CN113065722B - Continuous multi-step prediction road intelligent maintenance system based on deep learning - Google Patents

Continuous multi-step prediction road intelligent maintenance system based on deep learning Download PDF

Info

Publication number
CN113065722B
CN113065722B CN202110498608.1A CN202110498608A CN113065722B CN 113065722 B CN113065722 B CN 113065722B CN 202110498608 A CN202110498608 A CN 202110498608A CN 113065722 B CN113065722 B CN 113065722B
Authority
CN
China
Prior art keywords
road
maintenance
performance index
data
prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110498608.1A
Other languages
Chinese (zh)
Other versions
CN113065722A (en
Inventor
李家乐
张志帅
王雪菲
马国伟
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.)
Hebei University of Technology
Original Assignee
Hebei 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 Hebei University of Technology filed Critical Hebei University of Technology
Priority to CN202110498608.1A priority Critical patent/CN113065722B/en
Publication of CN113065722A publication Critical patent/CN113065722A/en
Application granted granted Critical
Publication of CN113065722B publication Critical patent/CN113065722B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • 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
    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Strategic Management (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Molecular Biology (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Educational Administration (AREA)
  • Animal Behavior & Ethology (AREA)
  • Databases & Information Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a continuous multi-step prediction road intelligent maintenance system based on deep learning, which can predict road diseases which can develop into serious diseases in the next years through a prediction model and a residual correction model. According to a life cycle strategy of a road, life cycle maintenance management is that under the limited maintenance fund scale, a medium-long term maintenance strategy plan covering the life cycle of the road is made by comprehensively considering the current operation situation of a roadbed and the road surface and combining with the currently adopted preventive maintenance technology, the best time is determined for the best road section, effective maintenance measures such as reconstruction, maintenance and improvement are taken for the worst road section, the maximized fund utilization effect and the road condition improvement effect are realized, a user is guided to carry out scientific road maintenance, and the fund waste caused by the poor maintenance method is avoided.

