Equipment life cycle cost management method and system

Through technologies such as multi-dimensional cost analysis model and reinforcement learning algorithm, the complexity and inefficiency of the cost management of equipment throughout the life cycle in the existing technology are solved, and the accurate prediction and optimization management of the cost of equipment throughout the life cycle is realized, which improves the improvement of the cost of equipment throughout the life cycle. Equipment maintenance efficiency and economic benefits.

CN118628097BActive Publication Date: 2024-11-15STATE GRID SHAANXI TENDERING CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411100653.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-11-15
Estimated Expiration
2044-08-12

AI Technical Summary

Technical Problem

The cost management technology of existing equipment throughout the life cycle depends on manual operations and repeated data entry. It is complex and inefficient, and lacks intelligent cost prediction and optimization methods, making it difficult to cope with complex and changing equipment operating environment and maintenance needs.

Method used

Using multi-dimensional cost analysis model, reinforcement learning algorithm, collaborative optimization mechanism, transfer learning technology and intelligent scheduling algorithm, data is collected in real time through sensor networks, multi-dimensional cost analysis model is built, cost allocation ratio dynamically optimized, and real-time cost prediction and optimization are carried out. Maintenance plan configuration.

Benefits of technology

It realizes accurate prediction and optimization management of the entire life cycle cost of the equipment, improves the accuracy and comprehensiveness of cost analysis, dynamically adjusts maintenance strategies, and improves equipment maintenance efficiency and economic benefits.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118628097B_ABST
    Figure CN118628097B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of equipment management technology, and in particular to a method and system for equipment life cycle cost management. The method collects equipment operation data through a sensor network, performs data cleaning and standardization processing, uses data mining technology and machine learning algorithms to build a multi-dimensional cost analysis model, and dynamically optimizes equipment costs through reinforcement learning algorithms and collaborative optimization mechanisms. Based on historical data and real-time data, a cost prediction model is built and optimized, and transfer learning and intelligent scheduling algorithms are used to perform real-time cost predictions to generate optimal maintenance plans and resource allocation plans. Dynamic adjustments are made in conjunction with a feedback optimization mechanism, and the optimization results are finally stored. The present invention uses advanced data analysis and intelligent algorithms to achieve accurate prediction and optimal scheduling of equipment operation and maintenance costs, significantly improves the efficiency and economic benefits of equipment management, and provides strong support for enterprises.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of equipment management, and in particular to a method and system for managing equipment's full life cycle cost. Background Art

[0002] Equipment life cycle cost management can comprehensively track all costs of equipment from purchase to scrapping throughout its life cycle, providing enterprises with comprehensive cost analysis and optimization basis. This management method can not only improve the utilization rate of equipment and extend the service life of equipment, but also reduce maintenance and operating costs through refined management.

[0003] However, the existing technology (Chinese invention patent, publication number: CN111882162B, name: A method and device for equipment full life cycle cost management) has significant deficiencies in implementing full life cycle cost management. For example, the existing technology adopts a method of synchronizing the equipment hierarchy structure in the production management system to the asset management system, obtaining the maintenance work order from the production management system, and importing it into the asset management system. This method realizes the full life cycle cost management of the equipment by detailed allocation and calculation of the labor cost, material cost, and machine shift fee of the maintenance work order. Although this method can achieve equipment cost management to a certain extent, it is highly dependent on manual operation and repeated entry of data, resulting in complex operation and low efficiency. In addition, this method lacks intelligent means in cost forecasting and optimization, and it is difficult to dynamically adjust and optimize the cost management strategy, and cannot fully cope with the complex and changeable equipment operating environment and maintenance needs. Summary of the invention

[0004] In response to the many problems existing in the above-mentioned prior arts, the present invention provides a method and system for equipment full life cycle cost management. The present invention realizes accurate prediction and optimal management of equipment full life cycle cost through the comprehensive application of multi-dimensional cost analysis model, reinforcement learning algorithm, collaborative optimization mechanism, transfer learning technology and intelligent scheduling algorithm.

[0005] A method for managing equipment life cycle costs, comprising the following steps:

[0006] The sensor network collects equipment operation data, material usage data and labor time record data in real time, forms a comprehensive data set after data cleaning and standardization, and integrates, unifies the format, verifies and normalizes the comprehensive data set to form a standardized data set;

[0007] Classify the cost data in the normalized data set, identify direct costs and indirect costs, and use data mining technology and machine learning algorithms to build a multi-dimensional cost analysis model. Analyze the correlation between equipment by introducing graph neural networks, and use a dynamic multi-level analysis framework to improve the accuracy of the model and generate equipment cost analysis result data;

[0008] Using reinforcement learning algorithms and collaborative optimization mechanisms, the equipment cost analysis result data is dynamically optimized, the cost allocation ratio of each equipment in different maintenance tasks is calculated, and the allocation strategy is adjusted through fuzzy clustering algorithms and adaptive allocation mechanisms to generate the final cost allocation results;

[0009] Based on historical data and real-time data, we build and optimize cost prediction models, use transfer learning and intelligent scheduling algorithms to make real-time cost predictions, generate optimal maintenance plans and resource allocation plans, combine feedback optimization mechanisms for dynamic adjustments, and finally store the optimization results.

[0010] Preferably, the multi-dimensional cost analysis model includes:

[0011] Cluster analysis method is used to classify cost data, and K-means algorithm is used to cluster cost data to determine different types of cost categories;

[0012] Use neural network algorithms to build cost prediction models, and train multi-layer feedforward neural networks based on clustering results and historical cost data to predict future costs;

[0013] Graph neural network is introduced to analyze the correlation between devices, the adjacency matrix is ​​used to represent the connection relationship between devices, and the multi-layer perceptron is used to quantitatively analyze the impact between devices.

[0014] Preferably, the K-means algorithm specifically includes iteratively calculating cluster centers, minimizing intra-cluster variance, allocating cost data points to the closest cluster center until convergence, and determining different types of cost categories.

[0015] Preferably, the graph neural network represents the correlation between devices through an adjacency matrix, and quantitatively analyzes the impact between devices through a multi-layer perceptron.

[0016] Preferably, the reinforcement learning algorithm includes a Q-learning algorithm, which is trained by historical data and real-time data, constructs a state-action value function, and adjusts the optimization strategy in real time to minimize the equipment life cycle cost, wherein the update formula of the Q-learning algorithm is:

[0017]

[0018] in, Indicates the current state; Indicates the current action; Indicates immediate reward; represents the learning rate; represents the discount factor; Indicates the next state; Indicates the next action; Indicates the current status Next action The Q value of the current strategy is used to evaluate the quality of the current strategy. is in the next state All possible actions In the example, the value of the action with the largest Q value is selected, which represents the optimal cumulative reward in the future; represents the optimal future cumulative reward after discount, weighing long-term and short-term benefits; Indicates the next state Next action The Q value is used to evaluate and update the current strategy.

[0019] Preferably, the collaborative optimization mechanism includes combining a genetic algorithm and a reinforcement learning algorithm, optimizing the initial population using a genetic algorithm, generating a new generation of populations through selection, crossover and mutation operations, and optimizing the new generation of populations using a reinforcement learning algorithm to achieve multi-objective optimization and balance cost and resource utilization.

[0020] Preferably, the cost prediction model uses a transfer learning method to extract features from data of different devices and scenarios, and migrates these features to the cost prediction model in the new environment for cost prediction in different environments.

[0021] Preferably, the specific formula of the transfer learning method is:

[0022]

[0023] in, Represents the loss function of the transfer learning method; is the number of samples of source domain data; Represents the loss function of the source domain data; and Represent source domain samples and labels respectively; is the number of samples of target domain data; A distance metric representing the characteristics of the target domain and the source domain; and Feature extractors representing source domain and target domain respectively; is the regularization parameter; It is a model trained on source domain data and used to make predictions on source domain data; It is a feature extractor that extracts features from source domain data; It is a feature extractor that extracts features from the target domain data for cost prediction in the target domain.

