量的维度；N为DNA序列S的长度；n ij为碱基对B iB j在DNA序列S中出现的次数；碱基含量
    Figure PCTCN2020141464-appb-100003
    碱基位置比
    Figure PCTCN2020141464-appb-100004
    DNA序列S中碱基B i出现的位置标记为S i，s i是S i中的值。
  4. 根据权利要求1～3之一所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于，步骤3)的具体实现过程包括：
    A)将车辆部件故障基因V s随机划分为训练集和测试集；初始化多目标灰狼优化算法的迭代次数m、预期精度；
    B)将所述训练集、ESNs深度回声状态网络模型储蓄池的初始层数θ 0和每一层储蓄池矩阵谱的初始半径κ 0作为ESNs深度回声状态网络模型的输入，以具有储蓄池层数θ m和储蓄池矩阵谱半径κ m的ESNs深度回声状态网络模型作为输出，训练ESNs深度回声状态网络模型；
    C)将所述测试集、储蓄池层数θ m和储蓄池矩阵谱半径κ m作为多目标灰狼优化算法两个目标优化函数的输入，计算两个目标优化函数的值；
    D)根据两个所述目标优化函数的值的乘积，更新ESNs深度回声状态网络储蓄池层数和每一层储蓄池矩阵谱半径的搜索路径，使得下一次两个目标函数值的乘积大于当前次两个目标函数值的乘积，从而得到新的储蓄池层数θ m+1和储蓄池矩阵谱半径κ m+1；
    E)迭代次数加1，将新的储蓄池层数θ m+1和储蓄池矩阵谱半径κ m+1作为多目标灰狼优化算法目标优化函数的输入，返回步骤C)，直至多目标灰狼优化算法目标优化函数值达到预期精度或完成所设定的迭代次数，完成ESNs深度回声状态网络训练，并获取最优参数θ optimal和κ optimal，该最优参数θ optimal和κ optimal对应的ESNs深度回声状态网络模型即预测模型。
  5. 根据权利要求4所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于，两 个所述目标优化函数表达式为：
    Figure PCTCN2020141464-appb-100005
    Figure PCTCN2020141464-appb-100006
    Figure PCTCN2020141464-appb-100007
    Figure PCTCN2020141464-appb-100008
    Figure PCTCN2020141464-appb-100009
    Figure PCTCN2020141464-appb-100010
    Figure PCTCN2020141464-appb-100011
    其中，θ是储蓄池层数，κ是储蓄池矩阵谱半径，
    Figure PCTCN2020141464-appb-100012
    是将θ和κ代入ESNs模型输出的的预测值，
    Figure PCTCN2020141464-appb-100013
    是所有预测值的平均值；V t是DNA序列的真实值，
    Figure PCTCN2020141464-appb-100014
    是所有真实值的平均值；N是DNA序列的长度，且1≤t≤N，下标CT表示车体故障，ZXJ表示转向架故障，QY表示牵引传动控制系统故障，ZD表示制动系统故障，LJ表示车端连接装置故障，SL表示受流装置故障，SB表示车辆内部设备和驾驶室设备故障；
    Figure PCTCN2020141464-appb-100015
    Figure PCTCN2020141464-appb-100016
  6. 根据权利要求1～5之一所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于，还包括：将预判定的候选车辆部件故障基因V s作为聚类模型的输入，搭建模板库。
  7. 根据权利要求6所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于， 搭建模板库的具体实现过程包括：
    步骤1：将连续投影法降维得到的预判定的候选车辆部件故障基因V s作为随机邻接嵌入算法的输入，得到高维数据点V i和V j的条件概率p j|i、低维数据点v i和v j的条件概率q j|i，将条件概率最小化，得到最小化的高维数据的条件概率p j|i和最小化的低维数据的条件概率q ij；
    步骤2：依据条件概率最小化结果计算出高低维条件概率差异的最小值p ij，
    Figure PCTCN2020141464-appb-100017
    通过梯度下降法最小化代价函数L：
    Figure PCTCN2020141464-appb-100018
    得到最优解
    Figure PCTCN2020141464-appb-100019
    将所述最优解
    Figure PCTCN2020141464-appb-100020
    作为tSNE聚类算法的聚类结果输出；所述聚类结果对应ART城市智轨车辆的模板库template：template＝[CT,ZXJ,QY,ZD,LJ,SL,SB]；
    其中，CT,ZXJ,QY,ZD,LJ,SL,SB为DNA序列模板库中的故障类别；CT：车体故障；ZXJ：转向架故障；QY：牵引传动控制系统故障；ZD：制动系统故障；LJ：车端连接装置故障；SL：受流装置故障；SB：车辆内部设备和驾驶室设备故障；KL表示散度。
  8. 根据权利要求7所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于，步骤4)之后，还包括：
    5)判断所述预测模型输出的预测序列对应故障类别是否与所述模板库中的故障类别匹配，若所述故障类别从属于模板库中某一故障类别中的子类别，则将所述故障类别划入该故障类别中，记为旧类故障
    Figure PCTCN2020141464-appb-100021
    若所述故障类别不属于模板库中的任何一类，则将预测序列对应的故障类别添加进模板库，并标记为新的故障类型
    Figure PCTCN2020141464-appb-100022
  9. 根据权利要求1～8之一所述的一种智慧城市智轨车辆故障基因预测方法，其特征在于，还包括：
    利用预测模型对实时采集的振动数据进行预测，再利用DNA螺旋序列解码及虚拟模板库实现预测结果的可视化；具体实现过程包括：对预测模型输出的预测结果进行二进制逆编码转换，其中，二进制逆编码转换后的预测结果中的腺嘌呤、胸腺嘧啶的结合碱基对被解码后对应数字0，即设备故障程度未达到警戒线阈值，鸟嘌呤、C 胞嘧啶的结合碱基对被解码后对应数字1，即设备故障程度达到了警戒线阈值，必须进行检修修复。
  10. 一种智慧城市智轨车辆故障基因预测系统，其特征在于，包括计算机设备；所述计算机设备被配置或编程为用于执行权利要求1～9之一所述方法的步骤。
PCT/CN2020/141464 2020-12-30 2020-12-30 一种智慧城市智轨车辆故障基因预测方法及系统 WO2022141213A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2020/141464 WO2022141213A1 (zh) 2020-12-30 2020-12-30 一种智慧城市智轨车辆故障基因预测方法及系统
AU2020471210A AU2020471210B2 (en) 2020-12-30 2020-12-30 Fault gene predicting method and system for Autonomous-Rail Rapid Transit vehicle in smart city

