Industrial device group intelligent operation and maintenance and energy consumption optimization method based on edge computing

Abstract

The application provides an industrial equipment group intelligent operation and maintenance and energy consumption optimization method based on edge computing, relates to the field of industrial equipment intelligent operation and maintenance, and comprises the following steps: collecting data through an edge node and time alignment, constructing double-layer time sequence feature representation, calculating the causal relationship between equipment, forming a fault propagation tree and a prediction atlas, constructing a multi-objective optimization model based on risk assessment to formulate a load distribution scheme, and issuing control instructions according to the risk level classification. The application realizes cooperative operation and maintenance of equipment groups, improves fault prediction accuracy, reduces energy consumption, and prolongs the service life of equipment.

Description

Technical Field

[0001] This invention relates to intelligent operation and maintenance technology for industrial equipment, and more particularly to an intelligent operation and maintenance and energy consumption optimization method for industrial equipment clusters based on edge computing. Background Technology

[0002] Intelligent operation and maintenance (O&M) and energy consumption optimization of industrial equipment clusters are key technologies for achieving efficient, safe, and energy-saving industrial production. Traditional equipment O&M methods mainly rely on periodic inspections and experience-based diagnosis, which are difficult to adapt to the complexity and high-efficiency requirements of modern industrial production. Edge computing, as an emerging distributed computing model, can provide computing, storage, and network services close to the data source, reducing the burden on central servers and lowering data transmission latency, thus providing a new technical approach for intelligent O&M and energy consumption optimization of industrial equipment clusters.

[0003] Existing industrial equipment monitoring systems typically employ a centralized architecture, requiring large amounts of data to be transmitted to a central server for processing. This results in network bandwidth pressure and processing latency, failing to meet the demands for real-time monitoring and rapid response. This is particularly problematic when dealing with a large number of devices and a surge in data volume, making it difficult to guarantee system real-time performance and reliability. Traditional equipment fault diagnosis methods mostly target single devices or simple combinations of devices, lacking in-depth analysis and modeling capabilities for the collaborative relationships among device groups. They cannot effectively identify and predict fault propagation paths and impact ranges between devices, leading to unreasonable allocation of maintenance resources, difficulty in implementing preventative maintenance, and challenges in achieving global energy consumption optimization while ensuring reliable equipment operation. This is especially true in complex production environments with dynamically changing loads, where the lack of adaptable energy consumption control mechanisms makes it difficult to balance the dual objectives of operational efficiency and energy consumption. Summary of the Invention

[0004] This invention provides a method for intelligent operation and maintenance and energy consumption optimization of industrial equipment groups based on edge computing, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for intelligent operation and maintenance and energy consumption optimization of industrial equipment clusters based on edge computing, comprising: By collecting operational status data and energy consumption data of industrial equipment groups through distributed edge nodes, and using a multi-point timestamp calibration algorithm for time alignment processing, equipment-level time series data is obtained. Wavelet transform decomposition is performed on device-level time series data to extract operational trend features and abnormal fluctuation features, and a two-layer time series feature representation is constructed by combining the physical connection sequence of the devices. Based on the dual-layer time-series feature representation, the strength of causal influence and transmission delay between devices are calculated, and an operation and maintenance coupling relationship matrix and an energy consumption transmission relationship matrix are constructed to generate device group collaborative features. Based on the collaborative characteristics of the equipment group, a fault propagation tree is constructed and a multi-path propagation simulation is performed to calculate the fault risk value and energy consumption evolution trend of each equipment node and generate a time series prediction map. Based on the time-series prediction map, a set of risk-exceeding equipment is identified, a multi-objective optimization model with maintenance resource constraints is constructed, a load allocation scheme is calculated and iteratively corrected through equipment state change feedback, and control commands are generated. Control commands are divided into multi-level execution queues according to risk priority. Load adjustment commands and maintenance strategies are issued to the set of devices with risks exceeding the threshold in a hierarchical manner. The collaborative characteristics of the device group are corrected by the device operating status, and the maintenance execution order is dynamically adjusted.

[0006] By collecting operational status data and energy consumption data of industrial equipment clusters through distributed edge nodes, and performing time alignment processing using a multi-point timestamp calibration algorithm, the equipment-level time-series data is obtained, including: The industrial equipment group is divided into multiple acquisition areas based on communication distance and equipment density. Master-slave distributed edge nodes are set up in the acquisition areas to build a hierarchical edge node communication network and generate a multi-layer acquisition network topology. Based on the analysis of the multi-layer acquisition network topology, the operating load of the equipment and the network communication load are analyzed to calculate the optimal sampling period and buffer capacity, and an adaptive sampling scheme is generated. Based on the adaptive sampling scheme, the operating status data and energy consumption data of industrial equipment groups are collected through distributed edge nodes, and a data collection timestamp and transmission timestamp mapping table is established to form the raw sampling data. A distributed clock synchronization mechanism is constructed, which uses edge nodes to calculate transmission path delay and node clock drift, generates timestamp correction compensation values, and performs distributed time alignment on the original sampled data. Multi-dimensional feature extraction and data consistency verification are performed on the time-aligned data. Based on the verification results, the data is reconstructed to obtain device-level time-series data.

[0007] Wavelet transform decomposition is performed on device-level time series data to extract operational trend features and abnormal fluctuation features. A two-layer time series feature representation is constructed by combining the physical connection sequence of the devices, including: A fluctuation analysis matrix is ​​constructed for equipment-level time series data, the fluctuation trend is calculated, and wavelet transform decomposition is performed on equipment-level time series data at different time scales based on the fluctuation trend to generate multi-scale wavelet decomposition coefficients. The trend and fluctuation components of equipment-level time series data are reconstructed using multi-scale wavelet decomposition coefficients. The phase relationship and amplitude ratio between the components are extracted to construct the operation trend characteristics. A feature analysis matrix is ​​constructed for the wave components to extract wave energy distribution and wave morphology features. Abnormal wave characteristics and wave propagation paths are calculated through wave feature analysis. Construct the device connection structure according to the physical connection order of the devices, mark the physical connection relationships of the devices, establish the device transmission path, and generate the device connection matrix; By mapping operational trend characteristics and abnormal fluctuation characteristics to the device connection matrix, and combining the spatial correlation between fluctuation propagation path and device transmission path, a two-layer temporal feature representation containing temporal and spatial characteristics is constructed.

[0008] Based on a two-layer time-series feature representation, the strength of causal influence and transmission delay between devices are calculated. An operation and maintenance coupling matrix and an energy consumption transmission matrix are constructed, generating device group collaborative features including: Multi-scale state decomposition is performed on the two-layer time series feature representation. State change features are extracted through fluctuation feature analysis and amplitude reconstruction. The state transition law is calculated by combining the change features to obtain the intensity and depth of causal influence between devices and form a causal feature sequence. Based on the causal characteristic sequence analysis of the state change process, combined with the state transition law, the equipment state response time and response amplitude are calculated, the transmission delay and transmission link between equipment are determined, and the transmission characteristic matrix is ​​constructed. The transmission feature matrix is ​​mapped and matched with the device connection order to analyze the direct impact transmission caused by state changes between adjacent devices, as well as the interval impact transmission caused by intermediate devices. The coupling degree between devices is calculated, and the operation and maintenance coupling relationship matrix is ​​constructed by combining the causal impact strength and impact depth. Based on the operation and maintenance coupling relationship matrix, energy transfer paths are identified, energy transfer efficiency is calculated by combining transfer delay and transfer link, energy transfer direction and transfer gain are determined by coupling degree analysis, and energy consumption transfer relationship matrix is ​​generated. The operation and maintenance coupling relationship matrix and the energy consumption transfer relationship matrix are fused to generate the collaborative features of the equipment group.