Description

Continuous multi-step prediction road intelligent maintenance system based on deep learning
Technical Field
The invention relates to the technical field of road maintenance, in particular to a continuous multi-step prediction road intelligent maintenance system based on deep learning.
Background
By the end of 2020, the total mileage of roads in China reaches 510 kilometers and is the first place in the world. After the road is built and put into use, a lot of road surface diseases such as cracks, ruts, subsidence, pits, wave congestion, looseness and the like can appear after a period of time under the influence of factors such as vehicle load, climate and the like, and the normal service life of the road can be influenced if the road is not treated in time. The maintenance of the road can ensure safe, comfortable and smooth driving and save the transportation cost and time.
The defects of the prior art are that the traditional maintenance method has hysteresis, the maintenance mode is passive, repeated investment is often needed, the repair scale is large, and the maintained pavement is difficult to achieve the original use effect. CN 111105332A is a highway intelligent pre-maintenance method and system based on artificial neural network, which improves the prediction precision and prediction foresight, but the prediction mode can not realize multi-step prediction of road surface performance, and can not know in advance which diseases will develop into more serious diseases, so that the judgment of which diseases need to be repaired in focus can not be made.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to solve the technical problem of providing a continuous multi-step prediction road intelligent maintenance system based on deep learning.
The invention solves the technical problem by designing a continuous multi-step prediction road intelligent maintenance system based on deep learning, which comprises the following contents:
1) establishing a prediction model based on a GRU neural network, wherein the prediction model can calculate to obtain a future road performance index predicted value according to historical road performance index data and road environment data; the prediction model represents the change rules of the road performance index values at different positions of the road in different years, historical road environment data and the causal relationship between the road performance index data and future road performance index data; and then subtracting the predicted value of the pavement performance index from the true value of the pavement performance index to obtain a residual error.
2) Establishing a residual error correction model based on a GRU neural network, wherein the characteristic value of the residual error correction model is the same as that of the prediction model, namely, the input of the residual error correction model is road surface performance index data and road environment data of a fixed time step length, and the target value of the residual error correction model is a residual error; the residual correction model can accurately predict a residual predicted value, and then the predicted value of the pavement performance index calculated by the prediction model is added with the predicted value of the residual calculated by the residual correction model to obtain a corrected predicted value of the pavement performance index; the prediction model can obtain a predicted value of the pavement performance index, the residual correction model can obtain a predicted value of the residual, the predicted value of the residual is added with the predicted value of the pavement performance index to obtain a corrected predicted value of the pavement performance index, and the corrected predicted value of the pavement performance index is used as data of the pavement performance index of the corresponding year; and obtaining a trained prediction model and a residual error correction model through iterative training.
3) And inputting the road performance index data and the road environment data of the corresponding time step as characteristic values, calculating a predicted value of the road performance index of the next year by using the trained prediction model, inputting the corrected predicted value of the road performance index of the corresponding time step and the road environment data as characteristic values, calculating a residual predicted value by using the trained residual correction model, adding the residual predicted values to obtain a corrected predicted value of the road performance index of the next year, and continuously performing iterative prediction according to the process to realize continuous multi-step prediction of the road performance index.
4) A question-answering system is established, and questions and answers of the user can be answered quickly and accurately. Firstly, establishing a knowledge graph by utilizing python and neo4j, then establishing a semantic analysis model based on an LSTM neural network, extracting entities and relations based on a Bi-SLSTM neural network (bidirectional slice long-short term memory neural network) and a Convolutional Neural Network (CNN), and finally constructing a Cypher query statement and outputting a result.
Compared with the prior art, the invention has the beneficial effects that:
1, through a prediction model and a residual error correction model, which road diseases can be predicted to develop into more serious diseases in the next years. According to a life cycle strategy of a road, life cycle maintenance management is that under the limited maintenance fund scale, a medium-long term maintenance strategy plan covering the life cycle of the road is made by comprehensively considering the current operation situation of a roadbed and the road surface and combining with the currently adopted preventive maintenance technology, the best time is determined for the best road section, effective maintenance measures such as reconstruction, maintenance and improvement are taken for the worst road section, the maximized fund utilization effect and the road condition improvement effect are realized, a user is guided to carry out scientific road maintenance, and the fund waste caused by the poor maintenance method is avoided.