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/141464 WO2022141213A1 (zh) 2020-12-30 2020-12-30 一种智慧城市智轨车辆故障基因预测方法及系统

Publications (1)

Publication Number Publication Date
WO2022141213A1 true WO2022141213A1 (zh) 2022-07-07

Family

ID=82259989

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/141464 WO2022141213A1 (zh) 2020-12-30 2020-12-30 一种智慧城市智轨车辆故障基因预测方法及系统

Country Status (2)

Country Link
AU (1) AU2020471210B2 (zh)
WO (1) WO2022141213A1 (zh)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115174639A (zh) * 2022-09-07 2022-10-11 吉林信息安全测评中心 一种基于物联网的工业设备管理方法和系统
CN115333990A (zh) * 2022-10-14 2022-11-11 中国人民解放军海军工程大学 一种基于郊狼与灰狼混合优化的高效传输路由方法
CN116028822A (zh) * 2023-03-30 2023-04-28 国网福建省电力有限公司 一种电能表误差状态评估方法、系统、设备及存储介质
CN116165274A (zh) * 2023-02-17 2023-05-26 哈尔滨工业大学 基于贝叶斯全局稀疏概率主成分分析的城市轨道损伤识别方法
CN116366477A (zh) * 2023-05-30 2023-06-30 中车工业研究院(青岛)有限公司 一种列车网络通信信号检测方法、装置、设备及存储介质
CN116579112A (zh) * 2023-05-17 2023-08-11 西南交通大学 一种基于轨道交通网络的应急桥接公交驻车点选址方法
CN116861218A (zh) * 2023-07-25 2023-10-10 上海华菱电站成套设备股份有限公司 一种矿井提升机关键设备状态监测预警系统
CN116956197A (zh) * 2023-09-14 2023-10-27 山东理工昊明新能源有限公司 基于深度学习的能源设施故障预测方法、装置及电子设备
CN117132130A (zh) * 2023-08-31 2023-11-28 江西航天国数信息科技有限公司 一种实时采集信息进行建模的智慧城市预测系统
CN117290764A (zh) * 2023-11-23 2023-12-26 湖南省交通科学研究院有限公司 基于数据特征分析的治超系统故障智能识别及诊断的方法
CN117302293A (zh) * 2023-08-14 2023-12-29 北京城建智控科技股份有限公司 轮径值预测方法、装置、电子设备及存储介质
CN117649876A (zh) * 2024-01-29 2024-03-05 长春大学 基于gwo算法在gwas数据上检测与复杂疾病相关snp组合的方法
WO2024077642A1 (zh) * 2022-10-12 2024-04-18 大连理工大学 针对航空发动机故障预警的量子回声状态网络模型构建方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7610130B1 (en) * 2005-07-20 2009-10-27 Sandia Corporation Physical context management for a motor vehicle
CN107886193A (zh) * 2017-10-27 2018-04-06 太原理工大学 一种基于灰狼优化回声状态网络的时序预测方法
CN108267312A (zh) * 2017-12-25 2018-07-10 北京建筑大学 一种基于快速搜索算法的地铁列车轴承智能诊断方法
CN108304960A (zh) * 2017-12-29 2018-07-20 中车工业研究院有限公司 一种轨道交通设备故障诊断方法
CN110232415A (zh) * 2019-06-12 2019-09-13 中南大学 一种基于生物信息特征的列车转向架故障识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7610130B1 (en) * 2005-07-20 2009-10-27 Sandia Corporation Physical context management for a motor vehicle