[0009] Based on the collaborative characteristics of the equipment group, a fault propagation tree is constructed and a multi-path propagation simulation is performed. The fault risk value and energy consumption evolution trend of each equipment node are calculated, and a time-series prediction map is generated, including: The operation feature sequence and energy consumption feature sequence are extracted from the collaborative features of the equipment group. The fluctuation cycle analysis is performed on the operation feature sequence to calculate the state change frequency vector, and the amplitude decomposition is performed on the energy consumption feature sequence to obtain the state change amplitude vector. The state evolution matrix is ​​constructed by fusing the state change frequency vector and the state change amplitude vector, and a fault propagation tree is generated based on the state evolution matrix. The location of the propagation point is determined based on the fault propagation tree. The propagation speed value is calculated using the state change frequency in the state evolution matrix. The propagation attenuation coefficient is obtained through the state change amplitude. The propagation speed value and the propagation attenuation coefficient are combined to construct the propagation range matrix. The propagation channel matrix is ​​formed using the propagation range matrix. The propagation range matrix and the propagation channel matrix are combined to generate a state transition sequence. The state transition sequence is used to calculate the transition probability. The fluctuation characteristics are constructed by combining the state change sequence. The equipment node failure risk value is calculated by using the transition probability and the fluctuation characteristics. The fault risk values ​​of equipment nodes are mapped to the state evolution matrix. The energy consumption evolution trend is calculated by combining the state change frequency vector and the state change amplitude vector. The fault risk values ​​of equipment nodes and the energy consumption evolution trend are arranged in the time dimension to generate a time series prediction map.

[0010] Based on time-series prediction maps, a set of devices with risk exceeding the threshold is identified. A multi-objective optimization model constrained by maintenance resources is constructed, a load allocation scheme is calculated, and iterative correction is performed based on feedback from equipment state changes. Control commands are generated, including: The fault risk value of each device node is extracted from the time series prediction map, and the device nodes whose fault risk value exceeds the preset risk threshold are screened out to form a set of devices with risk exceeding the threshold. Extract the fault propagation path and energy consumption transmission relationship of each device in the set of devices with risk exceeding the threshold. Calculate the fault impact weight between devices based on the fault propagation path and the energy consumption coupling coefficient between devices based on the energy consumption transmission relationship. Use the fault impact weight and energy consumption coupling coefficient as device association parameters. Combine the total maintenance resources and load transfer capacity to construct a multi-objective optimization model that includes the objectives of minimizing fault risk and minimizing energy consumption. Input the load values ​​of each device in the set of devices with risk exceeding the threshold into the multi-objective optimization model to calculate the initial load allocation scheme; The initial load allocation scheme is distributed to the set of devices with risk exceeding the threshold. The changes in fault risk value and energy consumption after execution are collected. The fault impact weight and energy consumption coupling coefficient are updated. The process is repeated until the changes in fault risk value and energy consumption meet the convergence condition. The final load allocation scheme is then output, and control commands are generated based on the final load allocation scheme.

[0011] Control commands are divided into multi-level execution queues according to risk priority. Load adjustment commands and maintenance strategies are issued in a tiered manner to the set of devices with risks exceeding the threshold. The collaborative characteristics of the device group are corrected by the device operating status, and the maintenance execution sequence is dynamically adjusted, including: Extract the fault risk value and maintenance time window parameters of each device contained in the control instructions, calculate the risk urgency value of each device, sort the control instructions according to the risk urgency value, and divide them into multi-level execution queues; Based on the risk urgency value range corresponding to each execution level in the multi-level execution queue, the devices in the risk exceeding the threshold device set are assigned to different execution levels, and corresponding load adjustment instructions and maintenance strategies are issued to the devices at different execution levels respectively; Collect equipment operation status data of each device in the risk-exceeding device set after executing load adjustment commands and maintenance strategies, calculate the change in fault risk value and the change in energy consumption transfer relationship between devices before and after execution, and update the collaborative characteristics of the device group based on the change in fault risk value and the change in energy consumption transfer relationship. Extract the fault propagation path change information of each device in the updated device group collaboration features, recalculate the risk urgency value of each device, adjust the device execution order in the multi-level execution queue based on the recalculated risk urgency value, and generate the adjusted maintenance execution order.

[0012] A second aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0013] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0014] In this embodiment, wavelet transform decomposition is used to extract operational trend features and abnormal fluctuation features. Combined with the physical connection sequence of equipment, a two-layer temporal feature representation is constructed. This representation includes both the temporal characteristics of individual equipment and the physical relationships between equipment, improving the expressive power of the features. The relationship between fault propagation and energy consumption impact among equipment is quantitatively described, enabling the system to grasp the collaborative operation characteristics of the equipment group from a holistic perspective. The propagation path and impact range of equipment faults are predicted, achieving accurate early warning of potential faults. Under the constraint of limited maintenance resources, optimal maintenance sequencing and load allocation are achieved, improving maintenance efficiency and resource utilization. By issuing control commands through a hierarchical execution queue and dynamically adjusting maintenance strategies based on equipment operating status feedback, a closed-loop control mechanism is formed, reducing the overall energy consumption of the equipment group, achieving energy conservation and emission reduction, and improving overall equipment efficiency. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the intelligent operation and maintenance and energy consumption optimization method for industrial equipment clusters based on edge computing, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the analysis of equipment failure risk and energy consumption evolution in an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0018] Figure 1 This is a flowchart illustrating the intelligent operation and maintenance and energy consumption optimization method for industrial equipment clusters based on edge computing, as described in an embodiment of the present invention. Figure 1 As shown, the method includes: By collecting operational status data and energy consumption data of industrial equipment groups through distributed edge nodes, and using a multi-point timestamp calibration algorithm for time alignment processing, equipment-level time series data is obtained. Wavelet transform decomposition is performed on device-level time series data to extract operational trend features and abnormal fluctuation features, and a two-layer time series feature representation is constructed by combining the physical connection sequence of the devices. Based on the dual-layer time-series feature representation, the strength of causal influence and transmission delay between devices are calculated, and an operation and maintenance coupling relationship matrix and an energy consumption transmission relationship matrix are constructed to generate device group collaborative features. Based on the collaborative characteristics of the equipment group, a fault propagation tree is constructed and a multi-path propagation simulation is performed to calculate the fault risk value and energy consumption evolution trend of each equipment node and generate a time series prediction map. Based on the time-series prediction map, a set of risk-exceeding equipment is identified, a multi-objective optimization model with maintenance resource constraints is constructed, a load allocation scheme is calculated and iteratively corrected through equipment state change feedback, and control commands are generated. Control commands are divided into multi-level execution queues according to risk priority. Load adjustment commands and maintenance strategies are issued to the set of devices with risks exceeding the threshold in a hierarchical manner. The collaborative characteristics of the device group are corrected by the device operating status, and the maintenance execution order is dynamically adjusted.

[0019] In one optional implementation, operational status data and energy consumption data of industrial equipment clusters are collected through distributed edge nodes, and time alignment processing is performed using a multi-point timestamp calibration algorithm to obtain equipment-level time-series data, including: The industrial equipment group is divided into multiple acquisition areas based on communication distance and equipment density. Master-slave distributed edge nodes are set up in the acquisition areas to build a hierarchical edge node communication network and generate a multi-layer acquisition network topology. Based on the analysis of the multi-layer acquisition network topology, the operating load of the equipment and the network communication load are analyzed to calculate the optimal sampling period and buffer capacity, and an adaptive sampling scheme is generated. Based on the adaptive sampling scheme, the operating status data and energy consumption data of industrial equipment groups are collected through distributed edge nodes, and a data collection timestamp and transmission timestamp mapping table is established to form the raw sampling data. A distributed clock synchronization mechanism is constructed, which uses edge nodes to calculate transmission path delay and node clock drift, generates timestamp correction compensation values, and performs distributed time alignment on the original sampled data. Multi-dimensional feature extraction and data consistency verification are performed on the time-aligned data. Based on the verification results, the data is reconstructed to obtain device-level time-series data.