And 2, after the pavement performance is predicted, importing the prediction result into a question-answering system, wherein the question-answering system can answer which diseases may appear on the user road in which time period according to the established knowledge map, the reasons for the diseases, how to avoid the diseases and give a proper road maintenance strategy.
3 can help the user to answer the road maintenance related problems automatically and quickly, and reduce the time wasted by the user for looking up various data.
In a word, the method can efficiently and accurately predict the pavement performance index value in a plurality of years in the future, provides accurate data support for pavement pre-curing work, and improves the maintenance foresight through multi-step prediction. Through the question-answering system, the user can be scientifically guided to implement maintenance work, the utilization rate of maintenance funds is improved, and the service life of the pavement is prolonged.
Drawings
FIG. 1 is a flow chart of GRU neural network prediction correction;
FIG. 2 is a flow chart of knowledge-graph and question-answering system construction.
Detailed Description
The present invention will be described in detail below with reference to the following examples and the accompanying drawings. The specific embodiments are merely illustrative and explanatory of the invention in further detail, and do not limit the scope of the invention as claimed.
The invention relates to a continuous multi-step prediction road intelligent maintenance system based on deep learning, which comprises the following steps:
step 1, data acquisition and preprocessing
Collecting environmental factors and pavement performance index values, and collecting pavement performance index data (I) from the first year to the t + n year after pavement construction1，I2，...，It，...，It+n) And road environment data from the first year to the t + n year after the pavement is built (E)1，E2，...，Et，...，Et+n)。ItRepresents a road surface performance index value of the t-th year, EtIndicating the road environment data of the t-th year. The pavement performance indexes include: road surface damage condition index (PCI), road surface Running Quality Index (RQI), road surface Rut Depth Index (RDI), road surface anti-skid performance index (SRI), and road surface anti-skid performance index (SRI)Structural Strength Index (PSSI), and the like. The environmental data includes temperature, rainfall, traffic volume, etc.
And secondly, carrying out standardization processing on the data, and mapping the data into a (0, 1) interval. A normalized formula:
Figure BDA0003055510870000031
wherein X isnormFor normalized data, x is the raw data. max is the maximum value of the column of x and min is the minimum value of the column of x.
Thirdly, according to the difference of the stake numbers, the data are processed according to the following steps of 7: the scale of 3 is divided into a training set and a test set. Assuming that road surface performance index data and road environment data from the first year to the t + n year of a road section with a shared pile number of 1km to 100km are taken as a training set, the data from 1km to 70km is taken as a training set, (the data of all pile numbers are trained together for years, but the data are not trained together all year by year, a fixed step length is selected each time, the fixed step length refers to the selection of years, namely, the training is performed by using a fixed parameter length, the parameter length is t in the embodiment, and the data from 71km to 100km is taken as a test set. Every 1km is a stake number.
Step 2, establishing a prediction model based on a GRU neural network
GRU neural network formula:
reset gate calculation formula:
Figure BDA0003055510870000032
candidate value
Figure BDA0003055510870000033
The calculation formula of (2) is as follows:
Figure BDA0003055510870000034
the calculation formula for the update gate zt is:
Figure BDA0003055510870000035
hidden state at present time:
Figure BDA0003055510870000036
wherein sigma is a sigmoid function; tan h is a hyperbolic tangent function; h istHidden state at the present moment, ht-1For hidden states transmitted at the last moment, xtIs the state at the current time, ztTo control the gating of updates, rtGating for reset; wr、Wh、WzAnd giving initialization values to the parameters to be trained according to a random initialization method.
Establishing a prediction model based on a GRU neural network, and obtaining (I)1，I2，...，It，E1，E2，...，Et) As an input characteristic value, It+1As the target value. And adjusting the hyper-parameters and training a prediction model.
Obtaining a prediction model of causal relationship between historical road environment data and road performance index data and future road performance index data through training, wherein the model can calculate a target value according to an input characteristic value; the test set is used to test the prediction model, and the hyper-parameters are adjusted until the model meets the prediction accuracy requirement, where the accuracy requirement in this embodiment is that the Mean Absolute Percentage Error (MAPE) is less than 5%. The predicted value of the road surface performance index can be obtained through the prediction model
Figure BDA0003055510870000037
f1Refers to a prediction model of the device,
Figure BDA0003055510870000038
and (4) representing the predicted value of the road surface performance index in the t +1 year.