CN107886193A (zh) * 2017-10-27 2018-04-06 太原理工大学 一种基于灰狼优化回声状态网络的时序预测方法
CN108267312A (zh) * 2017-12-25 2018-07-10 北京建筑大学 一种基于快速搜索算法的地铁列车轴承智能诊断方法
CN108304960A (zh) * 2017-12-29 2018-07-20 中车工业研究院有限公司 一种轨道交通设备故障诊断方法
CN110232415A (zh) * 2019-06-12 2019-09-13 中南大学 一种基于生物信息特征的列车转向架故障识别方法

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115174639A (zh) * 2022-09-07 2022-10-11 吉林信息安全测评中心 一种基于物联网的工业设备管理方法和系统
CN115174639B (zh) * 2022-09-07 2022-12-06 吉林信息安全测评中心 一种基于物联网的工业设备管理方法和系统
WO2024077642A1 (zh) * 2022-10-12 2024-04-18 大连理工大学 针对航空发动机故障预警的量子回声状态网络模型构建方法
CN115333990A (zh) * 2022-10-14 2022-11-11 中国人民解放军海军工程大学 一种基于郊狼与灰狼混合优化的高效传输路由方法
CN115333990B (zh) * 2022-10-14 2023-02-03 中国人民解放军海军工程大学 一种基于郊狼与灰狼混合优化的高效传输路由方法
CN116165274A (zh) * 2023-02-17 2023-05-26 哈尔滨工业大学 基于贝叶斯全局稀疏概率主成分分析的城市轨道损伤识别方法
CN116165274B (zh) * 2023-02-17 2023-11-14 哈尔滨工业大学 基于贝叶斯全局稀疏概率主成分分析的城市轨道损伤识别方法
CN116028822A (zh) * 2023-03-30 2023-04-28 国网福建省电力有限公司 一种电能表误差状态评估方法、系统、设备及存储介质
CN116579112A (zh) * 2023-05-17 2023-08-11 西南交通大学 一种基于轨道交通网络的应急桥接公交驻车点选址方法
CN116579112B (zh) * 2023-05-17 2024-02-02 西南交通大学 一种基于轨道交通网络的应急桥接公交驻车点选址方法
CN116366477B (zh) * 2023-05-30 2023-08-18 中车工业研究院(青岛)有限公司 一种列车网络通信信号检测方法、装置、设备及存储介质
CN116366477A (zh) * 2023-05-30 2023-06-30 中车工业研究院(青岛)有限公司 一种列车网络通信信号检测方法、装置、设备及存储介质
CN116861218A (zh) * 2023-07-25 2023-10-10 上海华菱电站成套设备股份有限公司 一种矿井提升机关键设备状态监测预警系统
CN117302293B (zh) * 2023-08-14 2024-06-07 北京城建智控科技股份有限公司 轮径值预测方法、装置、电子设备及存储介质
CN117302293A (zh) * 2023-08-14 2023-12-29 北京城建智控科技股份有限公司 轮径值预测方法、装置、电子设备及存储介质
CN117132130A (zh) * 2023-08-31 2023-11-28 江西航天国数信息科技有限公司 一种实时采集信息进行建模的智慧城市预测系统
CN116956197A (zh) * 2023-09-14 2023-10-27 山东理工昊明新能源有限公司 基于深度学习的能源设施故障预测方法、装置及电子设备
CN116956197B (zh) * 2023-09-14 2024-01-19 山东理工昊明新能源有限公司 基于深度学习的能源设施故障预测方法、装置及电子设备
CN117290764B (zh) * 2023-11-23 2024-02-09 湖南省交通科学研究院有限公司 基于数据特征分析的治超系统故障智能识别及诊断的方法
CN117290764A (zh) * 2023-11-23 2023-12-26 湖南省交通科学研究院有限公司 基于数据特征分析的治超系统故障智能识别及诊断的方法
CN117649876A (zh) * 2024-01-29 2024-03-05 长春大学 基于gwo算法在gwas数据上检测与复杂疾病相关snp组合的方法
CN117649876B (zh) * 2024-01-29 2024-04-12 长春大学 基于gwo算法在gwas数据上检测与复杂疾病相关snp组合的方法

