CN-121981332-A - Port equipment maintenance decision optimization method and system with edge cloud cooperation

CN121981332ACN 121981332 ACN121981332 ACN 121981332ACN-121981332-A

Abstract

The invention discloses a port equipment maintenance decision optimization method and system with edge cloud cooperation, and relates to the technical field of port equipment maintenance. The method comprises the steps of collecting multisource heterogeneous data through an edge terminal, extracting feature vectors, constructing a semantic dynamic digital twin model, constructing a lightweight analysis model based on GCN architecture embedded physical constraints, deploying the lightweight analysis model on an edge gateway to output equipment health indexes, fault modes and data quality scores in real time, uploading relevant data and local sub-models to a cloud end at the edge side when the data quality does not reach the standard, constructing a target optimization model containing structural constraints by combining a PINN algorithm with a physical model library, solving a multi-target optimization problem by improving an NSGA-II algorithm, and generating an optimal maintenance scheme considering maintenance cost, equipment availability and fault risk. The invention integrates physical constraint and digital twin technology to ensure the reliability of decision, effectively improves the real-time performance and accuracy of maintenance decision, and is suitable for the high-efficiency operation and maintenance of special port equipment.

Inventors

LI XIANRUI
LIU JIAN
LIU LEILEI
XU BIN
ZHAO HAOXU
LI BINGSHUAI
Zhou Sanbo
WANG ZHE
ZHANG LEI
LI ZHANGYUN

Assignees

交通运输部天津水运工程科学研究所

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (7)

