CN119004118B - A method for predicting the life of thermal pipelines - Google Patents
A method for predicting the life of thermal pipelines Download PDFInfo
- Publication number
- CN119004118B CN119004118B CN202411487349.2A CN202411487349A CN119004118B CN 119004118 B CN119004118 B CN 119004118B CN 202411487349 A CN202411487349 A CN 202411487349A CN 119004118 B CN119004118 B CN 119004118B
- Authority
- CN
- China
- Prior art keywords
- data
- time
- pipeline
- feature
- weight
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/14—Pipes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medical Informatics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a thermal pipeline service life prediction method, and relates to the field of machine learning. The thermal pipeline life prediction process comprises the steps of constructing a thermal pipeline related data set, capturing time dynamic characteristics of data through a gate control circulation unit GRU, capturing data characteristics of different angles through a self-adaptive characteristic selection mechanism, capturing global characteristics of the data through a mixed calculation module, fully utilizing key information related to the thermal pipeline life in the data, weighting and fusing all the characteristics to obtain comprehensive information characteristic representation, and calculating to obtain a thermal pipeline life prediction result through a fully connected neural network, so that model prediction accuracy is effectively enhanced.
Description
Technical Field
The invention belongs to the technical field of machine learning, and particularly relates to a thermal pipeline service life prediction method.
Background
The service life of the heating power pipeline refers to the length of time that the heating power pipeline can normally run and has no major faults under the designed use condition, and along with the long-term use of the heating power pipeline, pipeline materials can be gradually aged due to factors such as high temperature, pressure, corrosion, fatigue and the like, so that the bearing capacity of the heating power pipeline is reduced, and finally accidents such as pipeline leakage or cracking and the like can be possibly caused, so that serious potential safety hazards and economic loss are brought, the method has important significance for accurately predicting the service life of the heating power pipeline, can help to discover potential problems in advance, timely maintain or replace the heating power pipeline, avoid major accidents and ensure the stable operation of a heating system.
The current methods in the time series prediction field mainly comprise a traditional statistical model, a machine learning model and a deep learning model, and although the methods show certain advantages in time dependence and trend modeling, the methods generally process data from a single angle, and have difficulty in comprehensively capturing deep influences of the data on the service life of the pipeline from multiple angles, so that prediction accuracy is reduced and adaptability to complex situations is insufficient.
The related data of the thermal pipeline comprise a plurality of key variables which have important influences on the service life of the pipeline, the data are fully analyzed, multi-level information implicit in the data is captured through a plurality of angles, meanwhile, the importance of the data is dynamically adjusted according to different influence degrees of each angle on the target variable, the influence of the data on the service life of the thermal pipeline can be effectively captured, and the prediction precision and the robustness of the model in coping with complex conditions are enhanced.
Disclosure of Invention
The invention provides a thermal pipeline life prediction method, which aims to model the time dependence of data by using a gate control circulation unit GRU so as to better capture time dynamic characteristics in thermal pipeline data, capture multi-level information implicit in the data from a plurality of angles, dynamically adjust the importance of the information of each angle by combining a self-adaptive weight selection mechanism, thereby extracting more representative and effective data characteristics, obtain the sharing information of the data by a mixed calculation strategy on the basis, further improve the prediction capability of a model by combining global and local characteristics, and finally carry out weighted fusion on all the data characteristics so as to realize the accurate prediction of the thermal pipeline life.
The technical method adopted by the invention for achieving the purpose comprises the following steps:
s1, collecting related data of a thermal pipeline, including temperature, humidity, pressure, corrosion degree of the inner surface of the pipeline, corrosion degree of the outer surface of the pipeline, pipeline material, operation time length, periodic load and soil pH value, preprocessing the collected data, and providing a reliable data base for the subsequent life prediction of the thermal pipeline;
S2, performing feature embedding on the preprocessed data by utilizing a feedforward neural network, and mapping the input to a high-dimensional feature space;
s3, further capturing time dependence of data through a gate control circulation unit GRU, wherein the GRU selectively reserves or forgets information from a long-time sequence by utilizing a reset gate and update gate mechanism, so that the model can process the long-time sequence data, and simultaneously reserves important historical information related to life prediction of a thermal pipeline so as to accurately model time dynamic characteristics;
s4, constructing an adaptive feature selection mechanism, firstly capturing data features from different angles by utilizing a plurality of feedforward neural networks, and then dynamically fusing feature data from different angles by utilizing adaptive weights to ensure that the influence of key variables on the service life of a pipeline is fully modeled;
s5, constructing a hybrid computing module, which is used for capturing context information and processing commonalities among the different angle characteristics so as to obtain global characteristics among the different angle characteristics, so that the model can consider the global and local characteristics;
s6, fusing various data information, comprehensively considering the importance of each data characteristic, and generating a final unique characteristic representation for predicting the service life of the heating power pipeline;
And S7, further calculating the weighted and fused characteristic representation through the fully-connected neural network to generate a thermal pipeline life prediction result.