And 3, calculating a residual error, establishing a residual error correction model based on the GRU neural network, and calculating a road surface performance index correction predicted value.
Firstly, the real value I of the road surface performance index is utilizedt+1Subtracting the predicted value of the road surface performance index
Figure BDA0003055510870000039
Obtaining a residual error, wherein the residual error calculation formula is as follows:
Figure BDA00030555108700000310
Rt+1representing the residual of year t + 1.
Then, building a residual error correction model based on the GRU neural network, and obtaining (I)1，I2，...，It，E1，E2，...，Et) As input characteristic value, Rt+1As the target value. Obtaining a residual error correction model of causal relation among road environment data, road surface performance index data and residual errors through training; and testing the residual error correction model by using the test set, and adjusting the hyperparameter until the precision requirement is met, wherein the precision requirement is that the average absolute percentage error (MAPE) is less than 5%. By residual correction of the modulus
Figure BDA0003055510870000041
The model can obtain residual prediction value
Figure BDA0003055510870000042
f2The model refers to a residual error correction model and represents a residual error predicted value of the t +1 th year. Adding the residual error predicted value and the pavement performance index predicted value to obtain the final pavement performance index correction predicted value
Figure BDA0003055510870000043
Comprises the following steps:
Figure BDA0003055510870000044
and 2, respectively establishing a prediction model and a residual error correction model in the step 3. By using
Figure BDA0003055510870000045
And other existing data, and the predicted value of the pavement performance index correction in the t +2 year is calculated through a prediction model and a residual correction model
Figure BDA0003055510870000046
The calculation process is as follows:
Figure BDA0003055510870000047
Et+1is the original existing data, the collected data includes1，E2，...Et+1，...，Et+n). The length of the parameters that are input into the model each time is fixed, so the first input is E1. The second input is E2.
Figure BDA0003055510870000048
Figure BDA0003055510870000049
Figure BDA00030555108700000410
This process is repeated until calculated
Figure BDA00030555108700000411
In the prediction, the time step is t, the historical data is data of 1 to t + n years, t and n are integers which are more than 1, because the sizes of n and t are not definite, when n < t,
Figure BDA00030555108700000412
the calculation formula of (2) is as follows:
Figure BDA00030555108700000413
Figure BDA00030555108700000414
Figure BDA00030555108700000415
Figure BDA00030555108700000416
when n is greater than t, the first and second groups,
Figure BDA00030555108700000417
the calculation formula of (2) is as follows:
Figure BDA00030555108700000418
Figure BDA0003055510870000051
Figure BDA0003055510870000052
Figure BDA0003055510870000053
according to the method, a prediction model and a residual error correction model can be trained according to the pavement performance indexes of the roads from the 1 st year to the t + n th year and the road environment data. The road performance indexes of continuous fixed time step lengths before the year to be predicted (the road performance indexes can be historical road performance indexes or comprise road performance index predicted values obtained through prediction according to the steps of the application and are determined according to actual time step lengths) and road environment data can be used for continuously predicting the road performance indexes of the next years, multi-step prediction of the road performance indexes is achieved, a residual error correction model is added on the basis of the prediction model, the predicted values can be corrected in time, error propagation can be reduced when continuous multi-step prediction is conducted, and prediction accuracy is improved. In this embodiment, the data of t + n years is known, and the road environment data of the predicted year in actual use can be obtained by prediction in the existing road environment data prediction mode.
And finally, storing the corrected and predicted result of the road performance index to the local, wherein the stored data is the corrected and predicted result of the road performance index of each pile number in each year.
In the above
Figure BDA0003055510870000054
Wherein, yaThe actual value is represented by the value of,
Figure BDA0003055510870000055
indicates the predicted value, where n indicates the number of values.
Step 4, construction of knowledge graph and question-answering system
1) And (5) constructing a knowledge graph. And reading a corrected and predicted result of the road performance index, wherein the predicted result represents the corrected and predicted value of the road performance index in the year of the pile number. Finding a specification related to road maintenance, including: road construction period, road service period and road maintenance specifications, such as but not limited to road engineering technical Standard JTG B01-2014, road technical condition assessment Standard JTG5210-2018, road asphalt pavement maintenance technical Specification JTG 5142 and 2019. And preprocessing the prediction result and the normalized unstructured text data by adopting a rule matching and manual checking mode to obtain structured data and storing the structured data into an EXCEL table. The preprocessed data should be structured data in the form of triples, such as: (node 1, relationship, node 2).
Installing a py2neo tool box, connecting to a neo4j database platform by using python, reading the created EXCEL file, creating nodes and relations into a neo4j platform, and constructing a knowledge graph about road disease maintenance. All nodes and relations in the knowledge graph are constructed into a data set, and then each word is labeled by using a BIO labeling method to generate a BIO entity and relation labeling set.