[0024] Preferably, the intelligent scheduling algorithm includes dynamically generating the optimal maintenance plan and resource allocation plan based on real-time cost prediction data and feedback data during equipment operation and maintenance, and optimizing through genetic algorithms and reinforcement learning algorithms.

[0025] A system for implementing the equipment life cycle cost management method, comprising:

[0026] Sensor network, used to collect real-time data on equipment operation, material usage, and labor time records;

[0027] The data processing module is used to clean and standardize the collected data to form a comprehensive data set, and to integrate, unify the format, verify and normalize the comprehensive data set to form a standardized data set;

[0028] The multi-dimensional cost analysis module is used to classify the cost data in the normalized data set, identify direct costs and indirect costs, and build a multi-dimensional cost analysis model using data mining technology and machine learning algorithms. It analyzes the correlation between devices by introducing graph neural networks, and uses a dynamic multi-level analysis framework to improve the accuracy of the model and generate equipment cost analysis result data;

[0029] The optimization module is used to dynamically optimize the equipment cost analysis result data using the reinforcement learning algorithm and the collaborative optimization mechanism, calculate the cost allocation ratio of each equipment in different maintenance tasks, adjust the allocation strategy through the fuzzy clustering algorithm and the adaptive allocation mechanism, and generate the final cost allocation result;

[0030] The prediction and scheduling module is used to build and optimize the cost prediction model based on historical data and real-time data, use transfer learning and intelligent scheduling algorithms to perform real-time cost prediction, generate optimal maintenance plans and resource allocation plans, combine feedback optimization mechanisms for dynamic adjustments, and finally store the optimization results.

[0031] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0032] The present invention realizes accurate prediction and optimized management of the cost of the entire life cycle of equipment through a multi-dimensional cost analysis model, a reinforcement learning algorithm and an intelligent scheduling algorithm, solving the problems of complex operation and low efficiency in the prior art.

[0033] The present invention uses a multi-dimensional cost analysis model and graph neural network technology to achieve refined classification and correlation analysis of equipment costs, thereby improving the accuracy and comprehensiveness of cost analysis.

[0034] The present invention uses a reinforcement learning algorithm and a collaborative optimization mechanism to dynamically optimize the equipment maintenance strategy, balance costs and resource utilization, and achieve efficient equipment maintenance and resource allocation;

[0035] The present invention uses transfer learning technology to extract features from data of different devices and scenarios, thereby improving the adaptability and accuracy of the cost prediction model and ensuring that the cost can be accurately predicted in different environments;

[0036] The present invention uses an intelligent scheduling algorithm to dynamically generate optimal maintenance plans and resource allocation plans based on real-time cost prediction data and feedback data during equipment operation and maintenance, thereby optimizing the operating efficiency and economic benefits of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the process of the present invention;

[0038] Figure 2 Construct a flow chart for the multi-dimensional cost analysis model of the present invention;

[0039] Figure 3 It is a flow chart for optimizing the reinforcement learning algorithm in the present invention;

[0040] Figure 4 A flow chart of using the transfer learning method to build and optimize the model in the present invention;

[0041] Figure 5 This is a flowchart for optimizing the intelligent scheduling algorithm in the present invention;

[0042] Figure 6 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0043] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0044] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0045] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0046] like Figure 1 As shown, a method for managing equipment life cycle cost includes the following steps:

[0047] The sensor network collects equipment operation data, material usage data and labor time record data in real time, forms a comprehensive data set after data cleaning and standardization, and integrates, unifies the format, verifies and normalizes the comprehensive data set to form a standardized data set;

[0048] Through the sensor network, the real-time monitoring of equipment operation status, material usage and labor hours can be achieved to obtain comprehensive and accurate data. These data include equipment operation parameters such as vibration, temperature, pressure, material inventory, consumption and usage, as well as labor hours and work efficiency information. The application of sensor networks can ensure the timeliness and accuracy of data, thereby providing reliable data support for subsequent analysis and decision-making.

[0049] In practical applications, first of all, the raw data obtained through the sensor network needs to be cleaned and standardized. Data cleaning includes removing outliers, duplicate data, and missing data to ensure the integrity and accuracy of the data. For example, statistical methods are used to identify and delete abnormal data points that deviate from the normal range, and duplicate records are deleted by comparing timestamps and content, and missing values ​​are filled using interpolation or mean filling methods. Data standardization is to convert all data into a unified scale, usually by scaling the data values ​​to the [0, 1] interval through normalization methods to ensure consistency of data from different sources. This step can effectively avoid analysis bias caused by different data dimensions and ensure the effectiveness of subsequent data integration.

[0050] Next, the cleaned and standardized data is formed into a comprehensive data set, and data integration, format unification, verification and normalization are performed on it. Data integration is to integrate data from different sensors into a unified framework for unified management and analysis. Format unification refers to the consistency of data formats from different data sources to ensure that the data is readable and available on the same platform. Data verification is to check the accuracy and consistency of data through a series of verification rules to ensure that the data is correct. Normalization is to further standardize and adjust the data to meet the preset standards and specifications. Through these steps, the final normalized data set is a structured, error-free and unified data set, which provides a solid foundation for multi-dimensional cost analysis and optimization.

[0051] Through the above process, comprehensive monitoring and analysis of equipment operation status, material usage and labor hours can be achieved. This not only helps to improve the accuracy and consistency of data, but also provides high-quality data support for the subsequent multi-dimensional cost analysis model construction, intelligent optimization algorithm application, and cost forecasting and dynamic adjustment. For example, in the full life cycle management of equipment, through real-time monitoring and data processing, abnormal conditions in equipment operation can be discovered in time, equipment maintenance needs can be predicted, material usage and inventory management can be optimized, and the utilization efficiency of human resources can be improved.

[0052] like Figure 2 As shown, the cost data in the normalized data set is classified, direct costs and indirect costs are identified, and a multi-dimensional cost analysis model is constructed using data mining technology and machine learning algorithms. The correlation between equipment is analyzed by introducing a graph neural network, and a dynamic multi-level analysis framework is used to improve the accuracy of the model, generating equipment cost analysis result data;

[0053] Through detailed classification and in-depth analysis of cost data, the main factors affecting the cost of equipment throughout its life cycle can be identified, thereby achieving accurate management and optimization of equipment costs.

[0054] Classify the cost data in the normalized data set to identify direct costs and indirect costs. Direct costs include material costs and labor costs directly consumed in equipment operation, while indirect costs include management costs, equipment depreciation, and machine hours. The purpose of classification is to distinguish costs of different natures for subsequent analysis and management.

[0055] A multi-dimensional cost analysis model is constructed using data mining technology and machine learning algorithms. Data mining technology is used to extract valuable information and patterns from a large amount of cost data. For example, cost data can be classified through cluster analysis methods to identify different types of cost categories. Machine learning algorithms are used to build prediction models. For example, a multi-layer feedforward neural network is trained based on clustering results and historical cost data to predict future costs through a neural network algorithm. Such a multi-dimensional cost analysis model can comprehensively consider various factors that affect equipment costs and provide accurate cost estimates through complex data analysis and prediction algorithms.

[0056] In order to further improve the accuracy of the model, the correlation between devices is analyzed by introducing a graph neural network. The graph neural network uses an adjacency matrix to represent the connection relationship between devices, and uses a multi-layer perceptron to quantitatively analyze the impact between devices. For example, in a device network, the failure of a certain device may affect the operating status and maintenance requirements of other related devices. Through the graph neural network, these complex correlations can be captured, improving the adaptability and accuracy of the cost analysis model to the actual device network.

