Model cluster chain system applied to electromechanical equipment operation and maintenance

Abstract

The invention belongs to the technical field of electromechanical equipment, and particularly relates to a model cluster chain system applied to operation and maintenance of electromechanical equipment, which can realize the functions of fault prediction, false alarm reduction, decision assistance and the like, and greatly improves the efficiency and the intelligent degree of operation and maintenance work. The model cluster chain system can adapt to equipment of different manufacturers, new and old degrees, service life and environments, complete coverage of an operation and maintenance model is achieved, and meanwhile the generality and individual management requirements of the equipment are considered through construction of generality, intermediate and specific algorithm models. Unnecessary maintenance work is reduced through intelligent operation and maintenance, the operation and maintenance cost is reduced, meanwhile, major faults can be effectively prevented through a real-time monitoring and fault early warning mechanism, and safety management is enhanced. Through introduction of the operation and maintenance management platform, a whole-process closed loop from data acquisition to work order management is realized, systematicness and coherence of operation and maintenance work are ensured, and the overall operation and maintenance management level is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electromechanical equipment, and in particular relates to a model cluster chain system applied to the operation and maintenance of electromechanical equipment. Background Art

[0002] With the increasing number and complexity of electromechanical equipment in modern buildings, its operation and maintenance management faces unprecedented challenges. The traditional manual operation and maintenance model is not only inefficient, but also difficult to adapt to the situation of a wide variety of equipment, huge amounts of data, and different environmental conditions, which often leads to problems such as delayed fault response, high maintenance costs, and increased safety risks. Especially in large buildings, electromechanical facilities cover multiple systems such as air-conditioning systems, elevators, fire protection, and lighting, involving thousands of types of equipment, each of which has different maintenance requirements and life cycles, making operation and maintenance work extremely complex and highly professional. In the existing operation and maintenance system, common problems are the lack of systematic management solutions, scattered equipment maintenance records, backward fault prediction and diagnosis methods, and insufficient application of big data and artificial intelligence technologies.

[0003] How to effectively integrate electromechanical equipment data from different sources and with different characteristics, build an intelligent operation and maintenance system that can cover a wide range of equipment types and target the characteristics of individual equipment, achieve efficient management, fault prediction and rapid response of equipment operation and maintenance, and optimize operation and maintenance costs, has become a key technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a model cluster chain system applied to the operation and maintenance of electromechanical equipment, which improves the operation and maintenance efficiency and intelligence level, realizes functions such as fault prediction, false alarm reduction, and decision assistance, greatly improves the efficiency and intelligence of operation and maintenance work, and solves the problems in the prior art raised in the above background technology.

[0005] To achieve the above purpose, the present invention adopts the following technical solution: a model cluster chain system applied to the operation and maintenance of electromechanical equipment, comprising:

[0006] Data collection module, used to collect operation data, maintenance history, inspection records and maintenance information of electromechanical equipment of different equipment, different manufacturers, different degrees of oldness, different years of use and different environments;

[0007] A model building module, based on the data collected by the data acquisition module, builds multiple operation and maintenance models for each type of equipment, including but not limited to a model for reducing the false alarm rate of IoT equipment, a model for predicting equipment failure, and a model for assisting fault judgment;

[0008] The attribute analysis and algorithm module is used to conduct in-depth analysis of the operating data of building mechanical and electrical equipment, extract common attributes and weight them to build a common algorithm model, extract related attributes and weight them to build an intermediate algorithm model, and build a unique algorithm model for unique attributes;

[0009] The cluster chain building module uses common algorithm models as a link to link and integrate various AI algorithm models to form a cluster chain of intelligent operation and maintenance models for electromechanical equipment;

[0010] The operation and maintenance management platform integrates the functions of various modules to achieve real-time equipment status monitoring, automatic fault warning, closed-loop work order management, AI-assisted inspection, decision support and intelligent operation and maintenance.

[0011] Preferably, the data acquisition module includes a wireless sensor network for remotely acquiring equipment operating parameters in real time, and the formula is expressed as:

[0012] [D=\sum_{i=1}^{n}{P_i}], where D represents the total data set collected and P_i represents the parameter set of the i-th device.

[0013] Preferably, the model building module uses random forest to establish a prediction model, and the formula is: [F(\text{Fau lt})=RF(X;\theta)], where F(Fau lt) represents the predicted failure probability and X is the input feature vector.

[0014] Preferably, the commonality algorithm model extracts commonality features between devices through matrix decomposition technology, and the formula is: [U, S, V^T = SVD(C)], where C is the device feature matrix, and U, S, and V^T are the left singular matrix, diagonal matrix, and right singular matrix, respectively.

[0015] Preferably, the intermediate algorithm model is mined through association rules, and the formula is:

[0016] [Conf(A\RightB)=\frac{Sup(A\cupB)}{Sup(A)}], where Conf represents confidence, A and B are two events, and Sup represents support.