2) And (6) semantic parsing. To giveDefine a word W ═ c1，c2，...，cmCi represents the ith character in the word, and m represents m characters in the word W. Then the word vector corresponding to the word may be represented as vw, and the character vector sequence corresponding to the word may be represented as:
w＝{vc1，vc2，...，vcn}
encoding the character vector sequence using a Long Short Term Memory (LSTM) neural network, resulting in a character-level vector representation:
vc＝LSTM({vc1，vc2，...，vcn})
the word can eventually be expressed as:
V＝[vw，vc]
vwis a word vector for a word.
3) And (4) performing entity and relation extraction by using the BI-SLSTM-CNN neural network.
3.1) extraction of word features using BI-SLSTM neural networks
For question q ═ { q ═ q1，q2，...，qlThe question sentence can be expressed as V ═ V }1，v2，...，vlAnd l represents the number of words in the question q. Inputting the analyzed question V into a bidirectional long-short term memory (BI-SLSTM) neural network and extracting question sequence characteristics to obtain a hidden layer output vector h at the momenti＝SLSTM(vi，hi-1). Wherein h isi-1The hidden layer state is output at the last moment, and i represents the ith moment of the long-short term memory neural network. And splicing the forward hidden layer state and the reverse hidden layer state to obtain the output h of the bidirectional long-short term memory (Bi-SLSTM) neural network:
Figure BDA0003055510870000061
the output sequence h of Bi-directional slice long-short term memory (Bi-SLSTM) neural network layer is h ═ { h }1，h2，...，hl}。
3.2) predicting whether a word belongs to an entity or a relational vocabulary
The output sequence h of the Bi-SLSTM neural network is a question word sequence, a training set is given, and the training set comprises a certain number of question word sequences h ═ { h { (h) }1，h2，...，hjReal label t ═ t of entities and relations corresponding to these question sentences1，t2，...，tjAnd (6) training to obtain a CNN prediction model by taking the question word sequence as a characteristic value and the real label as a target value. The output of the CNN prediction model is compressed to between 0 and 1 using the sigmoid function. The probability of a certain vocabulary belonging to or entity or relationship can be predicted to be O through a CNN prediction modelk＝σ(W[hk-s，...，hk，...，hk+s]+ b), where W is a parameter of the CNN prediction model, b is a bias term, σ is an activation function, where sigmoid activation function is used, and the expression is:
Figure BDA0003055510870000062
k is the k-th word, and the total number from k-s to k + s is 2s +1, and 2s +1 is the sliding window size of the CNN prediction model. The sliding window can frame the word local context according to the specified unit length, so as to obtain the word context characteristics in the window. Then the probability that the kth word belongs to an entity or relationship component is: p (W)k∈TE|q)＝OkTE refers to entities and relationships. Finally, a probability threshold is set, for example, the threshold is set to 0.5, and when the probability P is greater than 0.5, the word is considered to be an entity or a component of a relationship. Words with a probability less than 0.5 are considered non-entities or relationship components.
4) And inquiring and outputting results. Through semantic analysis and entity and relation extraction, entities and relations involved in the problem can be obtained. And then constructing a Cypher query statement according to the entities and the relations. For example, if the identified Entity is rut and the relationship is a prevention method, a Cypher query statement match (a: Entity) - [ b: relationship ] - (c: Entity) where a name is '< rut >' and b.name is '< prior >' return c.name, for inquiring about a prevention method of rutting (rut). If the user asks multiple questions, the program constructs multiple query statements. Inputting a query statement into a neo4j platform through python to perform query, obtaining answers of the questions, and finally, according to calling a corresponding reply template, for example, if the relation extracted from the questions is a prevention method, replying the prevention method of "×", if the relation extracted from the questions is a reason, replying the reason causing "×", possibly "×", and presenting the result to the user.
The invention introduces a neural network of gated cyclic units (GRU), which is a variant of RNN, and the output of the neural network of gated cyclic units (GRU) is not only influenced by the current input characteristics, but also influenced by the output at the previous moment, so that the neural network has better time sequence performance, and has a reset gate and an update gate. The reset gate determines how the new input information is combined with the previous memorized information, and the update gate determines the information saved to the current time step by the previous memorization, thereby solving the problems of gradient disappearance and gradient explosion. The Slicing Long Short Term Memory (SLSTM) neural network carries out a plurality of slicing processes on an input sequence to form a plurality of minimum subsequences with equal length. Thus, the loop unit can process each sub-sequence simultaneously on each layer, achieve parallel computation, and then perform information transfer through a multi-layer network. The SLSTM neural network can accelerate the speed of extracting entities and relations, and has great advantages particularly when long texts are processed. The BI-directional slice long-short term memory neural network (BI-SLSTM) comprises a forward SLSTM and a backward SLSTM, and the influence of historical information and future information on prediction accuracy is considered.
Nothing in this specification is said to apply to the prior art.