[0057] In addition, the dynamic multi-level analysis framework is used to further improve the accuracy of the model. The dynamic multi-level analysis framework not only analyzes the cost of a single device, but also considers the mutual influence between equipment groups to form a hierarchical and multi-dimensional cost analysis model. For example, in the management of large-scale industrial equipment, the maintenance decision of a production line not only affects the cost of the production line itself, but also affects the operating efficiency and cost of the entire production system. Therefore, the dynamic multi-level analysis framework can comprehensively consider cost factors at different levels and dimensions, and provide more comprehensive and accurate cost analysis results.

[0058] The equipment cost analysis result data generated by the above method provides strong data support for the equipment life cycle cost management. These data can not only help managers understand the current cost structure of the equipment, but also predict future cost trends and optimize equipment operation and maintenance decisions. For example, through the detailed analysis of cost data, certain high-cost equipment or processes can be identified, so that targeted measures can be taken to control costs; through the analysis of the correlation between equipment, equipment maintenance plans can be optimized, chain reactions caused by equipment failures can be reduced, and the overall equipment operation efficiency and economic benefits can be improved.

[0059] Preferably, the multi-dimensional cost analysis model includes:

[0060] Cluster analysis method is used to classify cost data, and K-means algorithm is used to cluster cost data to determine different types of cost categories;

[0061] Use neural network algorithms to build cost prediction models, and train multi-layer feedforward neural networks based on clustering results and historical cost data to predict future costs;

[0062] Graph neural network is introduced to analyze the correlation between devices, the adjacency matrix is ​​used to represent the connection relationship between devices, and the multi-layer perceptron is used to quantitatively analyze the impact between devices.

[0063] Cluster analysis is used to classify cost data. Cluster analysis is a technique for grouping data so that data in the same group have high similarity, while data in different groups have large differences. In the present invention, the K-means algorithm is used to cluster cost data. The K-means algorithm iteratively calculates cluster centers (i.e., the average value of each cluster), minimizes intra-cluster variance, and assigns cost data points to the closest cluster center until convergence, thereby determining different types of cost categories. This method can effectively divide cost data into different categories, such as material costs, labor costs, equipment depreciation costs, etc., thereby helping to identify and understand different types of cost structures.

[0064] Use neural network algorithms to build cost prediction models. Based on the cost categories and historical cost data obtained from cluster analysis, a multi-layer feedforward neural network (MLP) is trained to predict future costs. The multi-layer feedforward neural network learns and fits the input historical cost data through multi-layer nonlinear transformation and back propagation algorithms, and can capture complex patterns and trends in the data, thereby predicting future cost changes. In specific implementation, the results of each cluster can be used as input features of the neural network, and then combined with historical cost data for training, so as to improve the accuracy of cost prediction. For example, the trained neural network can predict the future maintenance cost of a certain equipment, providing a reference for the formulation of maintenance plans.

[0065] Graph neural network (GNN) is introduced to analyze the correlation between devices. The correlation between devices has an important impact on cost management. For example, the failure of a certain device may cause the downtime of other devices or increase the maintenance requirements. Graph neural network uses graph structure data for deep learning, represents the connection relationship between devices through the adjacency matrix, and quantifies the impact between devices through the multi-layer perceptron (MLP). In practical applications, a graph representing the device network can be constructed, where each node represents a device and the edge represents the connection or impact between devices. In the device network graph, the edges of the graph are used to represent the relationship or interaction between different devices. This relationship or interaction is of great significance in the cost management of the entire life cycle of equipment, because the mutual impact between equipment will directly or indirectly affect the operating cost and maintenance requirements of the equipment. Specifically, it includes:

[0066] Connection relationship: This can refer to the physical connection between devices. For example, on a production line, the output of device A may be the input of device B, so there is a direct physical connection between them. This connection relationship means that the status of device A (such as operating speed or failure) may directly affect the status and performance of device B.

[0067] Impact relationship: This is not limited to physical connections, but also includes functional or operational interdependencies between devices. For example, the failure of a certain device may cause the entire production line to stop, which in turn affects the normal operation and maintenance arrangements of other devices. This impact relationship is reflected through the operational logic and dependency relationships between devices.

[0068] Through graph neural networks, the status of each device can be updated and propagated, thereby quantifying the impact of each device on the overall cost. For example, if the failure of a key device will significantly affect the operation of multiple related devices, the graph neural network can capture this complex correlation and reflect it in the cost analysis.

[0069] Through the above steps, the constructed multi-dimensional cost analysis model can comprehensively analyze and predict the cost of the entire life cycle of the equipment. The cluster analysis method can clearly classify and understand different types of costs; the use of neural network algorithms can accurately predict future cost trends; the introduction of graph neural networks can capture the complex correlation between equipment and improve the comprehensiveness and accuracy of cost analysis. The resulting equipment cost analysis result data provides managers with important decision-making basis, which can help optimize equipment maintenance plans, improve resource utilization efficiency, reduce operating costs, and achieve efficient management of the entire life cycle of equipment.

[0070] Preferably, the K-means algorithm specifically includes iteratively calculating cluster centers, minimizing intra-cluster variance, allocating cost data points to the closest cluster center until convergence, and determining different types of cost categories.

[0071] The application of K-means algorithm in the present invention aims to achieve accurate classification and identification of different types of costs through cluster analysis of cost data, thereby providing a basis for subsequent cost prediction and optimization.

[0072] First, the principle of the K-means algorithm is to divide the data points into K clusters so that the data points in each cluster are as similar as possible, while the data points between different clusters are quite different. In the present invention, the cost data may include equipment maintenance costs, operating costs, material costs, etc. These cost data points can be assigned to different clusters through the K-means algorithm to identify different types of cost categories. Specifically, the algorithm first randomly selects K initial cluster centers, and then iterates the following steps:

[0073] Assign data points: Calculate the distance from each cost data point to each cluster center and assign each data point to the nearest cluster center. Distance is usually calculated using Euclidean distance.

[0074] Update cluster center: Recalculate the center of each cluster, that is, calculate the average value of all data points in the cluster as the new cluster center.

[0075] Iterative calculation: Repeat the above two steps until the cluster center no longer changes or the change is less than the preset threshold, at which point the algorithm converges.

[0076] In the present invention, the K-means algorithm can effectively cluster complex cost data so that the cost data in the same cluster have similar characteristics. In this way, each type of cost can be analyzed and managed more carefully. For example, through the clustering results, it can be found that the maintenance cost of a certain type of equipment is generally high, so a more accurate maintenance plan can be formulated for this type of equipment to optimize resource allocation.

[0077] In one embodiment, for a large manufacturing enterprise, the K-means algorithm was used to perform cluster analysis on the equipment maintenance cost data of the past year. The results showed that the equipment maintenance costs can be divided into three categories: high-cost equipment, medium-cost equipment, and low-cost equipment. By further analyzing the maintenance records of high-cost equipment, it was found that these equipment had high maintenance costs due to frequent failures of key components. Therefore, the enterprise can adopt a preventive maintenance strategy for these high-cost equipment, replace vulnerable parts in advance, reduce the frequency of failures, and thus reduce maintenance costs.

[0078] In this way, the K-means algorithm not only helps companies identify and classify different types of costs, but also provides specific optimization directions and measures to improve the accuracy and efficiency of equipment life cycle management. Ultimately, by optimizing the cost structure, companies can achieve the goals of reducing operating costs, improving equipment utilization, and extending equipment life.

[0079] Preferably, the graph neural network represents the correlation between devices through an adjacency matrix, and quantitatively analyzes the impact between devices through a multi-layer perceptron.

[0080] The application of graph neural network (GNN) in this invention aims to optimize the cost management strategy by analyzing the complex relationships between devices and quantifying the impact of each device on the overall cost.