[0017] Preferably, the unique algorithm model processes device data through a convolutional neural network, and the formula is: [Y=CNN(X;W,b)], where Y is the output layer prediction result, X is the input data, and W and b represent the weight matrix and bias term respectively.

[0018] Preferably, the cluster chain construction module uses graph theory to construct a connectivity graph between device models and optimizes the model calling sequence through the shortest path algorithm, the formula is:

[0019] [d(i,j)=\min_{p\i nP_{ij}}\sum_{u,v\i np}w(u,v)], where d(i,j) represents the shortest distance from model i to model j, P_{ij} is the set of all possible paths, and w(u,v) is the weight of the edge.

[0020] Preferably, the operation and maintenance management platform includes a resource scheduling mechanism based on cloud computing, which is used to dynamically allocate computing resources according to model computing requirements. The formula is: [R_t=\α\cdotD_t+\β\cdotM_t], where R_t is the required amount of resources at time t, D_t is the amount of data to be processed, and M_t is the model complexity; α and β are weight coefficients determined based on experience or optimization algorithms.

[0021] Preferably, the automatic fault warning is combined with Bayesian theorem to perform probability update, and the formula is: [P(H|E)=\frac{P(E|H)\cdotP(H)}{P(E)}], where P(H|E) is the probability of fault hypothesis H after observing evidence E.

[0022] Preferably, the AI-assisted inspection integrates a deep reinforcement learning algorithm to enable the inspection robot or software agent to autonomously learn the optimal inspection path and inspection strategy, which is expressed as follows:

[0023] [Q(s,a)=Q(s,a)+\alpha[r+\gamma\max_{a'}Q(s',a')-Q(s,a)]], where Q(s,a) represents the expected return of taking action a in state s, α is the learning rate, r is the immediate reward, γ is the discount factor, s′ and a′ represent the next state and the corresponding best action, respectively.

[0024] Technical effects and advantages of the present invention: Compared with the prior art, the model cluster chain system for operation and maintenance of electromechanical equipment proposed by the present invention has the following advantages:

[0025] The present invention can automatically collect and integrate multi-source data, establish customized operation and maintenance models for each type of equipment, realize functions such as fault prediction, false alarm reduction, and decision-making assistance, and greatly improve the efficiency and intelligence of operation and maintenance work. The model cluster chain system can adapt to equipment from different manufacturers, newness, age, and environment, and achieve comprehensive coverage of the operation and maintenance model. At the same time, through the construction of common, intermediate, and unique algorithm models, it takes into account the common and individual management needs of the equipment. Intelligent operation and maintenance can reduce unnecessary maintenance operations and reduce operation and maintenance costs. At the same time, real-time monitoring and fault warning mechanisms can effectively prevent the occurrence of major failures and enhance safety management. The introduction of the operation and maintenance management platform realizes a closed loop of the entire process from data collection to work order management, ensures the systematicness and consistency of operation and maintenance work, and improves the overall operation and maintenance management level. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The block diagram of the model cluster chain system applied to the operation and maintenance of electromechanical equipment according to the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] The present invention provides Figure 1 The model cluster chain system shown is applied to the operation and maintenance of electromechanical equipment. The system includes the following components and steps: data acquisition module, model construction module, attribute analysis and algorithm module, cluster chain construction module and operation and maintenance management platform. The details are as follows:

[0029] Data collection module, used to collect operation data, maintenance history, inspection records and maintenance information of electromechanical equipment of different equipment, different manufacturers, different degrees of oldness, different years of use and different environments;

[0030] A model building module, based on the collected data, builds multiple operation and maintenance models for each type of equipment, including but not limited to a model for reducing the false alarm rate of IoT equipment, a model for predicting equipment failure, and a model for assisting fault judgment;

[0031] The attribute analysis and algorithm module conducts in-depth analysis on the operation data of thousands of devices in dozens of systems of building electromechanical equipment, extracts common attributes and weights them to build common algorithm models, extracts related attributes and weights them to build intermediate algorithm models, and builds unique algorithm models for unique attributes;

[0032] The cluster chain construction module uses common algorithm models as a link to link and integrate the above-mentioned AI algorithm models to form a cluster chain of intelligent operation and maintenance models for electromechanical equipment, achieving comprehensive coverage of the operation and maintenance model, improving efficiency, and reducing costs;

[0033] The operation and maintenance management platform integrates the functions of the above modules to achieve real-time equipment status monitoring, automatic fault warning, closed-loop work order management, AI-assisted inspection, decision support and intelligent operation and maintenance, significantly improving operation and maintenance efficiency and safety.