[0020] To achieve efficient data acquisition and time alignment of operational status and energy consumption data for industrial equipment clusters, this implementation proposes a data acquisition and time alignment processing method based on distributed edge nodes. First, the industrial equipment cluster is divided into multiple acquisition zones based on communication distance and equipment density. In an industrial workshop, equipment can be divided into three acquisition zones according to physical location and communication requirements: a core production zone, an auxiliary production zone, and a management zone. The core production zone has high equipment density and typically includes key equipment such as CNC machine tools and robots, with a communication distance set within 50 meters. The auxiliary production zone includes material handling equipment and testing equipment, with a communication distance of 50-100 meters. The management zone includes monitoring terminals and data servers, with a communication distance of up to 300 meters.

[0021] Within each region, corresponding master-slave distributed edge nodes are deployed based on the number of devices and communication requirements. In the core region, one master node and three slave nodes are configured for every 10 devices; in the auxiliary region, one master node and two slave nodes are configured for every 20 devices; and in the management region, one master node and one slave node are configured for every 50 devices. Through wireless communication links between master nodes and wired connections between slave nodes and devices, a tree-like hierarchical edge node communication network is formed, generating a multi-layered data acquisition network topology.

[0022] This study analyzes the operational load of devices and the network communication load based on a multi-layered acquisition network topology. By collecting one week's worth of historical data, the average and peak data generation rates for each device are calculated. Combined with the processing power and communication bandwidth of each edge node, the overall network communication load is assessed. For high-load areas, such as critical equipment in the core production area, a dynamic sampling strategy is adopted: when equipment operation is stable, the sampling period can be appropriately extended to 500 milliseconds; when a change in equipment status is detected to exceed a preset threshold (e.g., temperature change rate greater than 5℃ / minute), the sampling period is automatically shortened to 100 milliseconds to ensure the capture of transient changes. Simultaneously, based on node storage capacity and network bandwidth, an optimal caching strategy is calculated: the master node cache capacity is set to 3 times the average data generation rate, and the slave node cache capacity is set to 1.5 times, to cope with network congestion. Considering these factors, an adaptive sampling scheme incorporating different device types and time periods is generated to achieve reasonable resource allocation and utilization.

[0023] Based on an adaptive sampling scheme, operational status and energy consumption data of industrial equipment clusters are collected through distributed edge nodes. For devices supporting standard industrial communication protocols (such as Modbus and Profinet), operating parameters are directly read through a protocol conversion module. For older devices that do not support standard protocols, auxiliary data is collected through additional sensors (such as current and vibration sensors) to deduce the device's operational status. During data acquisition, three types of timestamps are recorded: data generation timestamp (device end), data acquisition timestamp (edge ​​node end), and data transmission timestamp (gateway end). A mapping table is established between these three, assigning a complete spatiotemporal identifier to each data entry. For high-frequency data, a local preprocessing mechanism is used at the edge nodes to calculate statistical characteristics such as mean and variance through a sliding window, reducing data transmission volume and forming structured raw sampled data.

[0024] A distributed clock synchronization mechanism is constructed to achieve data time alignment. A precise time protocol is adopted, with each edge node periodically synchronizing with its upper-layer nodes, and the top-level master node synchronizing with the network time server. During synchronization, the transmission path delay is measured through multiple bidirectional communications, calculated using the formula: Path Delay = (t4 - t1) - (t3 - t2) / 2, where t1 is the time of sending the synchronization message, t2 is the time of receiving the synchronization message, t3 is the time of sending the response message, and t4 is the time of receiving the response message. Simultaneously, the node clock drift rate is calculated based on the time difference of multiple consecutive synchronization processes. For nodes whose clock drift rate exceeds a threshold (e.g., drift greater than 1 millisecond per hour), their synchronization frequency is increased. Based on the calculated path delay and clock drift, a timestamp correction compensation value is generated, and distributed time alignment is performed on the original sampled data, ensuring the comparability of data collected by different devices and nodes in the time dimension.

[0025] Multi-dimensional feature extraction and data consistency verification are performed on the time-aligned data. Feature extraction includes time-domain features (mean, standard deviation, peak value, etc.), frequency-domain features (extracting the dominant frequency component through Fast Fourier Transform), and statistical features (data distribution, outlier detection). Data consistency verification is achieved through three mechanisms: physical model verification, checking whether the data conforms to the physical characteristics of the equipment (e.g., the relationship between motor power and current / voltage); time-series logic verification, checking whether the order of events is reasonable (e.g., the start command must precede the running state); and redundant data cross-verification, using multi-sensor data to corroborate each other (e.g., temperature rise and power increase should occur synchronously). For data that fails verification, different reconstruction strategies are adopted according to the type of verification failure: for deviations caused by noise interference, wavelet transform is used for denoising; for missing values, interpolation is performed based on historical data trends; for obviously erroneous values, they are marked and not included in subsequent analysis. After verification and reconstruction, high-quality equipment-level time-series data is formed, providing a reliable data foundation for subsequent equipment status monitoring, energy efficiency analysis, and fault prediction.

[0026] In one optional implementation, wavelet transform decomposition is performed on the device-level time series data to extract operational trend features and abnormal fluctuation features. A two-layer time series feature representation is constructed by combining the device's physical connection sequence, including: A fluctuation analysis matrix is ​​constructed for equipment-level time series data, the fluctuation trend is calculated, and wavelet transform decomposition is performed on equipment-level time series data at different time scales based on the fluctuation trend to generate multi-scale wavelet decomposition coefficients. The trend and fluctuation components of equipment-level time series data are reconstructed using multi-scale wavelet decomposition coefficients. The phase relationship and amplitude ratio between the components are extracted to construct the operation trend characteristics. A feature analysis matrix is ​​constructed for the wave components to extract wave energy distribution and wave morphology features. Abnormal wave characteristics and wave propagation paths are calculated through wave feature analysis. Construct the device connection structure according to the physical connection order of the devices, mark the physical connection relationships of the devices, establish the device transmission path, and generate the device connection matrix; By mapping operational trend characteristics and abnormal fluctuation characteristics to the device connection matrix, and combining the spatial correlation between fluctuation propagation path and device transmission path, a two-layer temporal feature representation containing temporal and spatial characteristics is constructed.

[0027] The device-level time-series data is first preprocessed, including denoising and standardization, to ensure data quality. A fluctuation analysis matrix is ​​then constructed from the processed time-series data. This matrix consists of sliding windows of time-series data, with each row corresponding to a data segment within a time window. The fluctuation trend is obtained by calculating the difference indices between adjacent windows. These difference indices can be Euclidean distance, correlation coefficient, or information entropy change, etc. Based on the obtained fluctuation trend, the number of wavelet transform decomposition levels and wavelet basis functions are determined. Commonly used wavelet basis functions include Daubechies, Symlet, or Coiflet; an appropriate basis function is selected based on the characteristics of the device time-series data.

[0028] Multi-level discrete wavelet transform is applied to the time-series data of each device to obtain approximation coefficients (A) and detail coefficients (D) at different scales. For example, for a 3-level decomposition, one approximation coefficient A3 and three detail coefficients D1, D2, and D3 are obtained, where the approximation coefficient represents the low-frequency trend component of the data, and the detail coefficients represent the high-frequency fluctuation component. These coefficients form a set of multi-scale wavelet decomposition coefficients.

[0029] Using the obtained multi-scale wavelet decomposition coefficients, trend components and fluctuation components are reconstructed separately through inverse wavelet transform. The trend component is mainly reconstructed from approximation coefficients, reflecting the long-term trend of equipment operation; the fluctuation component is reconstructed from detail coefficients, reflecting the short-term fluctuation characteristics of equipment operation. The phase difference between the trend component and the original data, as well as the amplitude ratio of the fluctuation component to the original data, are calculated to construct an operating trend feature vector. This feature vector includes phase relationship parameters and amplitude ratio parameters, which can reflect the stability and variation law of equipment operating status.

[0030] For the extracted fluctuation components, a feature analysis matrix is ​​constructed, where each row represents the fluctuation component data within a time window. Fluctuation energy distribution features, including the energy proportion of different frequency bands and energy entropy, are extracted from the feature analysis matrix. Simultaneously, fluctuation morphology features, such as peak value distribution, periodicity, and waveform steepness, are extracted. Combining the fluctuation energy and morphology features, abnormal fluctuation characteristics are identified by setting thresholds or applying clustering algorithms. By analyzing the time lag relationship of fluctuation components between adjacent devices, the fluctuation propagation path is determined; this path represents the order in which the abnormal state is transmitted between devices.