Preferably, in the step S1, the data related to the thermal pipeline is collected, including temperature, humidity, pressure, corrosion degree of the inner surface of the pipeline, corrosion degree of the outer surface of the pipeline, pipeline material, operation duration, periodic load and ph value of soil, and the missing values in the data set are complemented by using a mean filling method.
Preferably, in the step S2, a first step is givenHistorical observation sequence of seed dataWhereinFor the time step of the observation,Is the firstSeed data firstThe observation value of time adopts a point-by-point marking method, the observation value of each time point through a feedforward neural network is converted into a high-dimensional vector which can be processed by a model, and the specific calculation process is as follows:
;
In the middle of Is thatThe resulting high-dimensional vector representation of the mapping,As a matrix of weight parameters that can be learned,For the function to be activated by the ReLU,Is the dimension of the data.
Preferably, compared with the traditional low-dimensional representation mode, the high-dimensional representation can capture more implicit mode and structure information, each observation point in the time sequence is mapped into the high-dimensional vector representation through the feedforward neural network, the feature extraction capacity of the model on the sequence data is effectively enhanced, the gradient disappearance problem can be avoided by adopting the ReLU activation function, and the model is ensured to keep higher training efficiency in the deep learning process.
Preferably, in the step S3, the dependency of the high-dimensional vector on the time dimension is captured by using a gate-controlled loop unit GRU, and the time dynamic feature is modeled, and the specific calculation process is as follows:
;
;
;
;
In the middle of Is the firstA reset gate of the time step,Is the firstThe update gate of the time step is updated,Is the firstThe candidate hidden state of the time step,Is the firstThe hidden state of the time step,、、、、、As a parameter of the weight-bearing element,、、As a parameter of the bias it is possible,As a function of the sigmoid,Is Hadamard product.
Preferably, the GRU passes through the reset gate by introducing the gate control circulating unitUpdate doorThe mechanism dynamically adjusts the updating of the hidden state, avoids the problem of forgetting or excessive accumulation of information, can effectively process the history data of long-time span in the life prediction of the thermal pipeline, has less parameter quantity compared with the traditional circulating neural network, has higher training efficiency, can better process the problem of long-time dependency, enhances the capturing capability of long-term trend and short-term change, and further improves the robustness and accuracy of the life prediction of the pipeline.
Preferably, the step S4 is constructed firstThe characteristic capture feedforward neural network of each angle, in order to obtain the characteristic representation of each data under different angles, the concrete calculation process is as follows:
;
In the middle of Is the firstSeed data firstTime of day by the firstThe features of the individual angles capture a representation of the features obtained by the feed-forward neural network,To be used for capturing the firstA network of individual angular features,For the dimension of the data, the weight of each angle is calculated, and the characteristic representation of different angles is fused dynamically, wherein the specific calculation process is as follows:
;
;
In the middle of Is the firstSeed data at the firstThe weight distribution of each angle at the moment,As a parameter of the weight-bearing element,Is the firstSeed data at the firstTime of day (time)The weight of the individual angles is such that,Is the firstSeed data at the firstAnd finally fusing the multiple angles at the moment to obtain the representation.
Preferably, different characteristic representations of data are extracted from a plurality of angles by utilizing a feedforward neural network, multidimensional information in each data can be fully captured, then the characteristics of different angles are weighted and fused by dynamically calculating the weight of each angle, so that the characteristics of key angles are ensured to get more attention in final representation.