[0034] The data acquisition module further includes a wireless sensor network for remotely acquiring equipment operating parameters in real time. The formula is:

[0035] [D=\sum_{i=1}^{n}{P_i}], where D represents the total data set collected and P_i represents the parameter set of the i-th device.

[0036] Furthermore, the data acquisition module implements the scheme of remotely acquiring equipment operating parameters in real time through the following submodules:

[0037] Sensor deployment submodule: Design and deploy various wireless sensors, such as temperature sensors, vibration sensors, current and voltage sensors, etc., according to the type, layout and monitoring requirements of electromechanical equipment to ensure comprehensive coverage of key operating parameters. In this step, the type selection, installation location and network topology design of the sensor need to be considered to ensure the accuracy of data collection and the stability of the network.

[0038] Signal transmission submodule: It uses low-power wide area network (LPWAN) technology, such as LoRa, NB-IoT, etc., to wirelessly transmit the raw data collected by each sensor to the data aggregation node or cloud server. This module needs to solve problems such as long-distance transmission, energy consumption control and data encryption security to ensure efficient, reliable and secure data transmission.

[0039] Data preprocessing submodule: At the data receiving end, the raw data is preliminarily cleaned and formatted, including removing outliers, data smoothing, unit conversion, etc., for subsequent analysis. The formula [D=\sum_{i=1}^{n}{P_i}] is reflected in this link, and the parameter set P_i of each device i is accumulated to form a unified data set D.

[0040] Time synchronization submodule: To ensure the time consistency of collected data, this submodule is responsible for synchronizing the timestamps of all sensors in the network, using NTP (Network Time Protocol) or other precise clock synchronization technologies to ensure the accuracy of data analysis timing.

[0041] Remote configuration and management submodule: It provides a remote access interface, allowing operation and maintenance personnel to remotely configure the sensor network, adjust parameters, diagnose faults, and upgrade software. This not only improves operation and maintenance efficiency, but also ensures the flexibility and maintainability of the entire data acquisition system.

[0042] Through the collaborative work of the above five sub-modules, the data acquisition module can effectively obtain operating parameters remotely and in real time from widely distributed electromechanical equipment, providing a solid data foundation for subsequent intelligent operation and maintenance models.

[0043] Exemplarily, the model building module uses a machine learning algorithm, such as random forest (RF), to build a prediction model, and the formula is:

[0044] [F(\text{Fau lt})=RF(X;\theta)]; where F(Fau lt) represents the predicted failure probability and X is the input feature vector.

[0045] Furthermore, the model building module uses the random forest (RF) algorithm to build a prediction model, and the specific implementation method involves the following submodules:

[0046] Feature selection and engineering submodule: Based on equipment operation data, maintenance records, inspection reports and maintenance information, statistical analysis, correlation test and other methods are used to filter out features that are highly correlated with equipment failures from the original data to construct the input feature vector X. At the same time, necessary feature engineering, such as normalization and encoding conversion, is performed to optimize the model input quality.

[0047] Data segmentation submodule: Divide the sorted dataset into training and test sets, usually in a ratio of 70%:30% or 80%:20%, to ensure that the generalization ability can be effectively evaluated while the model is being trained. This step may involve cross-validation to improve model robustness.

[0048] Random forest model initialization submodule: Determine the basic parameters of the random forest, such as the number of trees n_trees, the percentage of feature selection m_try, the maximum depth of the tree max_depth, etc., and initialize the random forest model framework.

[0049] Decision tree generation submodule: For each decision tree, samples are randomly selected from the training set through the Bootstrapping method, and feature subsets are randomly selected at each node to find the best split point, gradually building the tree structure. This process is repeated until the predetermined tree depth or node purity standard is reached.

[0050] Model training and optimization submodule: Use the training set data to train each tree in the random forest, and use the ensemble learning principle to summarize the prediction results of all decision trees to obtain the final fault prediction probability F (Fau lt). The process may include tuning of hyperparameters to maximize model performance.

[0051] Model evaluation and verification submodule: Use the test set to evaluate the model, and use indicators such as accuracy, recall, F1 score or AUC-ROC curve to measure the model's ability to predict faults. Adjust model parameters based on evaluation feedback and perform iterative optimization.

[0052] Model deployment and monitoring submodule: Deploy the optimized random forest model to the operation and maintenance system, receive equipment operation data in real time, and output the fault prediction probability. At the same time, set up a monitoring mechanism to track model performance, and update or retrain the model regularly or as needed to ensure the continuity and accuracy of the prediction effect.

[0053] Through the orderly execution of these seven sub-modules, the model building module can effectively use the random forest algorithm to build an efficient model for predicting equipment failures, providing a scientific basis for the intelligent operation and maintenance of electromechanical equipment.

[0054] Furthermore, the commonality algorithm model extracts common features between devices through matrix decomposition technology. The formula is: [U, S, V^T = SVD(C)], where C is the device feature matrix, U, S, and V^T are the left singular matrix, diagonal matrix, and right singular matrix, respectively, which are used to construct the commonality feature vector.