[0031] Construct a device connection structure diagram based on the physical connection sequence of equipment in a real industrial scenario. Mark the physical connection relationships between devices in the diagram, such as series, parallel, or ring connections. Based on these physical connection relationships, determine the transmission paths for material flow, energy flow, or information flow between devices. Transform the device connection relationships into an adjacency matrix, generating a device connection matrix where matrix elements represent the connection status and connection strength between devices.

[0032] The operational trend features and abnormal fluctuation features extracted in the previous steps are mapped to the nodes of the corresponding devices in the device connection matrix. The consistency between the fluctuation propagation path and the physical transmission path of the devices is compared to identify abnormal propagation paths. Based on the trend features in the time dimension and the connection relationships in the spatial dimension, a two-layer temporal feature representation model is constructed. The first layer of this model contains the operational trend features of each device, and the second layer contains the abnormal fluctuation propagation features between devices, forming a complete spatiotemporal feature representation.

[0033] In practical applications, taking a chemical production line as an example, the equipment includes reactors, heat exchangers, and separation towers. Time-series data such as temperature, pressure, and flow rate of each device are collected, and wavelet decomposition is performed on this data to extract trend and fluctuation characteristics. Analysis reveals that when a slight temperature fluctuation occurs in the reactor, this fluctuation is transmitted to the heat exchanger along the material flow direction, and the fluctuation characteristics change during this transmission. Using the constructed two-layer time-series feature representation, not only can abnormal fluctuations in the heat exchanger be detected, but the source of the anomaly—the temperature control problem in the reactor—can also be traced, thus achieving accurate fault diagnosis.

[0034] This method can also be applied to power plant equipment chains, such as systems composed of boilers, steam turbines, and generators. By analyzing the trend and fluctuation characteristics of the parameters of each piece of equipment, and combining this with the physical connections between the equipment, the location and timing of potential failures can be predicted, providing decision support for equipment maintenance.

[0035] The above-mentioned two-layer time series feature representation method can not only analyze the operating status of individual devices, but also capture the fault propagation law between devices, providing a comprehensive analytical perspective for the condition monitoring and fault diagnosis of complex industrial systems, and improving the accuracy of fault prediction and the precision of fault location.

[0036] In one optional implementation, the strength of causal influence and transmission delay between computing devices are calculated based on a two-layer time-series feature representation, and an operation and maintenance coupling relationship matrix and an energy consumption transmission relationship matrix are constructed to generate device group collaborative features, including: Multi-scale state decomposition is performed on the two-layer time series feature representation. State change features are extracted through fluctuation feature analysis and amplitude reconstruction. The state transition law is calculated by combining the change features to obtain the intensity and depth of causal influence between devices and form a causal feature sequence. Based on the causal characteristic sequence analysis of the state change process, combined with the state transition law, the equipment state response time and response amplitude are calculated, the transmission delay and transmission link between equipment are determined, and the transmission characteristic matrix is ​​constructed. The transmission feature matrix is ​​mapped and matched with the device connection order to analyze the direct impact transmission caused by state changes between adjacent devices, as well as the interval impact transmission caused by intermediate devices. The coupling degree between devices is calculated, and the operation and maintenance coupling relationship matrix is ​​constructed by combining the causal impact strength and impact depth. Based on the operation and maintenance coupling relationship matrix, energy transfer paths are identified, energy transfer efficiency is calculated by combining transfer delay and transfer link, energy transfer direction and transfer gain are determined by coupling degree analysis, and energy consumption transfer relationship matrix is ​​generated. The operation and maintenance coupling relationship matrix and the energy consumption transfer relationship matrix are fused to generate the collaborative features of the equipment group.

[0037] In collaborative management of equipment groups, the first step is to preprocess the equipment data, including data cleaning and normalization. Multi-dimensional time-series data such as temperature, pressure, and energy consumption are collected, outliers are eliminated, and standardization is performed to ensure the comparability of data across different dimensions.

[0038] When performing multi-scale state decomposition on a two-level time series feature representation, wavelet transform is used to decompose the time series data into sub-signals of different frequency bands to distinguish between rapidly changing and slowly changing equipment state features. During fluctuation feature analysis, the energy distribution, peak distribution, and waveform characteristics of each sub-signal are calculated, and abnormal fluctuation points are identified using threshold judgments. Amplitude reconstruction involves weighting and fusing the features of each frequency band according to their energy contribution rate to reconstruct the equipment state change feature sequence. For a certain cooling tower, after multi-scale decomposition, the temperature sensor data can be identified as having a correlation between periodic temperature fluctuations in the 0.1Hz frequency band and normal equipment operation, while irregular fluctuations in the 2Hz frequency band may indicate an early sign of bearing failure.

[0039] The state transition law calculation is based on Markov process modeling. A state transition matrix is ​​constructed by analyzing the probability distribution of device state changes within adjacent time windows. The elements in this matrix represent the probability of a device transitioning from one state to another, thus quantifying the state transition law. The strength of causal influence between devices is calculated using the Granger causality test to analyze the degree of impact of a device's state change on other devices. The depth of influence is determined by recursively calculating the causal chain length, i.e., how many levels of indirect transmission a device's state change can indirectly affect other devices. For a system consisting of a fan and a heat exchanger, when the fan speed decreases from 1200 rpm to 900 rpm, the heat exchanger temperature increases by 2.3 degrees Celsius after 67 seconds, indicating a clear causal relationship between the two devices. The causal influence strength is 0.78, and the influence depth is 1.

[0040] Analyzing state change processes based on causal characteristic sequences requires detecting state change points and calculating the magnitude and rate of change. Equipment state response time refers to the time interval from a state change in one piece of equipment to a response in another, calculated through cross-correlation analysis. Response magnitude is the ratio of the magnitude of the state change in the responding equipment to the magnitude of the change in the triggering equipment. The propagation delay matrix records the state propagation delay time between all equipment pairs, while the propagation link describes the path of state change propagation within the equipment group. In the case of a compressor and refrigeration system, after the compressor starts, the refrigeration pipe pressure rises by 0.4 MPa within 3.2 seconds, followed by a 5.1-degree drop in evaporator temperature within 15.7 seconds. This continuous response constitutes a propagation link, and the corresponding propagation delay is recorded in the propagation characteristic matrix.

[0041] When mapping the transmission feature matrix to the device connection sequence, a device connection diagram is first established based on the physical topology. Then, the latency information in the transmission feature matrix is ​​analyzed in correspondence with the physical connections. Direct impact transmission between adjacent devices manifests as the correlation of state changes between physically connected devices; intermittent impact transmission is the indirect impact generated through intermediate devices. The coupling degree between devices is calculated by comprehensively evaluating three factors: causal impact strength, impact depth, and transmission latency, forming a coupling metric. Organizing the coupling metrics between each pair of devices into a matrix form constitutes the operation and maintenance coupling relationship matrix. For the pump station system, the coupling degree between the main pump and the standby pump is 0.92, indicating a high degree of correlation; while the coupling degree between the main pump and the remote monitoring equipment is only 0.31, indicating a low degree of correlation.

[0042] When identifying energy transfer paths based on the operation and maintenance coupling relationship matrix, a threshold is first set according to the coupling strength to filter out the main energy flow channels. Energy transfer efficiency is calculated by combining transfer delay and energy loss coefficient; the shorter the delay and the smaller the loss, the higher the transfer efficiency. The energy transfer direction is determined based on the sequence of equipment state changes, and the transfer gain is the ratio of the energy consumption change of downstream equipment to that of upstream equipment. These features are organized into a matrix to generate the energy transfer relationship matrix. In a certain boiler system, the burner energy transfer efficiency is 0.86, mainly transferred along the path of burner-boiler body-heat exchanger-circulating pump, with transfer gains of 1.2, 0.75, and 0.92 for each segment.