Claims (5)

1. A continuous multi-step prediction road intelligent maintenance system based on deep learning comprises the following contents:
establishing a prediction model based on a GRU neural network, wherein the prediction model can calculate to obtain a future road performance index prediction value according to historical road performance index data and road environment data; the prediction model represents the change rules of the road performance index values at different positions of the road in different years, historical road environment data and the causal relationship between the road performance index data and future road performance index data;
then subtracting the predicted value of the pavement performance index from the real value of the pavement performance index to obtain a residual error;
establishing a residual error correction model based on a GRU neural network, wherein the characteristic value of the residual error correction model is the same as that of the prediction model, namely, the input of the residual error correction model is road surface performance index data and road environment data of a fixed time step length, and the target value of the residual error correction model is a residual error; the residual error correction model is used for predicting a residual error predicted value;
then adding the predicted value of the pavement performance index calculated by the prediction model and the predicted value of the residual calculated by the residual correction model to obtain a corrected predicted value of the pavement performance index, and taking the corrected predicted value of the pavement performance index as the data of the pavement performance index of the corresponding year; obtaining a trained prediction model and a residual error correction model through iterative training;
taking the pavement performance index data and the road environment data of corresponding time step as characteristic values, inputting the trained prediction model to calculate a pavement performance index predicted value of the next year, simultaneously inputting the pavement performance index correction predicted value of corresponding time step and the road environment data as characteristic values into the trained residual correction model to calculate a residual predicted value, adding the residual predicted values to obtain the pavement performance index correction predicted value of the next year, and continuously performing iterative prediction according to the process to realize continuous multi-step prediction of the pavement performance index;
the maintenance system also comprises a knowledge map and a question-and-answer system, wherein after the pavement performance is predicted in multiple steps continuously, the corrected prediction result of the pavement performance index is stored locally, the stored data is the corrected prediction result of the pavement performance index of each stake mark in each year, the prediction result is led into the knowledge map and the question-and-answer system, and the question-and-answer system can answer which kind of diseases possibly occur to a user road in which time period and the reasons for the diseases, how to avoid the diseases and give a proper road maintenance strategy according to the established knowledge map;
the maintenance system can predict which road diseases will develop into more serious diseases in the coming years, comprehensively considers the current operation situation of the roadbed and the pavement and combines with the preventive maintenance technology which can be adopted currently under the condition of limited maintenance fund scale, makes a medium-long term maintenance strategy plan covering the whole life cycle of the pavement, determines the best time for the best road section, and adopts the maintenance measures of reconstruction, maintenance or improvement on the worst road section, thereby realizing the maximized fund utilization effect and road condition improvement effect, guiding the user to carry out scientific road maintenance and avoiding the waste of funds caused by the poor maintenance method.
2. A maintenance system according to claim 1, wherein the construction process of the knowledge-graph and question-and-answer system is: firstly, establishing a knowledge graph by utilizing python and neo4j, then establishing a semantic analysis model based on an LSTM neural network, extracting entities and relations based on a Bi-SLSTM neural network and a Convolutional Neural Network (CNN), and finally establishing a Cypher query statement and establishing a question-and-answer system.
3. A maintenance system according to claim 2, wherein the construction process of the knowledge-graph and question-and-answer system is as follows:
1) establishing a knowledge graph:
reading the corrected and predicted result of the pavement performance index, and searching the standard related to the road maintenance, wherein the standard comprises the following steps: in the road construction period, the road service period and the road maintenance standard, preprocessing unstructured text data in the standard and prediction results in a rule matching and manual inspection mode to obtain structured data and storing the structured data into an EXCEL table; the preprocessed data is structured data in a triple form, and the data comprises nodes and relations;
installing a py2neo tool box, connecting the python to a neo4j database platform, reading the created EXCEL file, creating nodes and relations into a neo4j platform, and constructing a knowledge graph about road disease maintenance; constructing all nodes and relations in the knowledge graph into a data set, and then labeling each word by using a BIO labeling method to generate a BIO entity and relation labeling set;
2) semantic parsing: encoding the character vector sequence by using a long-short term memory (LSTM) neural network to obtain a character-level vector; the character-level vector and the word vector are merged into a word,
3) and (3) extracting entities and relations by using a Bi-SLSTM neural network and a Convolutional Neural Network (CNN):
3.1) extraction of word features using BI-SLSTM neural networks
Inputting the question expressed by words into a BI-directional long-short term memory (BI-SLSTM) neural network, extracting the characteristics of a question sequence, outputting the sequence as a question word sequence,
3.2) predicting whether a word belongs to an entity or a relational vocabulary
Given a training set, the training set contains a large number of question word sequences h ═ h1，h2，...，hjReal label t ═ t of entities and relations corresponding to these question sentences1，t2，...，tjTraining to obtain a CNN prediction model by taking the question word sequence as a characteristic value and the real label as a target value; the output of the CNN prediction model is compressed to be between 0 and 1 by using a sigmoid function, and the probability that a certain vocabulary belongs to or entities or relations can be predicted through the CNN prediction model; setting a probability threshold, wherein when the probability P is larger than the probability threshold, the word is considered as a component of an entity or a relation, otherwise, the word is considered as a non-entity or a relation component;
4) result query and output
The method comprises the steps of obtaining entities and relations involved in questions through semantic analysis and entity and relation extraction, then constructing Cypher query sentences according to the entities and relations, if a plurality of questions are asked by a user, constructing a plurality of query sentences through a program, inputting the query sentences to a neo4j platform through python for query to obtain answers of the questions, and finally displaying results to the user according to calling of a corresponding reply template.
4. The maintenance system of claim 1, wherein the pavement performance indicators include: the road surface damage condition index (PCI), the road surface Running Quality Index (RQI), the road surface Rutting Depth Index (RDI), the road surface anti-skid performance index (SRI) and the road surface structural strength index (PSSI); the road environment data comprises temperature, rainfall and traffic volume; collecting pavement performance index data from the first year to the t + n year after pavement construction and road environment data from the first year to the t + n year after pavement construction, and carrying out standardized processing on the data; and dividing a test set and a training set according to different pile numbers for establishing the GRU neural network.
5. The curing system of claim 1, wherein a prediction model and a residual error correction model are trained according to road surface performance indexes from 1 st year to t + n year of the road and road environment data; and then, multi-step prediction of the road surface performance index is realized by using the trained prediction model and the residual correction model, the residual correction model is added on the basis of the prediction model, the predicted value can be corrected in time, and the error propagation can be reduced when continuous multi-step prediction is carried out.
CN202110498608.1A 2021-05-08 2021-05-08 Continuous multi-step prediction road intelligent maintenance system based on deep learning Active CN113065722B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110498608.1A CN113065722B (en) 2021-05-08 2021-05-08 Continuous multi-step prediction road intelligent maintenance system based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110498608.1A CN113065722B (en) 2021-05-08 2021-05-08 Continuous multi-step prediction road intelligent maintenance system based on deep learning