[0055] The commonality algorithm model uses matrix decomposition technology, especially singular value decomposition (SVD) to extract common features between devices. The specific implementation method may include the following submodules:

[0056] Data preprocessing submodule: collects the operation data, maintenance history, inspection records and maintenance information of each electromechanical equipment, performs data cleaning, removes outliers and missing values, unifies the data format, and ensures data quality. These multi-dimensional information are integrated into an equipment feature matrix C.

[0057] Matrix normalization submodule: Normalizes the device feature matrix C, such as Z-score normalization or Min-Max scaling, to make features of different scales comparable and facilitate subsequent matrix decomposition operations.

[0058] Singular value decomposition calculation submodule: Apply the SVD algorithm to decompose the device feature matrix C, and obtain C = U·S·V^T, where U is the left singular matrix, S is the diagonal matrix containing singular values, and V^T is the transpose of the right singular matrix. The singular values ​​are arranged from large to small, reflecting the main feature direction of the data and its importance.

[0059] Common feature extraction submodule: Based on the results of SVD decomposition, the singular vectors corresponding to the first k largest singular values ​​are selected to construct common feature vectors. These vectors contain the main common patterns of the data, which can effectively reduce the dimension while retaining most of the information of the data.

[0060] Feature reconstruction submodule: Use the selected common feature vectors to reconstruct a low-dimensional device feature matrix, which retains the main common features of the original matrix and can be used for further model construction and analysis.

[0061] Model training and optimization submodule: Based on the reconstructed low-dimensional common feature matrix, various operation and maintenance models are trained, such as prediction models, fault diagnosis models, etc. Model parameters are optimized through cross-validation, grid search and other methods to improve the model's prediction ability and generalization performance.

[0062] Model evaluation and application submodule: Use independent test sets to evaluate the performance of the model, including but not limited to accuracy, precision, recall, and F1 score. Adjust the model or feature selection strategy based on the evaluation results. Finally, integrate the optimized model into the operation and maintenance system to achieve real-time monitoring, fault warning and other functions, and continuously track and optimize the model effect.

[0063] Through the above seven sub-modules, SVD technology can be effectively used to extract and utilize the common features between devices, build an efficient and generalized operation and maintenance model, and provide strong support for the intelligent operation and maintenance of electromechanical equipment.

[0064] Furthermore, the intermediate algorithm model uses association rule mining, and the formula is:

[0065] [Conf(A\RightB)=\frac{Sup(A\cupB)}{Sup(A)}], where Conf represents confidence, A and B are two events, and Sup represents support, which is used to identify potential associations between devices.

[0066] The intermediate algorithm model uses association rule mining technology to identify potential associations between devices. The specific implementation can be achieved through the following submodules:

[0067] Data preparation and preprocessing submodule: First, extract relevant information from equipment operation data, maintenance records, inspection data, and maintenance data to form a transaction database. Then perform data cleaning, including removing invalid records, processing missing values, data type conversion, etc., to ensure data quality.

[0068] Transaction data conversion submodule: converts the preprocessed data into a transaction data format suitable for association rule analysis, that is, the operating status and maintenance history of each device are regarded as "items" and each record is regarded as a "shopping basket". For example, a "shopping basket" may contain events such as "Device A is operating normally" and "Device B has undergone a maintenance".

[0069] Support threshold setting submodule: According to business requirements and data characteristics, the minimum support threshold (min_sup) is set to filter out infrequent event combinations. Event combinations with higher support are more likely to reflect the actual association between devices.

[0070] Frequent item set mining submodule: Apply April algorithm or other frequent item set mining algorithms to find all frequent item sets that meet the minimum support threshold. The algorithm gradually expands the candidate item set and calculates the support through continuous iteration until no new frequent item sets can be found.

[0071] Association rule generation and evaluation submodule: Generate candidate association rules based on frequent item sets and calculate the confidence of each rule using the formula:

[0072] [Conf(A\RightarrowB)=\frac{Sup(A\cupB)}{Sup(A)}] The rules that meet the preset confidence threshold (min_conf) are screened out, and these rules are regarded as evidence of strong association between devices.

[0073] Rule interpretation and application submodule: interpret the mined strong correlation rules to understand the mutual influence relationship between devices, such as "after device A is repaired, the probability of device B failure increases". Apply these rules to operation and maintenance strategy formulation, fault prevention, equipment layout optimization, etc. to achieve intelligent operation and maintenance decision-making.

[0074] Through the collaboration of the above six sub-modules, the intermediate algorithm model can deeply explore the implicit correlations in the equipment data, provide a scientific basis for the intelligent operation and maintenance of electromechanical equipment, and effectively improve the operation and maintenance efficiency and preventive maintenance capabilities.