[0043] Finally, the operation and maintenance coupling relationship matrix and the energy consumption transfer relationship matrix are fused using feature fusion. Tensor decomposition is employed to extract shared and complementary features from the two matrices, forming a unified feature representation. After nonlinear transformation and normalization, the fused features generate the final equipment group collaborative features, which are used for subsequent equipment group collaborative operation and maintenance decisions and energy consumption optimization.

[0044] This implementation method achieves accurate characterization of the operating status of industrial equipment groups by constructing a two-layer temporal feature representation; it reveals the complex interaction relationships between equipment based on multi-scale state decomposition and causal analysis; and it effectively extracts the collaborative features of the equipment group by constructing an operation and maintenance coupling relationship matrix and an energy consumption transfer relationship matrix. This method significantly improves the synergy and accuracy of intelligent operation and maintenance of industrial equipment groups, making operation and maintenance decisions more consistent with the actual correlation mechanisms between equipment; at the same time, it provides a data foundation for energy consumption optimization by identifying energy transfer paths and calculating energy transfer efficiency.

[0045] like Figure 2 As shown, a flowchart illustrating the equipment failure risk and energy consumption evolution analysis of this embodiment is presented.

[0046] In one optional implementation, a fault propagation tree is constructed based on the collaborative characteristics of the equipment group, and a multi-path propagation simulation is performed to calculate the fault risk value and energy consumption evolution trend of each equipment node, generating a time-series prediction map, including: The operation feature sequence and energy consumption feature sequence are extracted from the collaborative features of the equipment group. The fluctuation cycle analysis is performed on the operation feature sequence to calculate the state change frequency vector, and the amplitude decomposition is performed on the energy consumption feature sequence to obtain the state change amplitude vector. The state evolution matrix is ​​constructed by fusing the state change frequency vector and the state change amplitude vector, and a fault propagation tree is generated based on the state evolution matrix. The location of the propagation point is determined based on the fault propagation tree. The propagation speed value is calculated using the state change frequency in the state evolution matrix. The propagation attenuation coefficient is obtained through the state change amplitude. The propagation speed value and the propagation attenuation coefficient are combined to construct the propagation range matrix. The propagation channel matrix is ​​formed using the propagation range matrix. The propagation range matrix and the propagation channel matrix are combined to generate a state transition sequence. The state transition sequence is used to calculate the transition probability. The fluctuation characteristics are constructed by combining the state change sequence. The equipment node failure risk value is calculated by using the transition probability and the fluctuation characteristics. The equipment node failure risk value is mapped to the state evolution matrix, and the energy consumption evolution trend is calculated by combining the state change frequency vector and the state change amplitude vector. The equipment node failure risk value and energy consumption evolution trend are arranged in the time dimension to generate a time series prediction map.

[0047] To construct a fault propagation tree and perform multipath propagation simulation based on the collaborative characteristics of the equipment group, it is first necessary to acquire the operational data of the equipment group. The collaborative characteristics of the equipment group are a multi-dimensional feature matrix containing the operating status, energy consumption level, and interrelationships of each device in the group. Using a feature separation algorithm, this matrix is ​​split into an operational feature sequence describing the operating status of the devices and an energy consumption feature sequence describing energy consumption changes. The operational feature sequence typically includes time-varying data on parameters such as temperature, vibration, and pressure, while the energy consumption feature sequence includes parameters related to energy consumption, such as current, power, and flow rate. In a production line case, the original collaborative characteristics of the equipment group collected by the edge computing unit contained 30 dimensions, which, after feature separation, resulted in an 18-dimensional operational feature sequence and a 12-dimensional energy consumption feature sequence.

[0048] When performing fluctuation cycle analysis to calculate the state change frequency vector of the operating characteristic sequence, the Fast Fourier Transform (FFT) technique is used to convert the time-domain signal into a frequency-domain signal, extracting the amplitude and phase information of each frequency component. By analyzing the distribution characteristics of the main frequency components, the main cycle of state changes for each device is determined. The state change frequency vector records the frequency characteristics of the state changes for each device, reflecting the dynamic characteristics of the device operation. Amplitude decomposition is performed on the energy consumption characteristic sequence. Wavelet transform is used to decompose the energy consumption signal into detail coefficients and approximation coefficients at different scales. By analyzing the amplitude distribution of the coefficients at each scale, the state change amplitude vector is obtained. This vector records the amplitude characteristics of the energy consumption changes of the device, characterizing the intensity of energy consumption fluctuations. For a certain cooling equipment group, the operating characteristic sequence analysis shows that the main state change cycle is 3 hours, corresponding to a state change frequency vector of [0.33, 0.12, 0.08] times / hour; the state change amplitude vector of the energy consumption characteristic sequence is [12.5, 8.3, 4.7] kilowatt-hours, representing the magnitude of energy consumption changes between different states.

[0049] Fusing the state change frequency vector and the state change amplitude vector to construct the state evolution matrix is ​​a crucial step in modeling the state evolution process of equipment. The fusion process employs tensor product operations, cross-combining the features of each dimension of the two vectors to form a two-dimensional matrix representing the comprehensive characteristics of the state. Each element in this matrix represents the state transition characteristics of the equipment under specific frequency and amplitude conditions. When generating a fault propagation tree based on the state evolution matrix, a hierarchical clustering algorithm is used to cluster state nodes with similar characteristics in the state evolution matrix, forming a tree structure. The root node of the tree represents the initial fault point, branch nodes represent possible fault propagation paths, and leaf nodes represent the endpoints of the fault's impact. In a refrigeration system case, a group of 10 devices has a state evolution matrix of dimension 10×15. The fault propagation tree generated after hierarchical clustering contains 1 root node, 6 branch nodes, and 15 leaf nodes, clearly showing the potential fault propagation paths.

[0050] When determining the location of propagation points based on a fault propagation tree, the tree's topology is analyzed to identify the connectivity and hierarchical position of nodes. Nodes with high connectivity are typically key propagation points. The propagation velocity is calculated using the frequency of state changes in the state evolution matrix; the higher the frequency, the faster the fault propagation. The propagation velocity can be understood as the probability that a fault's impact propagates from one device to an adjacent device per unit time. The propagation attenuation coefficient is obtained through the amplitude of state changes; the larger the amplitude, the slower the propagation attenuation and the wider the fault's impact range. The propagation velocity and propagation attenuation coefficient are combined to construct a propagation range matrix, which describes the impact range of a fault propagating from each possible starting point to each possible ending point. The propagation range matrix is ​​used to form a propagation channel matrix, where each element represents the channel characteristics of fault propagation between two device nodes, including channel capacity and channel state. In a pumping station equipment group, the propagation velocity of the main water pump is 0.85, and the propagation attenuation coefficient is 0.12, indicating that the fault propagates quickly from the main water pump and has a wide impact range. The constructed propagation range matrix shows that a fault in the main water pump can affect 90% of the related equipment within 15 minutes.

[0051] To generate a state transition sequence by combining the propagation range matrix and the propagation channel matrix, the fault propagation process needs to be simulated chronologically along the path of the fault propagation tree, recording the changes in the state of each node. The transition probability is calculated using the state transition sequence. A sliding window method is employed, statistically analyzing historical state transition data to calculate the conditional probability of a state transitioning from one mode to another. Fluctuation characteristics are constructed based on the state change sequence, including fluctuation frequency, fluctuation amplitude, and fluctuation trend. The fault risk value of each equipment node is calculated using the transition probability and fluctuation characteristics. This risk value is a weighted combination of the transition probability and fluctuation characteristics, reflecting the likelihood of equipment failure. In a production line example, edge computing analysis shows that for critical motor equipment, the state transition probability matrix indicates a transition probability of 0.23 from "normal operation" to "minor anomaly," a fluctuation characteristic value of 0.68, and a comprehensive fault risk value of 0.57, indicating a moderate level of fault risk for this equipment.