[0081] The principle of graph neural network is to represent the relationship between data through graph structure, and use neural network to extract features and model the relationship between nodes (devices) in the graph. In the present invention, the correlation between devices can be represented by an adjacency matrix. The adjacency matrix is ​​a square matrix in which each element represents the connection relationship between two devices. If two devices are directly related, the corresponding position in the matrix is ​​1, otherwise it is 0. In this way, the topological structure of the device network can be represented as a graph.

[0082] Specifically, the graph neural network uses the adjacency matrix and the initial feature vector of the device to update the features of each device. Through the multi-layer perceptron (MLP), the graph neural network can extract the high-level features of the device layer by layer and quantify the impact between devices. The calculation process of each layer includes neighbor aggregation and feature update. The specific steps are as follows:

[0083] Neighbor aggregation: For each device node, the feature vectors of its neighbor nodes are collected and weighted averaged or summed to form aggregate features.

[0084] Feature update: The aggregated features are input into the multi-layer perceptron (MLP), and after nonlinear transformation, the feature vector of the device node is updated.

[0085] Through the above iterative calculations, graph neural networks can effectively capture the complex relationships between devices and evaluate the impact of each device on the overall cost through quantitative analysis.

[0086] An example of application in the present invention is as follows: A manufacturing company has multiple interconnected production equipment, and there are complex relationships between these equipment, such as sharing production resources and serving as spare parts for each other. By building an adjacency matrix between equipment, the company can use graph neural networks to model the correlation between equipment. By extracting and updating equipment features layer by layer through a multi-layer perceptron (MLP), the company can quantify the impact of each equipment on the overall production cost.

[0087] For example, through the analysis results, it was found that the failure of certain key equipment would have a significant impact on the operation of the entire production line, resulting in production interruptions and increased costs. Enterprises can formulate more stringent maintenance plans for these key equipment to detect and prevent failures in advance, thereby reducing downtime and costs. At the same time, through in-depth analysis of the relationship between equipment, enterprises can also optimize the allocation of production resources and improve production efficiency.

[0088] In this way, graph neural networks not only help companies identify and quantify complex relationships between equipment, but also provide specific measures to optimize equipment management and cost control, thereby improving the accuracy and effectiveness of equipment life cycle management. Ultimately, companies can achieve the goals of reducing operating costs, improving production efficiency, and extending equipment life. This not only has a direct effect on improving the economic benefits of the company, but also helps the company gain greater advantages in market competition.

[0089] like Figure 3 As shown, the reinforcement learning algorithm and collaborative optimization mechanism are used to dynamically optimize the equipment cost analysis result data, calculate the cost allocation ratio of each equipment in different maintenance tasks, adjust the allocation strategy through the fuzzy clustering algorithm and adaptive allocation mechanism, and generate the final cost allocation result;

[0090] The core of this process is to achieve refined management of equipment costs through advanced algorithms, thereby optimizing resource allocation, reducing maintenance costs, and improving equipment operating efficiency and economic benefits.

[0091] Reinforcement learning algorithm is a machine learning method that continuously adjusts strategies to maximize long-term benefits by interacting with the environment. In the present invention, the reinforcement learning algorithm is used to dynamically optimize the equipment cost analysis result data. The principle is to define the state, action and reward function, construct the state-action value function (Q function), and continuously learn and adjust the optimal maintenance strategy of the equipment in different states. For example, in a certain state of the equipment, choosing to perform preventive maintenance may have higher long-term benefits than choosing to wait for a fault to occur before repairing. Through the Q-learning algorithm, training can be performed based on historical data and real-time data, and the optimization strategy can be adjusted in real time to ensure that the cost of the equipment is minimized throughout its life cycle.

[0092] The collaborative optimization mechanism combines multiple optimization algorithms to achieve multi-objective optimization. In the present invention, the collaborative optimization mechanism combines genetic algorithms and reinforcement learning algorithms. Genetic algorithms optimize the initial population and generate a new generation of populations by simulating the natural selection process using selection, crossover and mutation operations. Reinforcement learning algorithms are used to further optimize individual strategies in the new generation of populations to achieve a balance between cost and resource utilization. For example, through collaborative optimization, the optimal cost-sharing strategy for different equipment in different maintenance tasks can be determined to ensure maximum cost-effectiveness under limited resource conditions.

[0093] During the specific implementation, the reinforcement learning algorithm is first used to preliminarily optimize the equipment cost analysis result data, and the cost sharing ratio of each device in different maintenance tasks is calculated. Then, the preliminary optimization results are refined and adjusted through the fuzzy clustering algorithm. The fuzzy clustering algorithm is an algorithm that assigns data points to multiple clusters. Each data point can belong to multiple clusters, but has different memberships. In the present invention, the fuzzy clustering algorithm is used to refine the cost sharing ratio of different maintenance tasks to ensure the flexibility and accuracy of the sharing strategy. For example, for some equipment, its maintenance cost may be affected by multiple factors. Through the fuzzy clustering algorithm, its cost can be allocated more accurately to avoid waste of resources.

[0094] The adaptive allocation mechanism dynamically adjusts the allocation strategy based on real-time data and historical data to generate the final cost allocation result. The adaptive allocation mechanism can respond to equipment status and environmental changes in real time to ensure the rationality and dynamic optimization of the allocation strategy. For example, by monitoring the real-time operating status and maintenance records of the equipment, the adaptive allocation mechanism can adjust the cost allocation ratio of different equipment in different maintenance tasks to ensure that the overall maintenance cost is minimized.

[0095] In one embodiment, in a manufacturing enterprise, the maintenance cost of the equipment on the production line is optimized through reinforcement learning algorithm and collaborative optimization mechanism. First, through the Q-learning algorithm, according to the historical fault records and maintenance data of the equipment, the state-action value function is established to preliminarily optimize the maintenance strategy of each device. Then, through the fuzzy clustering algorithm, the preliminary optimization results are refined and adjusted to ensure that the cost sharing ratio of different equipment in different maintenance tasks is more accurate. Finally, through the adaptive sharing mechanism, the sharing strategy is dynamically adjusted according to the real-time data to generate the final cost sharing result. In this way, the enterprise can not only reduce the overall maintenance cost, but also improve the operating efficiency of the equipment and the stability of the production line.

[0096] Preferably, the reinforcement learning algorithm includes a Q-learning algorithm, which is trained by historical data and real-time data, constructs a state-action value function, and adjusts the optimization strategy in real time to minimize the equipment life cycle cost, wherein the update formula of the Q-learning algorithm is:

[0097]

[0098] in, Indicates the current state; Indicates the current action; Indicates immediate reward; represents the learning rate; represents the discount factor; Indicates the next state; Indicates the next action; Indicates the current status Next action The Q value of the current strategy is used to evaluate the quality of the current strategy. is in the next state All possible actions In the example, the value of the action with the largest Q value is selected, which represents the optimal cumulative reward in the future; represents the optimal future cumulative reward after discount, weighing long-term and short-term benefits; Indicates the next state Next action The Q value is used to evaluate and update the current strategy.

[0099] Q-learning is a model-free reinforcement learning algorithm that constructs a state-action value function (Q function) to evaluate the long-term value of each state-action pair to determine the optimal strategy. Q-learning repeatedly updates the Q value to gradually approach the true state-action value function. Ultimately, in any state, the action with the largest Q value is selected as the optimal strategy.

[0100] In practical applications, suppose a company wants to optimize the maintenance strategy of its equipment to minimize the total cost of the equipment throughout its life cycle. The specific implementation steps are as follows:

[0101] Define various operating states of the equipment, such as normal operation, minor fault, major fault, etc.;

[0102] Define the actions that can be taken in each state, such as immediate maintenance, delayed maintenance, continued operation, etc.