[0075] Furthermore, the unique algorithm model processes device-specific data through a deep learning network, such as a convolutional neural network (CNN), and the formula is: [Y=CNN(X;W,b)], where Y is the output layer prediction result, X is the input data, and W and b represent the weight matrix and bias term, respectively.

[0076] The unique algorithm model uses the convolutional neural network (CNN) in deep learning to process device-specific data. The specific implementation can be achieved through the following submodules:

[0077] Data collection and preprocessing submodule: First, a large amount of relevant data is collected for specific equipment, including but not limited to sensor readings, operation logs, fault records, etc. The data preprocessing stage involves normalization and standardization, removal of outliers, and division of data into training sets, validation sets, and test sets to prepare for subsequent model training.

[0078] Feature Engineering Submodule: For specific types of data such as images or time series, feature extraction is performed, which may include edge detection of images, texture feature extraction, or frequency domain transformation of time series. For non-image data, it may also be necessary to design or select appropriate feature representations so that CNN can better capture patterns in the data.

[0079] Model design and construction submodule: Design the CNN model architecture according to the characteristics of the device data, including determining the number of convolutional layers, filter size, pooling layer type, whether to add a fully connected layer, etc. Select appropriate activation functions, loss functions, and optimization algorithms, such as ReLU, cross entropy loss, and Adam optimizer, etc. Use a deep learning framework (such as TensorFlow or PyTorch) to build the model.

[0080] Model training and tuning submodule: Train the CNN model on the training set, and adjust the weight matrix W and the bias term b through back propagation. Monitor the performance indicators (such as accuracy, recall, etc.) on the validation set, and apply strategies such as early stopping or learning rate decay to prevent overfitting. Adjust the model structure or hyperparameters based on the validation results, such as increasing the number of layers, changing the number of filters, etc., to perform model tuning.

[0081] Model evaluation and deployment submodule: Evaluate the generalization ability of the model on an independent test set to ensure that the model not only performs well on the training data, but also has stable performance on unseen data. After that, deploy the trained model to the production environment to achieve real-time monitoring, fault prediction, or performance evaluation of specific equipment. At the same time, establish a model update mechanism to regularly retrain or fine-tune the model with new data to maintain the timeliness and accuracy of the model.

[0082] Through the implementation of this series of sub-modules, the unique algorithm model can efficiently learn the specific data of different devices, thereby providing accurate operation and maintenance decision support.

[0083] Furthermore, the cluster chain building module uses graph theory to build a connectivity graph between device models and optimizes the model calling sequence through the shortest path algorithm. The formula is:

[0084] [d(i,j)=\min_{p\i nP_{ij}}\sum_{u,v\i np}w(u,v)], where d(i,j) represents the shortest distance from model i to model j, P_{ij} is the set of all possible paths, and w(u,v) is the weight of the edge.

[0085] The cluster chain building module optimizes the model calling sequence through graph theory, and the specific implementation method includes the following submodules:

[0086] Model node definition submodule: First, define a node in the graph for each constructed operation and maintenance model. Each node represents a specific operation and maintenance model, such as a fault prediction model, a fault judgment model, etc. The relationship between nodes reflects the information flow or dependency between models.

[0087] Edge construction and weight assignment submodule: Based on the relationship and interaction frequency between models, the edges in the graph are constructed. If the output of model i is the input of model j or the two have a direct cooperative relationship, an edge is established between i and j. At the same time, weights w(u,v) are assigned to the edges according to the cost of model call (such as computational complexity, response time, data transfer volume) or synergy benefits, which directly affects the selection of the shortest path.

[0088] Connectivity graph construction submodule: Integrate all model nodes and edges to form a directed graph, ensuring that the logical relationships and dependency structures between models are clearly presented. This step may include checking the connectivity of the graph to ensure that any two models can be connected through a series of edges.

[0089] Shortest path algorithm application submodule: Apply the shortest path algorithm (such as Dijkstra algorithm, Floyd-Warshall algorithm or A* algorithm, etc.) to calculate the shortest path [d(i,j)] between any two model nodes i and j in the graph. The formula is

[0090] [d(i,j)=\min_{p\i nP_{ij}}\sum_{u,v\i np}w(u,v)] is used to quantify this process, where P_{ij} is the set of all possible paths from i to j.

[0091] Model call sequence optimization submodule: Based on the shortest path calculation results, optimize the order of model calls to ensure efficient execution of models in the operation and maintenance process. Prioritize model combinations with shorter path lengths to reduce processing time and resource consumption, while taking into account the efficiency and effectiveness of collaborative work between models.

[0092] Dynamic adjustment and monitoring submodule: As time goes by and the device status changes, the model performance and dependencies may change. Therefore, it is necessary to continuously monitor the actual effect of the model call, such as response time, accuracy and other indicators, and dynamically adjust the structure and weight of the model graph as needed to ensure the continuous optimization and adaptability of the cluster chain.