[0052] The fault risk values ​​of equipment nodes are mapped to a state evolution matrix, and the state transition characteristics are corrected by using the risk values ​​as weighting factors in the state evolution matrix. The energy consumption evolution trend is calculated by combining the state change frequency vector and the state change amplitude vector. An autoregressive moving average model is used to predict the energy consumption trend of equipment in the future. The fault risk values ​​of equipment nodes and the energy consumption evolution trend are arranged along the time dimension to generate a time-series prediction map. This map is a three-dimensional data structure, with the horizontal axis representing the time dimension, the vertical axis representing equipment nodes, and the depth axis representing the fault risk value and energy consumption trend value. For a certain industrial cooling system, the time-series prediction map generated by the edge computing unit shows that in the next 72 hours, the fault risk value of the cooling tower will gradually increase from 0.32 to 0.78, while the energy consumption trend is expected to increase by 25%, providing maintenance personnel with clear early warning information and directions for energy consumption optimization.

[0053] This implementation constructs a fault propagation tree based on the collaborative characteristics of equipment groups, achieving accurate characterization and prediction of fault propagation patterns in industrial equipment groups. This method effectively identifies complex influence relationships between equipment, accurately simulates the propagation path and speed of faults within the equipment group, and precisely calculates the fault risk value and energy consumption evolution trend of each equipment node. Compared to traditional methods, this scheme not only considers the state characteristics of individual equipment but also comprehensively analyzes the interactive influences between equipment, making fault early warning more accurate and energy consumption prediction more reasonable. Through the application of edge computing technology, the entire analysis process can be completed rapidly on-site, significantly improving decision-making response speed.

[0054] In one optional implementation, a set of devices with risk exceeding the threshold is identified based on a time-series prediction map, a multi-objective optimization model constrained by maintenance resources is constructed, a load allocation scheme is calculated and iteratively corrected through feedback from equipment state changes, and control commands are generated, including: The fault risk value of each device node is extracted from the time series prediction map, and the device nodes whose fault risk value exceeds the preset risk threshold are screened out to form a set of devices with risk exceeding the threshold. Extract the fault propagation path and energy consumption transmission relationship of each device in the set of devices with risk exceeding the threshold. Calculate the fault impact weight between devices based on the fault propagation path and the energy consumption coupling coefficient between devices based on the energy consumption transmission relationship. Use the fault impact weight and energy consumption coupling coefficient as device association parameters. Combine the total maintenance resources and load transfer capacity to construct a multi-objective optimization model that includes the objectives of minimizing fault risk and minimizing energy consumption. Input the load values ​​of each device in the set of devices with risk exceeding the threshold into the multi-objective optimization model to calculate the initial load allocation scheme; The initial load allocation scheme is distributed to the set of devices with risk exceeding the threshold. The changes in fault risk value and energy consumption after execution are collected. The fault impact weight and energy consumption coupling coefficient are updated. The process is repeated until the changes in fault risk value and energy consumption meet the convergence condition. The final load allocation scheme is then output, and control commands are generated based on the final load allocation scheme.

[0055] First, the fault risk value of each equipment node is extracted from the time-series prediction graph. The time-series prediction graph is a multi-dimensional data structure that contains the fault risk value and energy consumption evolution trend of each equipment in the equipment group over time. By traversing the time-series prediction graph, the fault risk value of each equipment node within the current and future prediction time windows is extracted. The risk value is usually a value between 0 and 1, representing the probability of equipment failure. When filtering fault risk values, a preset risk threshold is set according to the safety requirements of the industrial site. This threshold can be determined based on the equipment type, production importance, and historical fault data. Generally, the threshold for critical equipment is set lower. The edge computing unit compares the extracted risk values ​​and filters out the equipment that exceeds the preset threshold, forming a set of equipment with risk exceeding the threshold. In a production line case, the edge computing unit extracted the fault risk values ​​of 12 equipment from the time-series prediction graph. The preset risk threshold is 0.65. It was found that the risk values ​​of the main motor, conveyor belt, and pressure valve are 0.78, 0.72, and 0.67, respectively, exceeding the preset threshold. Therefore, these three equipment constitute the set of equipment with risk exceeding the threshold.

[0056] Extracting the fault propagation paths and energy consumption transfer relationships for each device in the set of devices with risk exceeding the threshold is the foundation for constructing the optimization model. Fault propagation paths are extracted from the previously generated fault propagation tree, describing how a fault propagates from one device to others. Energy consumption transfer relationships are obtained from the energy consumption transfer relationship matrix, reflecting how changes in the energy consumption of one device affect the energy consumption levels of other devices. Based on the fault propagation paths, the fault impact weights between devices are calculated. Path analysis is used, considering path length, propagation probability, and impact degree, to quantify the intensity of the impact of a fault on other devices. Based on the energy consumption transfer relationships, the energy consumption coupling coefficients between devices are calculated. By analyzing the correlation and transfer efficiency of energy consumption data, the degree of correlation between energy consumption changes between devices is determined. The fault impact weights and energy consumption coupling coefficients are used as device correlation parameters, and a multi-objective optimization model is constructed by combining the total maintenance resources and load transfer capacity. The total maintenance resources refer to the sum of human, material, and time resources available for equipment maintenance; the load transfer capacity refers to the workload that can be transferred from one device to other devices without affecting normal production. The model includes two optimization objectives: minimizing fault risk and minimizing energy consumption.

[0057] The core step in the optimization process is to input the load values ​​of each device in the set of devices with risk exceeding the threshold into the multi-objective optimization model to calculate the initial load allocation scheme. The load value refers to the current workload or operating intensity of the device, usually expressed as a percentage of the device's rated load. The multi-objective optimization model uses a genetic algorithm to solve the problem. First, multiple sets of candidate load allocation schemes are generated, and then the fitness function is used to evaluate the merits of each scheme. The fitness function comprehensively considers two optimization objectives: the overall failure risk of the device group after the scheme is implemented, and the total energy consumption level of the device group after the scheme is implemented. During the fitness evaluation process, Pareto sorting is used to determine the set of non-dominated solutions, and a non-dominated sorting strategy is used to select the optimal solution as the initial load allocation scheme.

[0058] The initial load allocation scheme is distributed to the set of devices with risks exceeding the threshold. Collecting the changes in fault risk values ​​and energy consumption after execution is a crucial step in verifying the scheme's effectiveness. The load allocation scheme is converted into specific control parameters, such as device start / stop status, operating frequency, and pressure settings, by the edge computing unit and then distributed to the relevant devices for execution. After a certain execution time, the edge computing unit calculates the changes in fault risk values ​​and energy consumption from the real-time operating data of the devices. These two changes represent the degree of change in device fault risk and energy consumption relative to before the load adjustment. Based on these actual changes, the fault impact weight and energy consumption coupling coefficient are updated using a recursive least squares method. The correlation parameters in the model are corrected based on the observed changes, making the model more accurately reflect the actual situation. This iterative process is repeated, with the load allocation scheme recalculated and executed each time based on the updated parameters, until the changes in fault risk values ​​and energy consumption meet the preset convergence conditions. The convergence condition is typically set as the difference between the changes in two consecutive iterations being less than a certain threshold. In the aforementioned case, after the first iteration, it was found that the failure risk of the main motor actually decreased by 0.15, less than the expected 0.20, while energy consumption decreased by 6%, higher than the expected 5%. Based on this, the system adjusted the failure impact weight of the main motor and other equipment from 0.85 to 0.78, and the energy consumption coupling coefficient from 0.73 to 0.79. After three iterations, the changes met the convergence condition, and the final load distribution scheme was output.

[0059] Based on the final load distribution scheme, control commands are generated. These commands are a set of specific parameters and action instructions for equipment operation, including adjustments to equipment operating status, changes to parameter settings, and maintenance schedules. During command generation, the edge computing unit translates the load distribution scheme into a control language recognizable by each device, considering execution details such as equipment start-up / shutdown sequence and transition time to ensure that command execution does not cause production interruptions or safety hazards. In the final production line scheme, the edge computing unit generates a series of control commands, including "main motor frequency reduced to 35Hz," "start standby motor and set frequency to 25Hz," and "conveyor belt speed reduced to 70%," and also develops a maintenance plan for the main motor, scheduling maintenance during the next production off-peak period.