[0103] Design reward functions based on the immediate cost or benefit of each action; for example, timely maintenance of equipment can reduce the probability of future failures, thus obtaining positive rewards; delayed maintenance may lead to more serious failures, thus obtaining negative rewards;

[0104] Initialize the Q-values ​​of all state-action pairs to zero or random values;

[0105] Through simulation or actual operation, record the state transition of the device and the corresponding instant reward, and continuously update the Q value according to the Q-learning update formula;

[0106] In any state, the action with the largest Q value is selected as the optimal strategy, and the maintenance plan of the equipment is adjusted in real time.

[0107] Through the Q-learning algorithm, enterprises can dynamically optimize equipment maintenance strategies and adjust equipment operation and maintenance plans in real time to minimize the total cost of the entire life cycle. The specific effects include: by optimizing maintenance strategies, equipment failures can be discovered and handled in a timely manner, reducing the high maintenance costs caused by failures; through reasonable maintenance plans, unplanned downtime of equipment can be reduced, and equipment utilization and production efficiency can be improved; through timely maintenance and care, equipment wear and failure can be reduced, and the service life of equipment can be extended.

[0108] For example, a manufacturing company applied the Q-learning algorithm to optimize the maintenance strategy of its key production equipment. After several months of data training and strategy adjustment, the company significantly reduced the failure rate and maintenance costs of the equipment and improved the overall operating efficiency and economic benefits of the production line.

[0109] Preferably, the collaborative optimization mechanism includes combining a genetic algorithm and a reinforcement learning algorithm, optimizing the initial population using a genetic algorithm, generating a new generation of populations through selection, crossover and mutation operations, and optimizing the new generation of populations using a reinforcement learning algorithm to achieve multi-objective optimization and balance cost and resource utilization.

[0110] The application of this collaborative optimization mechanism in the present invention aims to achieve global optimization of equipment maintenance and operation strategies through advanced optimization technology, thereby improving the efficiency and effectiveness of equipment management.

[0111] Genetic algorithm is an optimization algorithm that simulates natural selection and genetic mechanism. It generates a new generation of population through selection, crossover and mutation operations, so as to gradually approach the optimal solution. In the present invention, the genetic algorithm is used to optimize the initial population, that is, the set of initial equipment maintenance strategies. The specific steps include: generating a set of initial equipment maintenance strategies, each strategy represents a possible maintenance plan; evaluating the fitness of each maintenance strategy, which is usually based on the cost-effectiveness of the strategy, that is, the balance between maintenance cost and equipment operation efficiency; selecting maintenance strategies with higher fitness for reproduction according to fitness, and eliminating strategies with lower fitness; performing crossover operations on the selected maintenance strategies, that is, exchanging some strategy genes to generate new maintenance strategies; performing mutation operations on the new strategies after crossover, that is, randomly changing some genes to increase the diversity of strategies. These steps are repeated until the predetermined optimization generation is reached or the fitness is stable.

[0112] After the new generation of population is generated, the reinforcement learning algorithm is used to further optimize the new generation of population. The reinforcement learning algorithm constructs a state-action value function (Q function) through interaction with the environment, and continuously adjusts and optimizes the strategy to maximize long-term benefits. In the present invention, the reinforcement learning algorithm is used to optimize the new strategy generated by the genetic algorithm to achieve multi-objective optimization. The specific process is: under the operating state of each device, the instant rewards of different maintenance actions are calculated based on historical data and real-time data, and the state-action value function is updated through the Q-learning algorithm, and the maintenance strategy is adjusted in real time to ensure that the optimal maintenance action is selected under different states, thereby minimizing the cost of the entire life cycle of the equipment.

[0113] By combining the collaborative optimization mechanism of genetic algorithms and reinforcement learning algorithms, multi-objective optimization can be achieved to balance cost and resource utilization. In the full life cycle management of equipment, this mechanism can effectively optimize maintenance strategies, reduce maintenance costs, and improve equipment utilization and operating efficiency. For example, in a manufacturing company, the maintenance strategy of key equipment is optimized through a collaborative optimization mechanism. First, the initial strategy set is generated using a genetic algorithm, and the optimization strategy is generated through multiple iterative selection, crossover and mutation operations. Then, these strategies are further optimized through a reinforcement learning algorithm, and the maintenance plan is dynamically adjusted based on the real-time operation data and historical maintenance records of the equipment. In the end, the company successfully reduced the overall maintenance cost of the equipment, improved the operating efficiency of the production line, and extended the service life of the equipment.

[0114] like Figure 4 As shown, based on historical data and real-time data, a cost prediction model is built and optimized, and transfer learning and intelligent scheduling algorithms are used to perform real-time cost prediction, generate optimal maintenance plans and resource allocation plans, and dynamically adjust them in combination with feedback optimization mechanisms, and finally store the optimization results.

[0115] By combining data analysis and intelligent algorithms, accurate prediction and optimized scheduling of equipment operation and maintenance costs can be achieved, thereby improving the efficiency and economic benefits of equipment management.

[0116] Building and optimizing the cost prediction model is the basis of the entire process. The cost prediction model is built based on the historical and real-time data of the equipment, using machine learning and data mining techniques to analyze and capture patterns and trends in the data. Specifically, by extracting features and modeling data such as historical maintenance records, equipment operating status, and material consumption, a model that can predict future maintenance costs is established. The model can predict the maintenance needs and corresponding costs of equipment under different operating conditions. For example, by training a multi-layer feedforward neural network model, the maintenance cost of a certain equipment in the next month can be predicted. Based on the current operating status of the equipment and historical maintenance records, the model can provide accurate cost prediction values.

[0117] Introducing transfer learning technology into the cost prediction model can further improve the adaptability and accuracy of the model. Transfer learning is a machine learning method that learns from one task and applies it to another related task. In the present invention, transfer learning can migrate data features from different devices and different operating environments to a new prediction model, thereby achieving accurate cost prediction for different devices and different operating environments. For example, the maintenance mode learned from the historical data of a certain type of equipment can be migrated to the prediction model of another type of equipment to improve its prediction performance.

[0118] The prediction results generated by the cost prediction model provide data support for the subsequent intelligent scheduling algorithm. The intelligent scheduling algorithm formulates the optimal maintenance plan and resource allocation plan by analyzing the prediction results. Specifically, the intelligent scheduling algorithm takes into account factors such as equipment maintenance requirements, available resources, and maintenance personnel's work schedule, and uses optimization algorithms (such as genetic algorithms, ant colony algorithms, etc.) to generate the optimal maintenance plan and resource allocation plan. For example, through the intelligent scheduling algorithm, it can be determined which equipment needs to be maintained and when, and which maintenance personnel and materials are assigned, so as to maximize resource utilization and minimize maintenance costs.

[0119] In actual applications, a manufacturing company has achieved real-time cost prediction and optimal scheduling of key equipment by building and optimizing cost prediction models, using transfer learning and intelligent scheduling algorithms. First, a multi-layer feedforward neural network model was built by analyzing the historical maintenance data and real-time operating status of the equipment, and the prediction performance of the model was improved through transfer learning technology. Then, using the intelligent scheduling algorithm, a maintenance plan and resource allocation plan for the next month were formulated based on the prediction results to ensure that maintenance costs and resource waste are minimized without affecting production.

[0120] In order to ensure the effectiveness and flexibility of the scheduling plan, dynamic adjustments are made in combination with the feedback optimization mechanism. The feedback optimization mechanism timely adjusts and optimizes the maintenance plan and resource allocation by real-time monitoring of the equipment's operating status and maintenance status. For example, if the operating status of a certain equipment changes or a sudden failure occurs, the feedback optimization mechanism can immediately adjust the original maintenance plan and reallocate resources to ensure the normal operation of the equipment and the timeliness of maintenance. Finally, the optimized results are stored in the central database to ensure the security and traceability of the data, providing data support for subsequent analysis and decision-making.

[0121] Preferably, the cost prediction model uses a transfer learning method to extract features from data of different devices and scenarios, and migrates these features to the cost prediction model in the new environment for cost prediction in different environments.