[0093] Through the cooperation of the above sub-modules, the cluster chain construction module can efficiently organize and optimize the calling sequence between models, optimize the operation and maintenance process, and improve the efficiency and intelligence level of equipment operation and maintenance.

[0094] Furthermore, the operation and maintenance management platform also includes a resource scheduling mechanism based on cloud computing, which dynamically allocates computing resources according to the model computing requirements to improve processing efficiency. The formula is: [R_t=\α\cdotD_t+\β\cdotM_t], where R_t is the required amount of resources at time t, D_t is the amount of data to be processed, and M_t is the model complexity; α and β are weight coefficients determined based on experience or optimization algorithms.

[0095] The resource scheduling mechanism in the operation and maintenance management platform takes advantage of cloud computing and implements dynamic resource allocation through the following submodules to improve processing efficiency:

[0096] Resource monitoring and analysis submodule: real-time monitoring of the usage of various computing resources on the cloud platform, including CPU occupancy, memory usage, storage space, etc., while analyzing the resource consumption trend and historical data of the current running model to provide basic information for resource allocation.

[0097] Demand forecasting submodule: Based on historical data and real-time monitoring information, time series analysis, machine learning and other methods are used to predict the amount of data to be processed Dt and the model complexity Mt at a future time point t, providing a basis for the calculation of resource demand.

[0098] Resource demand calculation submodule: According to the formula [R_t=\α\cdot D_t+\β\cdot M_t], the amount of resources R_t required at time t is calculated, where α and β are weight coefficients determined based on experience or optimization algorithms, reflecting the impact of data volume and model complexity on resource demand.

[0099] Resource scheduling strategy submodule: Design and implement resource scheduling strategies to dynamically adjust cloud resource allocation based on the calculated resource demand R_t. Strategies can include load balancing, priority scheduling, reserved resources, elastic scaling, etc., to ensure that there are sufficient resources to support high-load periods and release resources to save costs during low-load periods.

[0100] Automated deployment and migration submodule: Automatically deploy the model to the most suitable computing node according to the resource scheduling strategy, or automatically migrate the model to other available resources when resources are tight, to ensure task continuity and high availability, while reducing manual intervention and improving efficiency.

[0101] Monitoring and feedback optimization submodule: Continuously monitor the resource scheduling execution and model operation status, collect performance indicators and resource utilization efficiency data, and feed back to the demand forecasting and scheduling strategy module, continuously optimize the scheduling algorithm and parameter settings, and realize continuous optimization and automatic adjustment of resource allocation.

[0102] Through the close collaboration of the above sub-modules, the operation and maintenance management platform can efficiently and flexibly manage cloud computing resources and dynamically adjust resource allocation according to model computing requirements, thereby significantly improving equipment operation and maintenance processing efficiency and overall system performance.

[0103] Furthermore, the automatic fault warning function combines the Bayesian theorem to perform probability updates, and the formula is: [P(H|E)=\frac{P(E|H)\cdotP(H)}{P(E)}], where P(H|E) is the probability of the fault hypothesis H after observing evidence E, achieving more accurate fault prediction. The automatic fault warning function uses the Bayesian theorem to perform probability updates, and the specific implementation plan includes the following submodules:

[0104] Data collection and preprocessing submodule: First, real-time and historical data are collected from equipment sensors, log records, maintenance records and other channels, and then cleaned, standardized and normalized to ensure data quality and provide an accurate basis for subsequent analysis.

[0105] Feature extraction and selection submodule: Extract features closely related to equipment failure from preprocessed data, such as vibration amplitude, temperature anomaly, operating noise, etc. Use feature selection algorithms (such as recursive feature elimination and model-based feature selection) to select the feature set with the most predictive value.

[0106] Prior probability definition submodule: Based on historical fault data and equipment knowledge, define the fault hypothesis H (equipment failure) and the prior probability P(H) under each feature condition, as well as the probability of the feature appearing in the fault-free state. These prior probabilities provide a starting point for subsequent Bayesian updates.

[0107] Conditional probability calculation submodule: For each feature E, the conditional probability P(E|H) when the fault hypothesis H is established is calculated based on historical data, as well as the marginal probability P(E) of the feature appearing in all cases. This step is the core of applying Bayes' theorem.

[0108] Bayesian update submodule: When new evidence E (such as abnormal sensor readings) appears, the Bayesian theorem formula [P(H|E)=\frac{P(E|H)\cdot P(H)}{P(E)}] is used to update the posterior probability of the fault hypothesis H. This process is repeated continuously, and the accuracy and precision of fault prediction gradually improves with the addition of new data.