[0060] This implementation achieves intelligent operation and maintenance and energy consumption optimization for industrial equipment clusters through risk identification and multi-objective optimization based on time-series prediction maps. This method can accurately identify high-risk equipment, comprehensively consider the impact of fault propagation and energy consumption transfer, and find the optimal load allocation scheme to balance fault risk and energy consumption level under the constraint of limited maintenance resources. An iterative correction mechanism based on equipment state change feedback enables the optimization model to continuously adapt and adjust, more accurately reflecting the equipment correlation characteristics in the actual working environment. Compared with traditional methods, this solution not only focuses on the performance optimization of individual equipment but also considers the overall synergistic effect of the equipment cluster, achieving energy-efficient utilization while ensuring production safety.

[0061] In one optional implementation, control commands are divided into multi-level execution queues according to risk priority. Load adjustment commands and maintenance strategies are issued hierarchically to the set of devices with risks exceeding the threshold. The collaborative characteristics of the device group are corrected by the device operating status, and the maintenance execution order is dynamically adjusted, including: Extract the fault risk value and maintenance time window parameters of each device contained in the control instructions, calculate the risk urgency value of each device, sort the control instructions according to the risk urgency value, and divide them into multi-level execution queues; Based on the risk urgency value range corresponding to each execution level in the multi-level execution queue, the devices in the risk exceeding the threshold device set are assigned to different execution levels, and corresponding load adjustment instructions and maintenance strategies are issued to the devices at different execution levels respectively; Collect equipment operation status data of each device in the risk-exceeding device set after executing load adjustment commands and maintenance strategies, calculate the change in fault risk value and the change in energy consumption transfer relationship between devices before and after execution, and update the collaborative characteristics of the device group based on the change in fault risk value and the change in energy consumption transfer relationship. Extract the fault propagation path change information of each device in the updated device group collaboration features, recalculate the risk urgency value of each device, adjust the device execution order in the multi-level execution queue based on the recalculated risk urgency value, and generate the adjusted maintenance execution order.

[0062] First, the fault risk value and maintenance time window parameters for each device are extracted from the control instructions. The control instructions include key parameters such as device identification, fault risk value, suggested execution time, maintenance resource requirements, and expected maintenance duration. When calculating the risk urgency value, the product of the fault risk value and the device's impact level is divided by the maintenance time window length; the result is the risk urgency value. After sorting the control instructions in descending order according to their risk urgency values, they are divided into multi-level execution queues based on predetermined threshold ranges. In a production line case, the edge computing unit extracted the fault risk values ​​for the main motor, conveyor belt, and pressure valve, which were 0.78, 0.72, and 0.67, respectively; the maintenance time window lengths were 8 hours, 12 hours, and 24 hours, respectively; the device impact levels were 0.9, 0.8, and 0.7, respectively; and the calculated risk urgency values ​​were 0.088, 0.048, and 0.020, respectively. Based on the preset risk urgency classification criteria, control commands are divided into three levels of execution queues: high priority (risk urgency value greater than 0.05), medium priority (risk urgency value between 0.03 and 0.05), and low priority (risk urgency value less than 0.03). Accordingly, the main motor is assigned to the high priority queue, the conveyor belt to the medium priority queue, and the pressure valve to the low priority queue.

[0063] Based on the risk urgency range corresponding to each execution level in the multi-level execution queue, allocating devices in the risk-exceeding-threshold device set to different execution levels is a key step in implementing differentiated maintenance strategies. Execution levels typically include emergency handling, planned maintenance, and monitoring / observation levels. Different execution levels correspond to different load adjustment strategies and maintenance resource allocation schemes. When issuing corresponding load adjustment instructions and maintenance strategies to devices at different execution levels, the edge computing unit generates specific execution parameters, including the load adjustment magnitude, execution time point, and maintenance resource allocation details. In the aforementioned example, the main motor, as a high-priority device, receives the load adjustment instruction "Immediately reduce the load to 50%, and arrange maintenance team A for maintenance at 14:00 today"; the conveyor belt, as a medium-priority device, receives the instruction "Reduce the workload to 70%, and arrange maintenance team B for maintenance at 08:00 tomorrow"; and the pressure valve, as a low-priority device, receives the instruction "Maintain the current load, strengthen monitoring, and schedule routine maintenance next week."

[0064] Collecting equipment operating status data after executing load adjustment commands and maintenance strategies from each device in the risk-exceeding-threshold equipment set serves as the basis for evaluating maintenance effectiveness and updating the collaborative characteristics of the equipment group. Equipment operating status data includes various performance parameters, vibration data, temperature data, energy consumption data, etc. Calculating the change in fault risk value before and after execution requires using a pre-trained equipment health status assessment model. The equipment operating status data before and after execution is input into the model to obtain the corresponding fault risk value; the difference is the change. When calculating the change in energy consumption transmission relationships between devices, correlation analysis of energy consumption data is used to compare the degree of correlation between the energy consumption changes of each device before and after execution. Based on these two types of changes, the collaborative characteristics of the equipment group are updated, primarily updating the weight values ​​in the fault propagation path diagram and the energy consumption transmission relationship matrix. In the aforementioned case, after the main motor underwent load adjustment and maintenance, its fault risk value decreased from 0.78 to 0.35, a reduction of 0.43; while the fault risk value of the related conveyor belt decreased from 0.72 to 0.55, a reduction of 0.17; the energy transfer coefficient between the two changed from 0.73 to 0.65, a reduction of 0.08. Based on these changes, the edge computing unit updated the fault impact weight between the main motor and the conveyor belt in the equipment group collaboration feature, adjusting it from 0.85 to 0.75.

[0065] Extracting the fault propagation path change information of each device from the updated equipment group collaboration characteristics and recalculating the risk urgency value of each device is the core step in dynamically adjusting the maintenance execution order. The fault propagation path change information mainly includes changes in path weights and path structure. Based on this change information, the risk urgency value of each device is recalculated using the same method as the initial calculation. During the calculation process, in addition to considering the device's own fault risk value and maintenance time window parameters, it is also necessary to reassess the impact of device failures on the overall production line in conjunction with the updated fault propagation path. Adjusting the device execution order in the multi-level execution queue based on the recalculated risk urgency value may lead to an increase or decrease in the priority of some devices. In the aforementioned case, since the main motor had completed its maintenance, its risk urgency value significantly decreased from 0.088 to 0.012, and it was reassigned to the low-priority queue. The conveyor belt, however, saw its risk urgency value decrease slightly from 0.048 to 0.045 due to a reduced impact weight from the main motor's failure, remaining in the medium-priority queue. The pressure valve, due to other changes in the equipment group, saw its relative importance increase, with its risk urgency value rising from 0.020 to 0.033, and was moved to the medium-priority queue. Accordingly, the adjusted maintenance execution order places the conveyor belt first, the pressure valve second, and the main motor third, with maintenance resources correspondingly redistributed.

[0066] This implementation method achieves precise allocation and intelligent scheduling of maintenance resources for industrial equipment groups by dividing control commands into multi-level execution queues based on risk priority. Based on a risk urgency assessment mechanism, this method comprehensively considers fault risk values, maintenance time windows, and the degree of equipment impact to scientifically classify equipment, ensuring that limited maintenance resources are prioritized for high-risk equipment. The mechanism of issuing load adjustment commands and maintenance strategies in a tiered manner enables differentiated equipment management, avoiding waste and over-maintenance of maintenance resources. By updating the collaborative characteristics of the equipment group through equipment operating status feedback, the system can dynamically perceive structural changes and evolution of relationships within the equipment group, maintaining the accuracy and timeliness of the model. Based on the updated collaborative characteristics, the maintenance execution sequence is dynamically adjusted, enabling the maintenance plan to respond promptly to changes in on-site conditions and avoiding the lag and incompatibility that may result from fixed plans.