[0122] The application of transfer learning methods aims to enhance the generalization ability of cost prediction models by leveraging existing knowledge in different devices and scenarios, thereby improving their prediction performance in new environments.

[0123] Transfer learning is a machine learning method that improves the learning efficiency and effectiveness of new tasks by learning from one task and applying it to another related task. In the present invention, the transfer learning method is used to extract features from data of different devices and scenarios, and migrate these features to the cost prediction model in the new environment. Specifically, by analyzing and extracting features such as operating data, maintenance records, and material consumption of different devices, these features are incorporated into the new cost prediction model, enabling it to make accurate cost predictions in new devices and operating environments.

[0124] The principle of transfer learning is to transfer knowledge from the source domain (known devices and scenarios) to the target domain (new devices and scenarios) by sharing feature extractors. In the cost prediction model, the source domain data includes historical maintenance records, operating status, and cost data, while the target domain data is real-time data and operating records in new devices and scenarios. Through transfer learning, the existing knowledge in the source domain data can be effectively utilized, reducing the training requirements of the target domain data, thereby improving the training efficiency and prediction accuracy of the model.

[0125] In the specific implementation, we first extract features from the source domain data and build a feature extractor for the source domain. For example, for the maintenance records of a certain type of equipment, we extract key features such as failure mode, maintenance frequency, and cost trend through a convolutional neural network (CNN). Then, we migrate these features to the target domain, and through fine-tuning and retraining, we build a cost prediction model suitable for new equipment and scenarios. During the migration process, we can use techniques such as adversarial training and feature alignment to further improve the migration effect and the generalization ability of the model.

[0126] In this way, the cost prediction model can maintain efficient prediction performance under different equipment and scenarios. For example, in a manufacturing company, by extracting features from historical maintenance data of different types of equipment, a general cost prediction model is constructed using transfer learning technology. Then, the model is applied to newly introduced production line equipment, and the model parameters are fine-tuned through a small amount of real-time data to quickly establish high-precision cost prediction capabilities. In this way, the company can not only predict the maintenance needs of new equipment in advance, but also optimize resource allocation and reduce overall maintenance costs.

[0127] The application of transfer learning methods in cost forecasting not only improves the adaptability and accuracy of the model, but also significantly shortens the model training time and data requirements. For example, the maintenance mode learned from the historical data of a certain type of equipment can be directly applied to other similar equipment, thus avoiding the high cost and time consumption of training the model from scratch. This is of great significance for enterprises to maintain efficient equipment management and cost control in a rapidly changing production environment.

[0128] Preferably, the specific formula of the transfer learning method is:

[0129]

[0130] in, Represents the loss function of the transfer learning method; is the number of samples of source domain data; Represents the loss function of the source domain data; and Represent source domain samples and labels respectively; is the number of samples of target domain data; A distance metric representing the characteristics of the target domain and the source domain; and Feature extractors representing source domain and target domain respectively; is the regularization parameter; It is a model trained on source domain data and used to make predictions on source domain data; It is a feature extractor that extracts features from source domain data; It is a feature extractor that extracts features from the target domain data for cost prediction in the target domain.

[0131] The transfer learning method uses the above formula to realize the process of extracting features from source domain data and migrating them to the target domain model, thereby improving the adaptability and accuracy of cost prediction.

[0132] The core of transfer learning is to apply the knowledge in the source domain to the target domain by sharing feature extractors, thereby achieving efficient learning and prediction in a new environment.

[0133] The loss function in the above formula is It consists of two parts: the first part is the loss function of the source domain data, and the second part is the distance measurement between the target domain and the source domain features. Specifically, the first part Indicates that the model is trained on the source domain data so that the model can accurately fit the source domain samples and tags Part II It represents the distance metric between the target domain and the source domain features, by minimizing the number of target domain samples Feature extractors in source and target domains and The feature differences in , realize feature migration and sharing. Regularization parameter It is used to balance the weight between the source domain loss and the feature distance metric, thereby controlling the effect of transfer learning.

[0134] In the present invention, the application of this transfer learning method can significantly improve the adaptability and accuracy of the cost prediction model. For example, in a manufacturing enterprise, by extracting features from the historical maintenance data of different types of equipment, a feature extractor in the source domain can be constructed. When an enterprise introduces a new type of equipment, it collects a small amount of initial data on the new equipment. , and use the feature extractor of the target domain After processing these data, the distance between the source domain features and the target domain features is measured and minimized, thereby achieving feature migration of new equipment. In this way, even in a new environment, enterprises can quickly build a high-precision cost prediction model to accurately predict the maintenance needs and corresponding costs of new equipment.

[0135] In the specific implementation, we first extract features from the source domain data and build a feature extractor for the source domain. For example, for the maintenance records of a certain type of equipment, we extract key features such as failure mode, maintenance frequency, and cost trend through a convolutional neural network (CNN). Then, we migrate these features to the target domain, and through fine-tuning and retraining, we build a cost prediction model suitable for new equipment and scenarios. During the migration process, we can use techniques such as adversarial training and feature alignment to further improve the migration effect and the generalization ability of the model.

[0136] In this way, the transfer learning method can maintain efficient prediction performance under different equipment and scenarios. For example, in a manufacturing company, by extracting features from historical maintenance data of different types of equipment, a general cost prediction model is constructed using transfer learning technology. Then, the model is applied to newly introduced production line equipment, and the model parameters are fine-tuned through a small amount of real-time data to quickly establish high-precision cost prediction capabilities. In this way, the company can not only predict the maintenance needs of new equipment in advance, but also optimize resource allocation and reduce overall maintenance costs.

[0137] The application of transfer learning methods in cost forecasting not only improves the adaptability and accuracy of the model, but also significantly shortens the model training time and data requirements. For example, the maintenance mode learned from the historical data of a certain type of equipment can be directly applied to other similar equipment, thus avoiding the high cost and time consumption of training the model from scratch. This is of great significance for enterprises to maintain efficient equipment management and cost control in a rapidly changing production environment.

[0138] like Figure 5 As shown, preferably, the intelligent scheduling algorithm includes dynamically generating the optimal maintenance plan and resource allocation plan based on real-time cost prediction data and feedback data during equipment operation and maintenance, and optimizing through genetic algorithms and reinforcement learning algorithms.

[0139] In the present invention, the application of intelligent scheduling algorithm aims to achieve optimal management of equipment maintenance and resource allocation through real-time data analysis and intelligent optimization, thereby improving equipment operation efficiency and reducing maintenance costs.

[0140] The core of the intelligent scheduling algorithm is to use real-time cost prediction data and feedback data from equipment operation and maintenance to dynamically generate the optimal maintenance plan and resource allocation plan. First, the maintenance needs and corresponding cost data of the equipment in the future are obtained through the cost prediction model. These data provide prediction information on the possible failure time point of the equipment, maintenance frequency and required resources. Based on these prediction data, the intelligent scheduling algorithm can formulate maintenance plans in advance and reasonably allocate maintenance resources.

[0141] The intelligent scheduling algorithm is optimized using genetic algorithms and reinforcement learning algorithms. Genetic algorithms are an optimization algorithm that simulates natural selection and genetic mechanisms. They iteratively generate new generations of populations through operations such as selection, crossover, and mutation, thereby gradually approaching the optimal solution. In the present invention, genetic algorithms are used to generate initial maintenance plans and resource allocation plans. The specific steps include generating a set of initial plans, each of which represents a set of possible maintenance plans and resource allocations; evaluating the fitness of each plan, which is usually based on a balance between total maintenance costs and resource utilization; selecting plans with higher fitness for reproduction according to fitness, and eliminating plans with lower fitness. "Selecting plans with higher fitness for reproduction according to fitness" means selecting those plans with better performance from a set of candidate solutions as the basis for further combination and mutation to generate a new generation of solutions. This process simulates the mechanism of natural selection in the biological world, that is, the survival of the fittest, to gradually approach the optimal solution to the problem. Specifically:

[0142] Fitness is used to measure the pros and cons of each solution (i.e., individual) in solving a problem. In the present invention, fitness is usually based on the balance between the total maintenance cost and resource utilization of the maintenance plan and resource allocation plan. A solution with high fitness means that it can effectively utilize resources at a low cost and meet maintenance needs.