[0109] Early warning threshold setting and response submodule: The fault early warning threshold is set according to the posterior probability. When the failure probability of a certain device exceeds this threshold, the early warning mechanism is triggered. At the same time, the corresponding response strategy is designed, such as sending an alarm to the maintenance team, automatically arranging maintenance tasks, or starting the backup system, etc., to quickly respond to potential failures.

[0110] Through the interaction of these sub-modules, the automatic fault warning function can not only monitor the equipment status in real time, but also dynamically adjust the fault prediction probability based on historical and current data, achieve more accurate fault prediction, and effectively improve the operation and maintenance efficiency and safety of electromechanical equipment.

[0111] Furthermore, AI-assisted inspection integrates the deep reinforcement learning (DRL) algorithm, which enables the inspection robot or software agent to autonomously learn the optimal inspection path and inspection strategy to optimize the inspection efficiency. The formula is expressed as:

[0112] [Q(s,a)=Q(s,a)+\alpha[r+\gamma\max_{a'}Q(s',a')-Q(s,a)]], where Q(s,a) represents the expected return of taking action a in state s, α is the learning rate, r is the immediate reward, γ is the discount factor, s′ and a′ represent the next state and the corresponding optimal action, respectively, so as to achieve efficient automatic inspection and maintenance guidance for electromechanical equipment.

[0113] The AI-assisted inspection solution uses a deep reinforcement learning (DRL) algorithm to achieve autonomous learning and optimization of the inspection robot or software agent. The specific implementation method can be divided into the following sub-modules:

[0114] Environmental modeling and state representation submodule: First, a digital model of the electromechanical equipment inspection environment is constructed, including factors such as equipment layout, physical obstacles, and equipment status. Then, a state representation method is designed to encode environmental information into a machine-processable vector form, such as image pixels, sensor readings, etc.

[0115] Action space definition submodule: Define all possible actions that the inspection robot or software agent can perform, such as movement direction, inspection action, use of specific sensors, etc., to form an action space. These actions should fully cover all necessary operations in the inspection process.

[0116] Reward function design submodule: Design a reward r mechanism to guide AI to learn effective strategies. Rewards can be based on factors such as efficiency in completing tasks, timeliness in discovering faults, safety in avoiding collisions, etc., to ensure that the robot or agent can learn the optimal behavior during the exploration process.

[0117] Deep Q network architecture design submodule: Build a deep neural network architecture to approximate the Q function Q(s,a), where the network input is the current state s and action a, and the output is the expected reward. The network design needs to consider network depth, hidden layer size, activation function, etc. to optimize the learning effect.

[0118] Experience replay buffer management submodule: maintain an experience replay buffer to store the historical state-action-reward-new state quadruple (s, a, r, s′), and randomly sample from it for batch training to break the correlation between data and improve the stability and generalization ability of learning.

[0119] Strategy and value function update submodule: According to the formula:

[0120] [Q(s,a)=Q(s,a)+\al pha[r+\gamma\max_{a'}Q(s',a')-Q(s,a)]], update the network weights through optimization methods such as gradient descent, so that the Q function approaches the optimal strategy. The learning rate α and discount factor γ need to be adjusted appropriately to balance the learning speed and long-term benefits.

[0121] Exploration and Exploitation Balance Submodule: Implement ε-greedy strategy or other exploration mechanisms to balance the need to explore unknown areas and exploit known optimal strategies. High ε values ​​at the beginning encourage exploration, and gradually decrease as learning deepens to make more use of learned knowledge.

[0122] Simulation and real environment interaction submodule: First train in a simulation environment to accelerate the learning process and reduce costs. Then, deploy the trained model to the real inspection scene. Through continuous learning and adjustment, ensure that the algorithm can adapt to changes in the actual environment and achieve efficient and reliable automatic inspection.

[0123] Through the collaborative work of these sub-modules, the DRL-driven AI-assisted inspection system can continuously optimize its inspection paths and strategies, improve inspection efficiency and accuracy, and bring intelligent innovation to the maintenance and management of electromechanical equipment.

[0124] In summary, the present invention can automatically collect and integrate multi-source data, establish customized operation and maintenance models for each type of equipment, realize functions such as fault prediction, false alarm reduction, and decision-making assistance, and greatly improve the efficiency and intelligence of operation and maintenance work. The model cluster chain system can adapt to equipment from different manufacturers, newness, age, and environment, and achieve comprehensive coverage of the operation and maintenance model. At the same time, through the construction of common, intermediate, and unique algorithm models, it takes into account the common and individual management needs of the equipment. Intelligent operation and maintenance can reduce unnecessary maintenance operations and reduce operation and maintenance costs. At the same time, real-time monitoring and fault warning mechanisms can effectively prevent the occurrence of major failures and enhance safety management. The introduction of the operation and maintenance management platform realizes a closed loop of the entire process from data collection to work order management, ensures the systematicness and consistency of operation and maintenance work, and improves the overall operation and maintenance management level.