Publications (2)

Publication Number Publication Date
CN113065722A CN113065722A (en) 2021-07-02
CN113065722B true CN113065722B (en) 2022-07-01

Family

ID=76568342

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110498608.1A Active CN113065722B (en) 2021-05-08 2021-05-08 Continuous multi-step prediction road intelligent maintenance system based on deep learning

Country Status (1)

Country Link
CN (1) CN113065722B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114139460B (en) * 2022-02-01 2022-04-15 四川省公路规划勘察设计研究院有限公司 Method and system for constructing and diagnosing driving comfort evaluation model of continuous downhill road section
CN115841190B (en) * 2023-02-14 2023-06-06 中咨公路养护检测技术有限公司 Road PCI multi-step prediction method and device
CN116362714B (en) * 2023-02-21 2023-11-03 中国公路工程咨询集团有限公司 Prediction method and device for pavement maintenance period

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108335002A (en) * 2017-01-20 2018-07-27 亚信蓝涛（江苏）数据科技有限公司 A kind of visual road maintenance big data analysis system
CN108596396A (en) * 2018-04-28 2018-09-28 中国公路工程咨询集团有限公司 One kind is based on the modified pavement performance prediction of maintenance history and maintenance process and device
CN109918489A (en) * 2019-02-28 2019-06-21 上海乐言信息科技有限公司 A kind of knowledge question answering method and system of more strategy fusions
CN111105332A (en) * 2019-12-19 2020-05-05 河北工业大学 Highway intelligent pre-maintenance method and system based on artificial neural network
CN112527997A (en) * 2020-12-18 2021-03-19 中国南方电网有限责任公司 Intelligent question-answering method and system based on power grid field scheduling scene knowledge graph

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110334211A (en) * 2019-06-14 2019-10-15 电子科技大学 A kind of Chinese medicine diagnosis and treatment knowledge mapping method for auto constructing based on deep learning
CN111737916B (en) * 2020-06-22 2023-07-18 华南农业大学 Road and bridge disease analysis and maintenance decision method based on big data

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108335002A (en) * 2017-01-20 2018-07-27 亚信蓝涛（江苏）数据科技有限公司 A kind of visual road maintenance big data analysis system
CN108596396A (en) * 2018-04-28 2018-09-28 中国公路工程咨询集团有限公司 One kind is based on the modified pavement performance prediction of maintenance history and maintenance process and device
CN109918489A (en) * 2019-02-28 2019-06-21 上海乐言信息科技有限公司 A kind of knowledge question answering method and system of more strategy fusions
CN111105332A (en) * 2019-12-19 2020-05-05 河北工业大学 Highway intelligent pre-maintenance method and system based on artificial neural network
CN112527997A (en) * 2020-12-18 2021-03-19 中国南方电网有限责任公司 Intelligent question-answering method and system based on power grid field scheduling scene knowledge graph

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
残差自修正深度学习集成神经网络在短期电力负荷预测中的应用;陈卓;《中国优秀硕士学位论文全文数据库工程科技II辑》;20190815;论文第4节 *

Also Published As

Publication number Publication date
CN113065722A (en) 2021-07-02

Similar Documents

Publication Publication Date Title
CN113065722B (en) Continuous multi-step prediction road intelligent maintenance system based on deep learning
CN108520155B (en) Vehicle behavior simulation method based on neural network
CN104598611B (en) The method and system being ranked up to search entry
Chen et al. Soft computing methods applied to train station parking in urban rail transit
CN113918538B (en) New road maintenance data migration system based on artificial neural network
CN116050670B (en) Road maintenance decision method and system based on data driving
CN107688701A (en) Water quality hard measurement and body eutrophication evaluation method based on WASP models
CN108334943A (en) The semi-supervised soft-measuring modeling method of industrial process based on Active Learning neural network model
CN108665093A (en) Traffic accidents Severity forecasting method based on deep learning
CN112163669A (en) Pavement subsidence prediction method based on BP neural network
CN115948964A (en) Road flatness prediction method based on GA-BP neural network
CN117151346B (en) Civil aviation specialty teaching training system based on wisdom study
CN114973665A (en) Short-term traffic flow prediction method combining data decomposition and deep learning
CN110795522A (en) Method and device for predicting track position of mobile user
CN114239397A (en) Soft measurement modeling method based on dynamic feature extraction and local weighted deep learning
Kayadelen et al. Effects of maintenance, traffic and climate condition on International Roughness Index of flexible pavement
CN113378479A (en) Intelligent standard method and system based on automatic driving test intelligent platform vehicle
CN108764523A (en) Predictive Methods of Road Accidents based on unbiased nonhomogeneous gray model and geneva model
CN112241808A (en) Road surface technical condition prediction method, device, electronic equipment and storage medium
CN115964483B (en) Highway maintenance big data classification decision method based on neural network
CN107301226A (en) The automatic evaluation method of module is retrieved from a kind of question answering system
CN116306785A (en) Student performance prediction method of convolution long-short term network based on attention mechanism
CN116110219A (en) Traffic accident prediction method
Ming et al. The performance evaluation of expressway PPP project during operation period based on RF
CN114722704A (en) Wheel wear prediction network model training method and device

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