Also Published As

Publication number Publication date
AU2020471210B2 (en) 2023-10-26
AU2020471210A1 (en) 2022-07-14

Similar Documents

Publication Publication Date Title
WO2022141213A1 (zh) 一种智慧城市智轨车辆故障基因预测方法及系统
CN112734094B (zh) 一种智慧城市智轨车辆故障基因预测方法及系统
CN108629978B (zh) 一种基于高维路网和循环神经网络的交通轨迹预测方法
CN111832814A (zh) 一种基于图注意力机制的空气污染物浓度预测方法
CN111915059B (zh) 基于注意力机制的Seq2seq泊位占有率预测方法
Zhu et al. Dynamic prediction of traffic incident duration on urban expressways: A deep learning approach based on LSTM and MLP
CN113487061A (zh) 一种基于图卷积-Informer模型的长时序交通流量预测方法
CN103921743B (zh) 汽车行驶工况判别系统及其判别方法
Yan et al. Spatial-temporal chebyshev graph neural network for traffic flow prediction in iot-based its
WO2021013190A1 (zh) 基于气象参数的高速列车导航盲区定位方法及系统
CN112966871B (zh) 基于卷积长短期记忆神经网络的交通拥堵预测方法及系统
WO2021013192A1 (zh) 列车过隧道时长计算方法、系统及存储介质
CN112650204B (zh) 一种智慧轨道无人驾驶车辆故障基因识别方法及系统
CN114202120A (zh) 一种针对多源异构数据的城市交通行程时间预测方法
CN112966853A (zh) 基于时空残差混合模型的城市路网短时交通流预测方法
CN110929951B (zh) 一种用于电网告警信号的关联分析和预测方法
CN116631186B (zh) 基于危险驾驶事件数据的高速公路交通事故风险评估方法、系统
CN112990545A (zh) 一种高速公路交织区交通安全状态预测方法
CN113762338A (zh) 一种基于多重图注意力机制的交通流预测方法、设备及介质
CN107688556B (zh) 一种基于函数型主成分分析的实时行程时间计算方法
CN114882069A (zh) 基于lstm网络和注意力机制的出租车轨迹异常检测方法
CN115310677A (zh) 一种基于二进制编码表示与多分类的航迹预测方法及装置
CN116307152A (zh) 时空交互式动态图注意力网络的交通预测方法
CN116523177A (zh) 融合机理和深度学习模型的车辆能耗预测方法及装置
CN111311905A (zh) 一种基于粒子群优化小波神经网络的高速公路行程时间预测方法

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2020471210

Country of ref document: AU

Date of ref document: 20201230

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20967554

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20967554

Country of ref document: EP

Kind code of ref document: A1