[0143] The selected schemes are cross-operated, that is, some scheme genes are exchanged to generate new schemes; the new schemes after cross-operation are mutated, that is, some genes are randomly changed to increase the diversity of schemes. Through multiple iterations, the genetic algorithm can generate a set of better maintenance plans and resource allocation schemes.

[0144] After the preliminary plan is generated, it is further optimized through the reinforcement learning algorithm. The reinforcement learning algorithm constructs a state-action value function (Q function) through interaction with the environment, and continuously adjusts and optimizes the plan to maximize long-term benefits. In the present invention, the reinforcement learning algorithm is used to adjust the maintenance plan and resource allocation plan in real time to ensure that the optimal maintenance action is selected under different conditions. The specific process is: under the operating state of each device, the instant reward of different maintenance actions is calculated according to the real-time feedback data, and the state-action value function is updated through the Q-learning algorithm to adjust the maintenance plan and resource allocation in real time. For example, if the real-time operating status of a certain device shows that its failure probability has increased, the reinforcement learning algorithm can immediately adjust the original maintenance plan, arrange maintenance in advance, and avoid failures.

[0145] In actual applications, a manufacturing company optimizes its equipment maintenance and resource allocation strategies through intelligent scheduling algorithms. First, through the cost prediction model, the company obtains the maintenance demand and cost forecast data of key equipment in the future. Then, the genetic algorithm is used to generate preliminary maintenance plans and resource allocation plans, and a set of better plans are generated through multiple iterations of selection, crossover and mutation operations. Then, through the reinforcement learning algorithm, these plans are further optimized according to the real-time operating status and feedback data of the equipment, and the maintenance plan and resource allocation are adjusted in real time. For example, when the operating data of a certain equipment shows that its vibration frequency increases abnormally, the system can arrange maintenance immediately to avoid production downtime and high maintenance costs caused by equipment failure.

[0146] In this way, the intelligent scheduling algorithm can significantly improve the efficiency and effectiveness of equipment maintenance. First, the maintenance plan generated based on real-time data and predicted data is more accurate, and maintenance can be scheduled in advance to reduce equipment failures and downtime. Secondly, through the optimization of genetic algorithms and reinforcement learning algorithms, maintenance plans and resource allocation plans are more reasonable, which can maximize resource utilization and reduce maintenance costs. Finally, the dynamic adjustment mechanism ensures the flexibility and real-time nature of the maintenance plan, can quickly respond to changes in equipment status, and provide timely maintenance support.

[0147] like Figure 6 As shown, a system for implementing the equipment life cycle cost management method includes:

[0148] The sensor network is used to collect equipment operation data, material usage data and labor time record data in real time; the sensor network is used to collect equipment operation data, material usage data and labor time record data in real time, and is the basic module of the equipment life cycle cost management system. This module continuously monitors the operating status of the equipment, such as temperature, vibration, pressure, etc., as well as the usage of materials and the working time records of workers through various sensors deployed in the equipment and factories. These data are collected in real time and transmitted to the central data processing system. The core principle of the sensor network is to use the Internet of Things technology to connect sensors distributed everywhere through wireless or wired networks to realize real-time data collection and transmission. The effect is to provide a comprehensive view of the operation of the equipment, which provides a basis for subsequent data processing and analysis.

[0149] The data processing module is responsible for cleaning and standardizing the collected data to form a comprehensive data set, and integrating, unifying the format, verifying and normalizing the comprehensive data set to form a normalized data set. The data processing module is responsible for cleaning and standardizing the collected data to form a comprehensive data set, and integrating, unifying the format, verifying and normalizing the comprehensive data set to form a normalized data set. The principle of data cleaning and standardization is to remove noise and outliers in the data through a series of algorithms and rules to ensure the consistency and accuracy of the data. For example, remove duplicate data, fill in missing data, etc. Data integration and format unification are to uniformly format data from different sensors and systems to ensure that the data can be effectively used in subsequent analysis. Through these steps, the data processing module can generate high-quality normalized data sets and provide a reliable data basis for multi-dimensional cost analysis.

[0150] The multi-dimensional cost analysis module is used to classify the cost data in the normalized data set, identify direct costs and indirect costs, and use data mining technology and machine learning algorithms to build a multi-dimensional cost analysis model. The module analyzes the correlation between devices by introducing graph neural networks, and uses a dynamic multi-level analysis framework to improve the accuracy of the model, and generates equipment cost analysis result data. The multi-dimensional cost analysis module is used to classify the cost data in the normalized data set, identify direct costs and indirect costs, and use data mining technology and machine learning algorithms to build a multi-dimensional cost analysis model. This module analyzes the correlation between devices by introducing graph neural networks, and uses a dynamic multi-level analysis framework to improve the accuracy of the model, and generates equipment cost analysis result data. In principle, data mining technology discovers patterns and rules by analyzing a large amount of historical data. For example, the cost data is clustered by the K-means algorithm to identify different types of cost categories; the neural network algorithm predicts future costs by learning from historical cost data and clustering results. The graph neural network further analyzes the correlation between devices, represents the connection relationship between devices through the adjacency matrix, and uses a multi-layer perceptron to quantitatively analyze the impact between devices. The effect is to provide a refined, multi-dimensional view of cost analysis, providing a scientific basis for equipment management decisions.

[0151] The optimization module is used to dynamically optimize the equipment cost analysis result data using the reinforcement learning algorithm and the collaborative optimization mechanism, calculate the cost sharing ratio of each device in different maintenance tasks, adjust the sharing strategy through the fuzzy clustering algorithm and the adaptive sharing mechanism, and generate the final cost sharing result; the optimization module uses the reinforcement learning algorithm and the collaborative optimization mechanism to dynamically optimize the equipment cost analysis result data. This module calculates the cost sharing ratio of each device in different maintenance tasks, and adjusts the sharing strategy through the fuzzy clustering algorithm and the adaptive sharing mechanism to generate the final cost sharing result. The reinforcement learning algorithm constructs the state-action value function through interaction with the environment, and continuously adjusts and optimizes the strategy to minimize the equipment life cycle cost. Specifically, the Q-learning algorithm uses historical data and real-time data for training, and adjusts the optimization strategy in real time to achieve the goal of minimizing costs. The collaborative optimization mechanism combines the genetic algorithm and the reinforcement learning algorithm, optimizes the initial population through the genetic algorithm, generates a new generation of population through selection, crossover and mutation operations, and then uses the reinforcement learning algorithm to optimize the new generation of population to balance cost and resource utilization. The effect is to provide a dynamically optimized cost sharing strategy, which improves the efficiency and effectiveness of cost management.

[0152] The prediction and scheduling module is used to build and optimize the cost prediction model based on historical data and real-time data, use transfer learning and intelligent scheduling algorithms to make real-time cost predictions, and generate the optimal maintenance plan and resource allocation plan, and dynamically adjust it in combination with the feedback optimization mechanism, and finally store the optimization results. The prediction and scheduling module is used to build and optimize the cost prediction model based on historical data and real-time data, use transfer learning and intelligent scheduling algorithms to make real-time cost predictions, and generate the optimal maintenance plan and resource allocation plan, and dynamically adjust it in combination with the feedback optimization mechanism, and finally store the optimization results. The transfer learning method extracts features from the data of different equipment and scenarios, and migrates these features to the cost prediction model in the new environment to improve the adaptability and accuracy of the model. The intelligent scheduling algorithm dynamically generates the optimal maintenance plan and resource allocation plan through real-time cost prediction data and feedback data during equipment operation and maintenance. Specifically, the genetic algorithm is used to generate preliminary maintenance plans and resource allocation plans, and the reinforcement learning algorithm adjusts and optimizes these plans in real time to ensure that the optimal maintenance action is selected under different conditions. The effect is to achieve real-time scheduling and optimization of equipment maintenance, ensuring the efficiency and economy of equipment operation.