1. The port equipment maintenance decision optimization method with the edge cloud cooperation is characterized by comprising the following steps of: The edge terminal deploys a multi-type sensor, the edge terminal collects multi-source heterogeneous data and transmits the multi-source heterogeneous data to an edge gateway, the edge gateway carries out preprocessing on the multi-source heterogeneous data, extracts time domain, frequency domain, time-frequency domain and physical characteristics, and generates a structural feature vector; based on the structural feature vector, combining an improved FCA algorithm with a cross-domain bridge ontology mapping rule, extracting candidate nodes, and constructing a semantic dynamic digital twin model; taking the semantic dynamic digital twin model as a structural basis, adopting GCN as an infrastructure, embedding physical constraint, constructing a lightweight analysis model, deploying the lightweight analysis model on an edge gateway, and outputting equipment health index, fault mode and data quality score; Continuously monitoring the data quality through the lightweight analysis model, and uploading the corresponding structured feature vector, the output of the lightweight analysis model and the local submodel of the semantic dynamic digital twin model to a cloud platform if the data quality score of the continuous 3 monitoring periods is lower than a quality threshold; And the cloud platform receives the data, and builds a target optimization model by adopting PINN algorithm and combining a physical model library to generate an optimal maintenance scheme.
2. The method for optimizing maintenance decisions of a port equipment with edge-cloud cooperation as set forth in claim 1, wherein the extracting candidate nodes comprises performing discretization on historical sensor data and fault records to construct a formal background matrix S= (U, A, I), wherein U is a set of equipment core components, A is a set of attributes, is a union of fault modes, fault causes and monitoring features, and I is an association relation I If the actual association exists between the component U and the attribute a, the value of (U, a) epsilon I is recorded as 1, otherwise, the value is 0, the association strength of the attribute is quantized by adopting cosine similarity, a similarity threshold is set, the attributes exceeding the similarity threshold are combined, the inclusion comprises equipment components, a fault mode, a fault cause and monitoring characteristics, four-element candidate nodes (C, F, R and M) are extracted and formed, and the monitoring characteristics strongly associated with the fault mode are screened through mutual information entropy to serve as attribute nodes.
3. The port equipment maintenance decision optimization method based on edge cloud cooperation as claimed in claim 2, wherein the construction of the lightweight analysis model comprises the specific steps of model structural design and physical constraint embedding: The method comprises the steps of designing a lightweight analysis model structure, taking four-tuple candidate nodes (C, F, R, M) of a semantic dynamic digital twin model as entity nodes, taking the screened core monitoring characteristics as attribute nodes, and constructing initial edge connection according to a physical connection relation and a functional influence relation of equipment parts to form a GCN basic topological structure; physical constraint embedding, extracting key physical rules of equipment operation, converting the key physical rules into constraint terms to be integrated into a GCN loss function, wherein the formula of the total loss function is as follows: ; Wherein, the Cross entropy loss is classified for the failure mode, The physical constraint loss is obtained by calculating the residual error between the model predicted value and the theoretical value of the physical equation N is the number of samples, and, In order to lighten the predictive value of the analysis model, For the physical equation, θ is the set of physical parameters, Is an input feature.
4. The port equipment maintenance decision optimization method based on the edge cloud cooperation of claim 3, wherein the output layer structure of the lightweight analysis model comprises the steps of constructing three branch output layers based on a GCN framework, wherein the three branch output layers correspond to equipment health indexes, fault modes and data quality scores respectively, and an output layer activation function sequentially adopts Sigmoid, softmax, linear; the equipment health index is based on the component hierarchy relation of the semantic dynamic digital twin model, and a weighted summation formula is adopted to calculate the global health index: ; ; wherein K is the number of core components of the device, As the importance weight of the kth component, For the local health index of the kth component, deriving inferentially based on sensor characteristics and physical constraints, The result of the normalization of the sensor feature vector for the kth component, In order for the physical constraints to be satisfied, 、 Training parameters for the model; the model outputs probability distribution of M preset fault modes Satisfies the following conditions Taking the first 2 fault modes with the highest probability as recognition results; Scoring formula for data quality scoring , wherein, As an indicator of the integrity of the data, As an index of the consistency of the data, As an index of the rationality of the data, Is the corresponding index weight.
5. The method for optimizing maintenance decisions of a harbour site with edge cloud cooperation as claimed in claim 4, wherein said data quality score calculation includes three-dimensional indexes of integrity, consistency and rationality, and the specific steps and formulas are as follows: data integrity index , In order to actually receive the number of valid data points, The number of the data points is theoretically expected; Constructing a sensor data association matrix based on equipment physical logic by using data consistency indexes M is the number of sensors, and the consistency coefficient is calculated , For the theoretical correlation value of sensors i and j, As a result of the actual measurement value, 、 The maximum and minimum reasonable ranges of the association values are respectively; Data rationality index , For the data point of the g-th data point, Is a reasonable interval of data.
6. The method for optimizing maintenance decisions of port equipment by edge cloud cooperation as claimed in claim 4, wherein said adopting PINN algorithm to construct the target optimization model by combining with the physical model library specifically comprises: the structured feature vector is processed N is a characteristic dimension, normalization processing is carried out, and a normalization formula is as follows: , as the mean value of the i-th dimensional feature, Standard deviation of the i-th dimensional feature; extracting the equipment health index HI and the probability of a main fault mode output by the lightweight analysis model Data quality score Q, build initial state vector As a priori constraint term for PINN models; converting from the partial submodel of the semantic dynamic digital twin model to obtain PINN structural constraints including component association constraints and fault propagation constraints; the PINN model includes an input layer, a hidden layer, a physical constraint layer, and an output layer, the input layer receives the adapted structured feature vector X' and the initial state vector The hidden layer adopts a ReLU activation function, the physical constraint layer is embedded with four types of constraints which are semantic constraint, mechanism constraint, bearing constraint and structural constraint of a physical model library respectively, an Adam optimizer is adopted to set learning rate, iterative training is carried out until a loss function converges, a device fault evolution prediction model is obtained, and a fault evolution trend is output ; In the construction of the multi-objective optimization model, the maintenance cost is defined T is the total number of maintenance cycles, Is the cost of the unit of labor, The number of people is maintained for the t-th period, In order to be a unit cost of the spare parts, For the number of spare parts for the t-th cycle, The unit cost is lost for the shutdown, For the time length of the machine halt of the t-th period, the availability of the equipment Failure risk, failure probability based on PINN prediction And (3) calculating: K is the total number of failure modes, Is the severity coefficient of the kth fault; maintaining resource constraints , , For the maximum number of maintenance persons available, Upper limit of stock for spare parts, job plan constraint , Setting state variable threshold value for maximum allowable machine halt time length of single period based on component bearing limit of local submodel by digital twin constraint ; Generating N maintenance scheme individuals conforming to maintenance resource constraint, operation plan constraint and digital twin constraint, calculating three objective functions corresponding to each individual 、、 And calculating a total penalty value according to whether the individual violates three types of constraints, integrating the objective function value and the total penalty value to construct an adaptability evaluation function, generating a Pareto optimal solution set by adopting an improved NSGA-II algorithm, preferentially screening out the individual which completely meets the three types of constraints, and selecting an optimal compromise solution as a final optimal maintenance scheme based on a compromise balance principle of the multi-objective function.
7. A port equipment maintenance decision optimization system with edge-cloud cooperation, applying the port equipment maintenance decision optimization method with edge-cloud cooperation as claimed in any one of claims 1-6, comprising: the edge terminal is used for acquiring multi-source heterogeneous data and transmitting the multi-source heterogeneous data to the edge gateway, and the edge gateway is used for preprocessing the multi-source heterogeneous data and extracting time domain, frequency domain, time-frequency domain and physical characteristics to generate a structural characteristic vector; The twin model construction unit is used for extracting candidate nodes based on the structural feature vector and combining an improved FCA algorithm and a cross-domain bridge ontology mapping rule to construct a semantic dynamic digital twin model; The analysis model construction unit takes the semantic dynamic digital twin model as a structural basis, adopts GCN as a basic framework, embeds physical constraints, constructs a lightweight analysis model, and is deployed at an edge gateway and outputs equipment health index, fault mode and data quality score; the judging unit continuously monitors the data quality through the lightweight analysis model, and if the quality scores of the data in 3 continuous monitoring periods are lower than a quality threshold, the corresponding structured feature vector, the output of the lightweight analysis model and the local submodel of the semantic dynamic digital twin model are uploaded to a cloud platform; and the scheme generating unit is used for receiving the data by the cloud platform, constructing a target optimization model by adopting PINN algorithm and combining a physical model library, and generating an optimal maintenance scheme.