Preferably, in step S5, a hybrid computing module is constructed, and the hybrid computing module is configured to capture context information and process commonalities between features of different angles, extract a global feature representation and combine the global feature representation with the feature representation after multi-angle fusion to obtain a final feature representation, where a specific computing process is as follows:
;
;
;
In the middle of Is the firstSeed data at the firstThe global feature representation of the time of day sharing,In order to feed-forward the neural network,The weights that are occupied for the global feature representation,As a parameter of the weight-bearing element,Is the firstSeed data at the firstThe final characteristic representation of the moment in time.
Preferably, the global features in the data are effectively captured through the hybrid computing module, so that the common information among the features at different angles can be fully reflected, the introduction of the global features provides a macroscopic view for the model, the capturing of the context information is facilitated, in addition, the weight of the global features is dynamically adjusted through the Sigmoid function, the importance of the global features can be automatically distributed according to the data by the model, the weighted fusion of the global features and the multi-angle features is realized, and the prediction precision and the robustness of the model under a complex scene are improved.
Preferably, in the step S6, the feature vectors of each data at all time points are weighted and fused to obtain an overall feature vector representation of each data, and then the feature vectors of all data types are further weighted and fused to obtain a final feature vector for predicting the life of the thermal pipeline, which is specifically calculated as follows:
;
;
In the middle of As a feature vector for final use in thermodynamic pipe life prediction,Is the firstTime-series data of the seed data,Is the firstSeed data firstThe feature vector of the moment in time,The number of data types is determined,For the time step size of the time step,Is the firstTime-series data weights of the seed data,Is the firstSeed data firstThe weight of the moment in time is that,Is a weight parameter.
Preferably, in the step, the feature vectors in the time dimension are weighted and fused, so that the model can capture important time point information in the time sequence, the time dimension is normalized through a Softmax function, and the contribution of each time point can be dynamically adjusted by the model, so that the data characteristics of the key time points are highlighted, the information at the key time is ensured to have larger influence on model prediction, the model is facilitated to comprehensively process multi-source data, and the accuracy and the robustness of thermal pipeline life prediction are improved.
Preferably, in the step S7, the following steps are performedCalculating to obtain a final predicted value of the service life of the thermal pipeline through a fully-connected neural network:
;
In the middle of In order to fully connect the neural network,And the final life prediction value of the heating power pipeline.
Compared with the prior art, the method has the beneficial effects that dynamic information in the time dimension is effectively captured through the gate control circulation unit GRU, modeling capacity of the model on long-term dependency in a time sequence is enhanced, then, a self-adaptive weight selection mechanism is introduced to extract potential information in each data from multiple angles, and the characteristics of different angles are subjected to weighted fusion, so that important information is highlighted in a self-adaptive manner, then, shared information among different angles is further extracted through a mixed calculation module, the cognition capacity of the model on a global data mode is enhanced, finally, the characteristics of all the data are fused by the model, and a characteristic vector containing rich context information is generated, so that the accuracy and the robustness of life prediction of a thermal pipeline are remarkably improved.
Drawings
FIG. 1 is a step diagram of a thermal conduit life prediction method.
FIG. 2 is a gate cycle unit GRU diagram.
Fig. 3 is a schematic diagram of adaptive weight selection.
Fig. 4 is a block diagram of a hybrid calculation.
Fig. 5 is a graph of thermal pipeline life prediction effects.
Detailed Description
The invention provides a thermodynamic pipeline life prediction method, which is characterized in that a gate control circulation unit GRU effectively captures dynamic information in a time dimension, potential information in each data is extracted from a plurality of angles, characteristics of different angles are subjected to weighted fusion by utilizing a self-adaptive weight selection mechanism, shared information among different angles is further extracted by a mixed calculation module, finally, the characteristics of all the data are fused by a model to generate a characteristic vector containing abundant context information, and the thermodynamic pipeline life is accurately predicted by utilizing a fully connected network.
S1, collecting 90-day thermal pipeline related data, wherein the data comprise 9 types of data including temperature, humidity, pressure, pipeline inner surface corrosion degree, pipeline outer surface corrosion degree, pipeline material, operation time length, cycle load and soil pH value, selecting data of 5 days before and after data are subjected to mean filling, if the data of 5 days before and after under boundary conditions are used, filling the data by using the mean of available days, dividing a training set and a testing set according to the ratio of 8:2, and then adopting a sliding window method, wherein input data of each day is composed of the thermal pipeline related data of the first 14 days and used as a basis for model learning and prediction to ensure that a model can capture dynamic changes in time dimension.