[0125] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A model cluster chain system applied to electromechanical equipment operation and maintenance, characterized in that: include: Data collection module, used to collect operation data, maintenance history, inspection records and maintenance information of electromechanical equipment of different equipment, different manufacturers, different degrees of oldness, different years of use and different environments; A model building module, based on the data collected by the data acquisition module, builds multiple operation and maintenance models for each type of equipment, including but not limited to a model for reducing the false alarm rate of IoT equipment, a model for predicting equipment failure, and a model for assisting fault judgment; The attribute analysis and algorithm module is used to conduct in-depth analysis of the operating data of building mechanical and electrical equipment, extract common attributes and weight them to build a common algorithm model, extract related attributes and weight them to build an intermediate algorithm model, and build a unique algorithm model for unique attributes; The cluster chain building module uses common algorithm models as a link to link and integrate various AI algorithm models to form a cluster chain of intelligent operation and maintenance models for electromechanical equipment; The operation and maintenance management platform integrates the functions of various modules to achieve real-time equipment status monitoring, automatic fault warning, closed-loop work order management, AI-assisted inspection, decision support and intelligent operation and maintenance.

2. According to claim 1, a model cluster chain system applied to electromechanical equipment operation and maintenance is characterized in that: The data acquisition module includes a wireless sensor network for remotely acquiring equipment operating parameters in real time, and the formula is expressed as: [D=\sum_{i=1}^{n}{P_i}], where D represents the total data set collected and P_i represents the parameter set of the i-th device.

3. According to claim 1, a model cluster chain system applied to electromechanical equipment operation and maintenance is characterized in that: The model building module uses random forest to build a prediction model, and the formula is: [F(\text{Fault})=RF(X;\theta)], where F(Fault) represents the predicted fault probability and X is the input feature vector.

4. According to claim 1, a model cluster chain system applied to electromechanical equipment operation and maintenance is characterized in that: The commonality algorithm model extracts commonality features between devices through matrix decomposition technology, and the formula is: [U, S, V^T = SVD(C)], where C is the device feature matrix, and U, S, and V^T are the left singular matrix, diagonal matrix, and right singular matrix, respectively.

5. According to claim 1, a model cluster chain system applied to electromechanical equipment operation and maintenance is characterized in that: The intermediate algorithm model is mined through association rules, and the formula is: [Conf(A\RightarrowB)=\frac{Sup(A\cupB)}{Sup(A)}], where Conf represents confidence, A and B are two events, and Sup represents support.

6. A model cluster chain system applied to electromechanical equipment operation and maintenance according to claim 1, characterized in that: The unique algorithm model processes device data through a convolutional neural network, and the formula is: [Y=CNN(X;W,b)], where Y is the output layer prediction result, X is the input data, and W and b represent the weight matrix and bias term respectively.

7. A model cluster chain system for electromechanical equipment operation and maintenance according to claim 1, characterized in that: The cluster chain building module uses graph theory to build a connectivity graph between device models and optimizes the model calling sequence through the shortest path algorithm. The formula is: [d(i,j)=\min_{p\inP_{ij}}\sum_{u,v\inp}w(u,v)], where d(i,j) represents the shortest distance from model i to model j, P_{ij} is the set of all possible paths, and w(u,v) is the weight of the edge.

8. The model cluster chain system applied to electromechanical equipment operation and maintenance according to claim 1 is characterized in that: The operation and maintenance management platform includes a resource scheduling mechanism based on cloud computing, which is used to dynamically allocate computing resources according to the model computing requirements. The formula is: [R_t=\α\cdotD_t+\β\cdotM_t], where R_t is the required amount of resources at time t, D_t is the amount of data to be processed, and M_t is the model complexity; α and β are weight coefficients determined based on experience or optimization algorithms.

9. A model cluster chain system applied to electromechanical equipment operation and maintenance according to claim 1, characterized in that: The automatic fault warning is combined with Bayesian theorem to perform probability update, and the formula is: [P(H|E)=\frac{P(E|H)\cdotP(H)}{P(E)}], where P(H|E) is the probability of fault hypothesis H after observing evidence E.

10. The model cluster chain system applied to electromechanical equipment operation and maintenance according to claim 1, characterized in that: The AI-assisted inspection integrates a deep reinforcement learning algorithm to enable the inspection robot or software agent to autonomously learn the optimal inspection path and inspection strategy. The formula is expressed as: [Q(s,a)=Q(s,a)+\alpha[r+\gamma\max_{a'}Q(s',a')-Q(s,a)]], where Q(s,a) represents the expected return of taking action a in state s, α is the learning rate, r is the immediate reward, γ is the discount factor, s′ and a′ represent the next state and the corresponding best action, respectively.

Images