Description

Port equipment maintenance decision optimization method and system with edge cloud cooperation Technical Field The invention relates to the technical field of port equipment maintenance, in particular to a port equipment maintenance decision optimization method and system with edge-cloud cooperation. Background The harbour site is used as a key hub facility of a global supply chain, and long-term faces to complex operation environments such as high-frequency start-stop, heavy-load impact, salt spray corrosion, working condition fluctuation and the like, the health state of the harbour site directly determines the harbour operation efficiency and the logistics link stability, and the traditional maintenance mode has the following pain points: The problems of isomerism, noise interference and data deletion of the multi-source sensor data exist, high-quality data preprocessing is difficult to achieve by a single-edge node, fault feature extraction accuracy is insufficient, static analysis models cannot be dynamically adapted to equipment aging, component abrasion and working condition change, port edge equipment calculation force is limited, a traditional deep learning model is difficult to deploy, real-time response of an edge side and cloud global optimization are disjoint, local optimization and global suboptimal maintenance decision are easy to occur, only a single target of fault repair is concerned, and multi-dimensional requirements such as maintenance cost, equipment availability, operation continuity and the like are not met, so that maintenance resource configuration efficiency is low. Therefore, it is needed to construct a multi-objective global maintenance decision method for fusing multi-source data, so as to solve the problems in the prior art. Disclosure of Invention In view of the above, the invention provides a method and a system for optimizing maintenance decisions of a port device with edge-cloud cooperation, which are used for solving the problems in the background technology. In order to achieve the above purpose, the present invention adopts the following technical scheme: A port equipment maintenance decision optimization method with edge cloud cooperation comprises the following steps: The edge terminal deploys a multi-type sensor, the edge terminal collects multi-source heterogeneous data and transmits the multi-source heterogeneous data to an edge gateway, the edge gateway carries out preprocessing on the multi-source heterogeneous data, extracts time domain, frequency domain, time-frequency domain and physical characteristics, and generates a structural feature vector; based on the structural feature vector, combining an improved FCA algorithm with a cross-domain bridge ontology mapping rule, extracting candidate nodes, and constructing a semantic dynamic digital twin model; taking the semantic dynamic digital twin model as a structural basis, adopting GCN as an infrastructure, embedding physical constraint, constructing a lightweight analysis model, deploying the lightweight analysis model on an edge gateway, and outputting equipment health index, fault mode and data quality score; Continuously monitoring the data quality through the lightweight analysis model, and uploading the corresponding structured feature vector, the output of the lightweight analysis model and the local submodel of the semantic dynamic digital twin model to a cloud platform if the data quality score of the continuous 3 monitoring periods is lower than a quality threshold; And the cloud platform receives the data, and builds a target optimization model by adopting PINN algorithm and combining a physical model library to generate an optimal maintenance scheme. Preferably, the candidate node extraction comprises discretizing historical sensor data and fault records, and constructing a formal background matrix S= (U, A, I), wherein U is a device core component set, A is an attribute set, is a union of fault modes, fault reasons and monitoring characteristics, and I is an association relation IIf the actual association exists between the component U and the attribute a, the value of (U, a) epsilon I is recorded as 1, otherwise, the value is 0, the association strength of the attribute is quantized by adopting cosine similarity, a similarity threshold is set, the attributes exceeding the similarity threshold are combined, the inclusion comprises equipment components, a fault mode, a fault cause and monitoring characteristics, four-element candidate nodes (C, F, R and M) are extracted and formed, and the monitoring characteristics strongly associated with the fault mode are screened through mutual information entropy to serve as attribute nodes. Preferably, the construction of the lightweight analytical model comprises the specific steps of model structure design and physical constraint embedding as follows: The method comprises the steps of designing a lightweight analysis model structure, taking four-tuple candid