S2, embedding observed values in all data by using a point-by-point marking method and utilizing a feedforward neural network, so that input is mapped to a high-dimensional feature space.
Further, in the step S2, for each dataGiven a historical observation sequenceWhereinFor the time step of the observation,Is the firstSeed data at time pointIn order to convert the observed value of each time point into a high-dimensional vector which can be processed by a model, a point-by-point marking method is adopted, and the conversion is realized by using a feedforward neural network, wherein the specific calculation process is as follows:
;
In the middle of Is thatThe resulting high-dimensional vector representation of the mapping,As a matrix of weight parameters that can be learned,Activating the function for the ReLU.
S3, further capturing the time dependence of the data through a gate control circulation unit GRU, wherein the GRU is an improved circulation neural network, and the information from a long time sequence is selectively reserved or forgotten through a mechanism of resetting a gate and updating the gate, so that the model can effectively process key historical information related to the life prediction of the heating power pipeline in the long time sequence data.
Further, in the step S3, as shown in fig. 2, the dependency of the high-dimensional vector on the time dimension is captured by using a gate control loop unit GRU, and the GRU selectively retains or forgets information by resetting the gate and updating the gate, so as to enhance the capability of the model to process the time sequence, and the specific calculation process is as follows:
;
;
;
;
In the middle of Is the firstA reset gate of the time step,Is the firstThe update gate of the time step is updated,Is the firstThe candidate hidden state of the time step,In the state of being in a hidden state,、、、、、The parameters that are learned are trained for the model,、、As a parameter of the bias it is possible,As a function of the sigmoid,Is Hadamard product, i.e. per-element product.
And S4, constructing a self-adaptive feature selection mechanism, firstly capturing data features from different angles by utilizing a plurality of feedforward neural networks, and weighting the features extracted from each feedforward neural network by using a self-adaptive weight distribution mechanism so as to form a final fusion feature representation.
Further, in the step S4, as shown in fig. 3, first, a structure is constructedIndividual feed forward neural networks to captureCharacteristic of the angle, inThe specific calculation process is as follows:
;
In the middle of To pass through the firstThe first angle of the feed-forward neural networkSeed data at the firstThe characteristic of the time of day is indicative,As a dimension of the data it is,Is the firstA feed-forward neural network of individual angles, followed by for each time stepWeights under different angles are generated, feature representations of the different angles are dynamically fused, and the specific calculation process is as follows:
;
;
In the middle of Is the firstSeed data at the firstThe weight distribution of each angle at the moment,As a parameter of the weight-bearing element,Is the firstSeed data at the firstTime of day (time)The weight of the individual angles is such that,Is the firstSeed data at the firstAnd finally fusing the multiple angles at the moment to obtain the representation.
S5, constructing a hybrid computing module, and usingThe feedforward neural network captures commonalities among the different angle features to obtain global features among the different angle features, so that the model can consider the global and local features at the same time.
Further, in step S5, as shown in fig. 4, a hybrid computing module is constructed, global features are extracted and dynamically weighted and fused with feature representations fused at different angles, firstly, the global features are extracted through a feedforward neural network, and then an adaptive weight is generated for controlling the importance of the global features in the final feature fusion, and the specific computing process is as follows:
;
;
;
In the middle of Is the firstSeed data at the firstThe global feature representation of the time of day sharing,The weights that are occupied for the global feature representation,Is the firstSeed data at the firstThe final characteristic representation of the time of day,As a parameter of the weight-bearing element,Is a feed-forward neural network.
And S6, fusing various data information, comprehensively considering the characteristic information of different data sources, and generating a unique characteristic representation according to the importance of each data characteristic for predicting the service life of the thermal pipeline.
Further, in step S6, firstly, the weight of each time series feature vector corresponding to each data is calculated, the time series feature vectors are weighted and fused according to the weight, after the fusion of a single data type is completed, the feature vectors of all data types are further weighted and fused, and finally, the final feature vector for predicting the service life of the thermal pipeline is obtained, and the specific calculation process is as follows:
;
;
In the middle of Is the firstTime-series data weights of the seed data,Is the firstSeed data firstThe weight of the moment in time is that,As a parameter of the weight-bearing element,As a feature vector for final use in thermodynamic pipe life prediction,Is the firstTime-series data of the seed data,Is the firstSeed data firstThe feature vector of the moment in time,The number of data types is determined,In time steps.