[0153] The system of the present invention collects equipment operation data, material usage data and labor time record data in real time through the sensor network; the data processing module cleans, integrates and normalizes the data to generate a high-quality data set; the multi-dimensional cost analysis module constructs a multi-dimensional cost analysis model through data mining and machine learning algorithms to generate refined cost analysis results; the optimization module dynamically optimizes the cost sharing strategy through reinforcement learning and collaborative optimization mechanisms; the prediction and scheduling module generates the optimal maintenance plan and resource allocation plan in real time based on transfer learning and intelligent scheduling algorithms, and dynamically adjusts it in combination with the feedback mechanism to ensure the efficiency and economy of equipment operation. The present invention provides a comprehensive, accurate and efficient equipment life cycle cost management method, which effectively improves the overall level of equipment management.

[0154] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0155] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for managing equipment life cycle costs, characterized in that: The following steps are involved: The sensor network collects equipment operation data, material usage data and labor time record data in real time, forms a comprehensive data set after data cleaning and standardization, and integrates, unifies the format, verifies and normalizes the comprehensive data set to form a standardized data set; Classify the cost data in the normalized data set, identify direct costs and indirect costs, and use data mining technology and machine learning algorithms to build a multi-dimensional cost analysis model. Analyze the correlation between equipment by introducing graph neural networks, and use a dynamic multi-level analysis framework to improve the accuracy of the model and generate equipment cost analysis result data; Using reinforcement learning algorithms and collaborative optimization mechanisms, the equipment cost analysis result data is dynamically optimized, the cost allocation ratio of each equipment in different maintenance tasks is calculated, and the allocation strategy is adjusted through fuzzy clustering algorithms and adaptive allocation mechanisms to generate the final cost allocation results; Based on historical data and real-time data, we build and optimize cost prediction models, use transfer learning methods and intelligent scheduling algorithms to make real-time cost predictions, generate optimal maintenance plans and resource allocation plans, combine feedback optimization mechanisms to make dynamic adjustments, and finally store the optimization results; The use of the reinforcement learning algorithm and the collaborative optimization mechanism to dynamically optimize the equipment cost analysis result data includes: Reinforcement learning algorithms include Q-learning algorithms, which are trained through historical data and real-time data to build state-action value functions and adjust optimization strategies in real time to minimize the cost of the entire life cycle of the equipment. The update formula of the Q-learning algorithm is: ; in, Indicates the current state; Indicates the current action; Indicates immediate reward; represents the learning rate; represents the discount factor; Indicates the next state; Indicates the next action; Indicates the current status Next action The Q value of the current strategy is evaluated; is in the next state All possible actions In the example, the value of the action with the largest Q value is selected, which represents the optimal cumulative reward in the future; represents the optimal future cumulative reward after discount, weighing long-term and short-term benefits; Indicates the next state Next action The Q value is used to evaluate and update the current strategy. Q-learning repeatedly updates the Q value to gradually approach the true state-action value function. Finally, in any state, the action with the largest Q value is selected as the optimal strategy. The collaborative optimization mechanism combines genetic algorithms and reinforcement learning algorithms. The specific steps include: generating a set of initial equipment maintenance strategies, each strategy represents a possible maintenance plan; evaluating the fitness of each maintenance strategy, selecting maintenance strategies with higher fitness for reproduction according to the fitness, and eliminating strategies with lower fitness; performing crossover operations on the selected maintenance strategies to generate new maintenance strategies; performing mutation operations on the new strategies after crossover to increase the diversity of strategies; these steps are repeated until the predetermined optimization generation is reached or the fitness is stable; after generating a new generation of population, using reinforcement learning algorithms to further optimize the new generation of population; The use of transfer learning and intelligent scheduling algorithms to perform real-time cost prediction and generate optimal maintenance plans and resource allocation solutions includes: migrating knowledge in the source domain to the target domain through a shared feature extractor. In the cost prediction model, the source domain data includes historical maintenance records, operating status, and cost data, and the target domain data is real-time data and operating records in new equipment and scenarios. First, feature extraction is performed on the source domain data to build a feature extractor for the source domain; then, these features are migrated to the target domain, and a cost prediction model suitable for new equipment and scenarios is constructed through fine-tuning and retraining. Intelligent scheduling algorithms include dynamically generating optimal maintenance plans and resource allocation plans based on real-time cost forecast data and feedback data from equipment operation and maintenance, and optimizing them through genetic algorithms and reinforcement learning algorithms.

2. The method according to claim 1, characterized in that: The multi-dimensional cost analysis model includes: Cluster analysis method is used to classify cost data, and K-means algorithm is used to cluster cost data to determine different types of cost categories; Use neural network algorithms to build cost prediction models, and train multi-layer feedforward neural networks based on clustering results and historical cost data to predict future costs; Graph neural network is introduced to analyze the correlation between devices, the adjacency matrix is ​​used to represent the connection relationship between devices, and the multi-layer perceptron is used to quantitatively analyze the impact between devices.

3. The method according to claim 2, characterized in that The K-means algorithm specifically includes iteratively calculating cluster centers, minimizing intra-cluster variance, allocating cost data points to the closest cluster center until convergence, and determining different types of cost categories.

4. The method according to claim 2, characterized in that: The graph neural network represents the correlation between devices through an adjacency matrix, and quantitatively analyzes the impact between devices through a multi-layer perceptron.

5. The method according to claim 1, characterized in that The specific formula of the transfer learning method is: ; in, Represents the loss function of the transfer learning method; is the number of samples of source domain data; Represents the loss function of the source domain data; and Represent source domain samples and labels respectively; is the number of samples of target domain data; A distance metric representing the characteristics of the target domain and the source domain; and Feature extractors representing source domain and target domain respectively; is the regularization parameter; It is a model trained on source domain data and used to make predictions on source domain data; It is a feature extractor that extracts features from source domain data; It is a feature extractor that extracts features from the target domain data for cost prediction in the target domain.

6. A system for implementing the equipment life cycle cost management method according to any one of claims 1 to 5, characterized in that: include: Sensor network, used to collect real-time data on equipment operation, material usage, and labor time records; The data processing module is used to clean and standardize the collected data to form a comprehensive data set, and to integrate, unify the format, verify and normalize the comprehensive data set to form a standardized data set; The multi-dimensional cost analysis module is used to classify the cost data in the normalized data set, identify direct costs and indirect costs, and build a multi-dimensional cost analysis model using data mining technology and machine learning algorithms. It analyzes the correlation between devices by introducing graph neural networks, and uses a dynamic multi-level analysis framework to improve the accuracy of the model and generate equipment cost analysis result data; The optimization module is used to dynamically optimize the equipment cost analysis result data using the reinforcement learning algorithm and the collaborative optimization mechanism, calculate the cost allocation ratio of each equipment in different maintenance tasks, adjust the allocation strategy through the fuzzy clustering algorithm and the adaptive allocation mechanism, and generate the final cost allocation result; The prediction and scheduling module is used to build and optimize the cost prediction model based on historical data and real-time data, use transfer learning and intelligent scheduling algorithms to perform real-time cost prediction, generate optimal maintenance plans and resource allocation plans, combine feedback optimization mechanisms for dynamic adjustments, and finally store the optimization results.

Citation Information

Cited By

  • Dynamic management method and system for optimizing full life cycle cost of equipment

    CN120372502A