And S7, further calculating and processing the weighted and fused feature vectors through the fully connected neural network, so as to generate a thermal pipeline life prediction result.
Further, in the step S7Calculating to obtain a final predicted value of the service life of the thermal pipeline through a fully-connected neural network:
;
In the middle of In order to fully connect the neural network,And the final life prediction value of the heating power pipeline.
Further, the method is realized by adopting Python 3.8, the model is constructed and trained based on PyTorch framework and operates in CUDA11 environment to fully utilize GPU to accelerate calculation, the whole training process operates on NVIDIA RTX 3090 GPU, the batch size is 64 in the training process, and the learning rate is 64By usingAnd an optimizer.
Further, the prediction effect of the method is shown in fig. 5, the ordinate is the service life (day) of the thermal pipeline, the abscissa is the time (day), the gray straight line represents the service life of the real thermal pipeline, the black straight line represents the prediction value of the service life of the thermal pipeline, and the prediction result is highly consistent with the real value, so that the method has higher fitting degree and accuracy in the prediction of the service life of the thermal pipeline, and the effectiveness of the method is verified.
The foregoing is merely a preferred embodiment of the present invention, and it should be noted that modifications and improvements could be made by those skilled in the art without departing from the inventive concept, which fall within the scope of the present invention.
Claims (2)
1. The service life prediction method of the heating power pipeline is characterized by comprising the following steps of:
S1, collecting related data of a thermal pipeline, wherein the related data comprises 9 types of data, namely temperature, humidity, pressure, corrosion degree of the inner surface of the pipeline, corrosion degree of the outer surface of the pipeline, pipeline material, operation duration, periodic load and soil pH value, preprocessing the acquired data, and providing a reliable data base for the subsequent life prediction of the thermal pipeline;
S2, utilizing a feedforward neural network to perform feature embedding on the preprocessed data in a point-by-point marking mode, and mapping observed values of all the data to a high-dimensional feature space, wherein the method comprises the following specific steps of:
Input all types of data, taking the ith data as an example, its T time step historical observation sequence For the observed value of the ith data t E [1, T ] moment, converting the observed value of each time point into a high-dimensional vector which can be processed by a model by adopting a point-by-point marking method through a feedforward neural network, wherein the specific calculation process is as follows:
vi,t＝ReLU(Wxi,t);
In the middle of The resulting high-dimensional vector representation mapped for x i,t,As a learnable weight parameter matrix, reLU (·) is a ReLU activation function, and H is the dimension of data;
S3, processing long-time sequence data through a gate control circulation unit GRU, selectively reserving or forgetting information from the long-time sequence by utilizing a reset gate and update gate mechanism, reserving important historical information related to life prediction of a thermal pipeline, and accurately modeling time dynamic characteristics, wherein the method comprises the following specific steps of:
modeling time dynamic characteristics by utilizing the dependency of the capturing high-dimensional vector of the gate control loop unit GRU on the time dimension, and firstly utilizing a reset gate and an update gate to selectively reserve or forget information from a long time sequence, wherein the specific calculation process is as follows:
ri,t＝σ(vi,tWvr+hi,t-1Whr+br);
zi,t＝σ(vi,tWvz+hi,t-1Whz+bz);
In the middle of For the reset gate of the t-th time step,For the update gate of the t-th time step,B r as weight parameter,For the bias parameter, σ is a sigmoid function, and then the hidden vector of the gate control loop unit GRU is calculated according to the reset gate and the update gate, and the specific calculation process is as follows:
And Is the candidate hidden state of the t-th time step,In the hidden state of the t time step, W vh,As a parameter of the weight-bearing element,As bias parameters, σ is a sigmoid function, and as Hadamard product;
s4, constructing a self-adaptive feature selection mechanism, capturing features of different angles of data by using a plurality of feedforward neural networks, and then learning weights of all angles to fuse the features of the data of the different angles, so as to ensure that the influence of key variables on the service life of the pipeline is fully modeled, wherein the method comprises the following specific steps of:
firstly, constructing a characteristic capturing feedforward neural network of N angles to obtain characteristic representations of each data under different angles, then calculating the weight of each angle, and dynamically fusing the characteristic representations of different angles, wherein the specific calculation process is as follows:
ui,t,j＝FFNj(hi,t);
si,t＝softmax(Wshi,t);
In the middle of For a feature representation obtained by the feature capture network of the jth angle at the ith moment of time, FFN j (·) is the network used to capture the features of the jth angle, D is the dimension of the data,For the weight distribution of the ith data at each angle at the t-th moment,As a parameter of the weight-bearing element,For the weight of the ith data at the jth angle at the t-th time,S5, constructing a mixed calculation module, capturing commonalities among different angle features by using 1 feedforward neural network to obtain global features among the different angle features, adding the weighted global features and the data features fused with the different angles, so that the model can consider the global and local features at the same time, and the specific steps are as follows:
the method comprises the steps of constructing a hybrid computing module, capturing context information, processing global feature representations among different angle features, and carrying out dynamic weighted fusion on the feature representations fused with different angles to obtain the representation of the global and local features of the local function, wherein the specific computing process is as follows:
gi,t＝FFN(hi,t);
si,t,g＝Sigmoid(Wghi,t);
yi,t＝si,t,ggi,t+fi,t;
In the middle of For the global feature representation of the ith data obtained at time t, FFN (-) is a feed-forward neural network,The weights that are occupied for the global feature representation,As a parameter of the weight-bearing element,A final characteristic representation of the ith data at time t;
S6, fusing various data information, comprehensively considering the importance of each data characteristic, and generating a final unique characteristic representation for predicting the service life of the thermal pipeline, wherein the specific steps are as follows:
Firstly, carrying out weighted fusion on all final feature representations of time series data of all data to obtain an overall feature vector representation of each data, and then, further carrying out fusion on the overall feature vector representations of all data to obtain a comprehensive feature representation, wherein the specific calculation process is as follows:
wi＝softmax(yiWi);
In the middle of The time-series data weight for the i-th data,The weight at time t for the ith data,As a parameter of the weight-bearing element,As a feature vector for final use in thermodynamic pipe life prediction,Is time-series data of the ith data,The characteristic vector is the characteristic vector of the ith data at the T moment, M is the number of data types, and T is a time step;
and S7, further calculating the weighted and fused characteristic representation through the fully-connected neural network, and generating a thermal pipeline life prediction result.
2. The method for predicting the service life of a thermal pipeline according to claim 1, wherein, for the problem of predicting the service life of the thermal pipeline, data related to the thermal pipeline are collected, including temperature, humidity, pressure, corrosion degree of the inner surface of the pipeline, corrosion degree of the outer surface of the pipeline, pipeline material quality, operation duration, cyclic load and soil pH value, when the missing data is processed, data of 5 days before and after are adopted for average filling, if the data of 5 days before and after are less than under the boundary condition, the average value of available days is used for filling, and then the processed data set is divided into a training set and a verification set to be used as the basis for model learning and predicting the service life of the thermal pipeline.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411487349.2A CN119004118B (en) | 2024-10-24 | 2024-10-24 | A method for predicting the life of thermal pipelines |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411487349.2A CN119004118B (en) | 2024-10-24 | 2024-10-24 | A method for predicting the life of thermal pipelines |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN119004118A CN119004118A (en) | 2024-11-22 |
| CN119004118B true CN119004118B (en) | 2025-01-24 |
Family
ID=93482057
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411487349.2A Active CN119004118B (en) | 2024-10-24 | 2024-10-24 | A method for predicting the life of thermal pipelines |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119004118B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120217874A (en) * | 2025-03-20 | 2025-06-27 | 北京林业大学 | Remaining life prediction and maintenance method of heating pipelines based on deep reinforcement learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705176A (en) * | 2019-09-02 | 2020-01-17 | 北京市燃气集团有限责任公司 | Method and device for predicting residual life of gas pipeline |
| CN112182976A (en) * | 2020-10-12 | 2021-01-05 | 上海交通大学 | Method for predicting residual life of industrial equipment |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3671568A1 (en) * | 2018-12-17 | 2020-06-24 | IMEC vzw | Binary recurrent neural network inference technique |
| CN110377686B (en) * | 2019-07-04 | 2021-09-17 | 浙江大学 | Address information feature extraction method based on deep neural network model |
| US20230109096A1 (en) * | 2020-12-18 | 2023-04-06 | Strong Force Vcn Portfolio 2019, Llc | Maintenance Prediction and Health Monitoring for Robotic Fleet Management |
| EP4323940A4 (en) * | 2021-04-16 | 2025-02-19 | Strong Force VCN Portfolio 2019, LLC | Systems, methods, kits, and apparatuses for digital product network systems and biology-based value chain networks |
| CN115330094B (en) * | 2022-10-14 | 2023-04-07 | 成都秦川物联网科技股份有限公司 | Smart Gas Pipeline Life Prediction Method, Internet of Things System, Device and Medium |
| CN115618733B (en) * | 2022-10-24 | 2023-04-07 | 大连理工大学 | Multi-scale hybrid attention mechanism modeling method for predicting remaining service life of aircraft engine |
| CN116842379A (en) * | 2023-06-06 | 2023-10-03 | 山东省计算中心（国家超级计算济南中心） | A method for predicting the remaining service life of mechanical bearings based on DRSN-CS and BiGRU+MLP models |
| CN117538783A (en) * | 2023-11-23 | 2024-02-09 | 贵州金元绿链物流开发有限公司 | A lithium-ion battery state-of-charge estimation method based on time-domain fusion converter |
| CN118232355B (en) * | 2024-05-24 | 2024-08-09 | 山东和光智慧能源科技有限公司 | Depth peak shaving method and system based on Internet of things thermoelectric cooperation |
| CN118296973B (en) * | 2024-06-06 | 2024-07-30 | 山东和光智慧能源科技有限公司 | An intelligent leakage detection system for heating pipelines based on the Internet of Things |
-
2024
- 2024-10-24 CN CN202411487349.2A patent/CN119004118B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705176A (en) * | 2019-09-02 | 2020-01-17 | 北京市燃气集团有限责任公司 | Method and device for predicting residual life of gas pipeline |
| CN112182976A (en) * | 2020-10-12 | 2021-01-05 | 上海交通大学 | Method for predicting residual life of industrial equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN119004118A (en) | 2024-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115688579B (en) | Drainage basin multipoint water level prediction and early warning method based on generation countermeasure network | |
| CN112149879B (en) | A new energy medium and long-term electricity forecast method considering macro-volatility classification | |
| CN114648176B (en) | Wind-solar power supply consumption optimization method based on data driving | |
| CN119004118B (en) | A method for predicting the life of thermal pipelines | |
| CN111191835A (en) | IES incomplete data load prediction method and system based on C-GAN transfer learning | |
| CN111915092B (en) | Ultra-short-term wind power forecasting method based on long-short-term memory neural network | |
| CN114119273B (en) | Non-invasive load decomposition method and system for park comprehensive energy system | |
| CN113128113A (en) | Poor information building load prediction method based on deep learning and transfer learning | |
| CN109978283A (en) | A power prediction method of photovoltaic power generation based on branch evolutionary neural network | |
| CN114811713B (en) | Two-level network inter-user balanced heat supply regulation and control method based on mixed deep learning | |
| Kalogirou | Artificial neural networks and genetic algorithms for the modeling, simulation, and performance prediction of solar energy systems | |
| CN116300755B (en) | MPC-based two-layer optimization scheduling method and device for heating system with heat storage | |
| CN119904057A (en) | Dynamic scheduling method for integrated energy system based on multi-agent deep reinforcement learning | |
| Zhang et al. | Day-ahead wind power prediction using an ensemble model considering multiple indicators combined with error correction | |
| CN110119846A (en) | A kind of multiple target deepness belief network for mid-term electric load forecasting | |
| Wei et al. | A combination forecasting method of grey neural network based on genetic algorithm | |
| CN120049414A (en) | New energy power interval prediction method for association of excavation seasonal variation characteristics and meteorological characteristics | |
| Prema et al. | LSTM based deep learning model for accurate wind speed prediction | |
| CN118761691A (en) | Smart energy industry Internet data management method, system, medium and server | |
| Khan et al. | Towards net zero energy building: AI-based framework for power consumption and generation prediction | |
| Ou et al. | Comparative analysis of offshore wind power prediction models and clustering-based daily output classification | |
| CN117913778A (en) | Short-term wind power prediction method, system, electronic equipment and storage medium | |
| CN113836819B (en) | Bed temperature prediction method based on time sequence attention | |
| Lv et al. | Research on short-term load forecasting approach for smart grid | |
| Ferreira et al. | Application of Bio-Inspired Optimization Techniques for Wind Power Forecasting |
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 |