[0067] A second aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0068] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0069] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent operation and maintenance and energy consumption optimization of industrial equipment clusters based on edge computing, characterized in that, include: By collecting operational status data and energy consumption data of industrial equipment groups through distributed edge nodes, and using a multi-point timestamp calibration algorithm for time alignment processing, equipment-level time series data is obtained. Wavelet transform decomposition is performed on device-level time series data to extract operational trend features and abnormal fluctuation features, and a two-layer time series feature representation is constructed by combining the physical connection sequence of the devices. Based on the dual-layer time-series feature representation, the strength of causal influence and transmission delay between devices are calculated, and an operation and maintenance coupling relationship matrix and an energy consumption transmission relationship matrix are constructed to generate device group collaborative features. Based on the collaborative characteristics of the equipment group, a fault propagation tree is constructed and a multi-path propagation simulation is performed to calculate the fault risk value and energy consumption evolution trend of each equipment node and generate a time series prediction map. Based on the time-series prediction map, a set of risk-exceeding equipment is identified, a multi-objective optimization model with maintenance resource constraints is constructed, a load allocation scheme is calculated and iteratively corrected through equipment state change feedback, and control commands are generated. Control commands are divided into multi-level execution queues according to risk priority. Load adjustment commands and maintenance strategies are issued to the set of equipment with risks exceeding the threshold in a hierarchical manner. The collaborative characteristics of the equipment group are corrected by the equipment operating status, and the maintenance execution order is dynamically adjusted. Based on the collaborative characteristics of the equipment group, a fault propagation tree is constructed and a multi-path propagation simulation is performed. The fault risk value and energy consumption evolution trend of each equipment node are calculated, and a time-series prediction map is generated, including: The operation feature sequence and energy consumption feature sequence are extracted from the collaborative features of the equipment group. The fluctuation cycle analysis is performed on the operation feature sequence to calculate the state change frequency vector, and the amplitude decomposition is performed on the energy consumption feature sequence to obtain the state change amplitude vector. The state evolution matrix is ​​constructed by fusing the state change frequency vector and the state change amplitude vector, and a fault propagation tree is generated based on the state evolution matrix. The location of the propagation point is determined based on the fault propagation tree. The propagation speed value is calculated using the state change frequency in the state evolution matrix. The propagation attenuation coefficient is obtained through the state change amplitude. The propagation speed value and the propagation attenuation coefficient are combined to construct the propagation range matrix. The propagation channel matrix is ​​formed using the propagation range matrix. The propagation range matrix and the propagation channel matrix are combined to generate a state transition sequence. The state transition sequence is used to calculate the transition probability. The fluctuation characteristics are constructed by combining the state change sequence. The equipment node failure risk value is calculated by using the transition probability and the fluctuation characteristics. The equipment node failure risk value is mapped to the state evolution matrix, and the energy consumption evolution trend is calculated by combining the state change frequency vector and the state change amplitude vector. The equipment node failure risk value and energy consumption evolution trend are arranged in the time dimension to generate a time series prediction map. Based on time-series prediction maps, a set of devices with risk exceeding the threshold is identified. A multi-objective optimization model constrained by maintenance resources is constructed, a load allocation scheme is calculated, and iterative correction is performed based on feedback from equipment state changes. Control commands are generated, including: The fault risk value of each device node is extracted from the time series prediction map, and the device nodes whose fault risk value exceeds the preset risk threshold are screened out to form a set of devices with risk exceeding the threshold. Extract the fault propagation path and energy consumption transmission relationship of each device in the set of devices with risk exceeding the threshold. Calculate the fault impact weight between devices based on the fault propagation path and the energy consumption coupling coefficient between devices based on the energy consumption transmission relationship. Use the fault impact weight and energy consumption coupling coefficient as device association parameters. Combine the total maintenance resources and load transfer capacity to construct a multi-objective optimization model that includes the objectives of minimizing fault risk and minimizing energy consumption. Input the load values ​​of each device in the set of devices with risk exceeding the threshold into the multi-objective optimization model to calculate the initial load allocation scheme; The initial load allocation scheme is sent to the set of devices with risk exceeding the threshold. The changes in fault risk value and energy consumption after execution are collected. The fault impact weight and energy consumption coupling coefficient are updated. The process is repeated until the changes in fault risk value and energy consumption meet the convergence condition. The final load allocation scheme is then output, and control commands are generated based on the final load allocation scheme. Control commands are divided into multi-level execution queues according to risk priority. Load adjustment commands and maintenance strategies are issued in a tiered manner to the set of devices with risks exceeding the threshold. The collaborative characteristics of the device group are corrected by the device operating status, and the maintenance execution sequence is dynamically adjusted, including: Extract the fault risk value and maintenance time window parameters of each device contained in the control instructions, calculate the risk urgency value of each device, sort the control instructions according to the risk urgency value, and divide them into multi-level execution queues; Based on the risk urgency value range corresponding to each execution level in the multi-level execution queue, the devices in the risk exceeding the threshold device set are assigned to different execution levels, and corresponding load adjustment instructions and maintenance strategies are issued to the devices at different execution levels respectively; Collect equipment operation status data of each device in the risk-exceeding device set after executing load adjustment commands and maintenance strategies, calculate the change in fault risk value and the change in energy consumption transfer relationship between devices before and after execution, and update the collaborative characteristics of the device group based on the change in fault risk value and the change in energy consumption transfer relationship. Extract the fault propagation path change information of each device in the updated device group collaboration features, recalculate the risk urgency value of each device, adjust the device execution order in the multi-level execution queue based on the recalculated risk urgency value, and generate the adjusted maintenance execution order.

2. The method according to claim 1, characterized in that, By collecting operational status data and energy consumption data of industrial equipment clusters through distributed edge nodes, and performing time alignment processing using a multi-point timestamp calibration algorithm, the equipment-level time-series data is obtained, including: The industrial equipment group is divided into multiple acquisition areas based on communication distance and equipment density. Master-slave distributed edge nodes are set up in the acquisition areas to build a hierarchical edge node communication network and generate a multi-layer acquisition network topology. Based on the analysis of the multi-layer acquisition network topology, the operating load of the equipment and the network communication load are analyzed to calculate the optimal sampling period and buffer capacity, and an adaptive sampling scheme is generated. Based on the adaptive sampling scheme, the operating status data and energy consumption data of industrial equipment groups are collected through distributed edge nodes, and a data collection timestamp and transmission timestamp mapping table is established to form the raw sampling data. A distributed clock synchronization mechanism is constructed, which uses edge nodes to calculate transmission path delay and node clock drift, generates timestamp correction compensation values, and performs distributed time alignment on the original sampled data. Multi-dimensional feature extraction and data consistency verification are performed on the time-aligned data. Based on the verification results, the data is reconstructed to obtain device-level time-series data.

3. The method according to claim 1, characterized in that, Wavelet transform decomposition is performed on device-level time series data to extract operational trend features and abnormal fluctuation features. A two-layer time series feature representation is constructed by combining the physical connection sequence of the devices, including: A fluctuation analysis matrix is ​​constructed for equipment-level time series data, the fluctuation trend is calculated, and wavelet transform decomposition is performed on equipment-level time series data at different time scales based on the fluctuation trend to generate multi-scale wavelet decomposition coefficients. The trend and fluctuation components of equipment-level time series data are reconstructed using multi-scale wavelet decomposition coefficients. The phase relationship and amplitude ratio between the components are extracted to construct the operation trend characteristics. A feature analysis matrix is ​​constructed for the wave components to extract wave energy distribution and wave morphology features. Abnormal wave characteristics and wave propagation paths are calculated through wave feature analysis. Construct the device connection structure according to the physical connection order of the devices, mark the physical connection relationships of the devices, establish the device transmission path, and generate the device connection matrix; By mapping operational trend characteristics and abnormal fluctuation characteristics to the device connection matrix, and combining the spatial correlation between fluctuation propagation path and device transmission path, a two-layer temporal feature representation containing temporal and spatial characteristics is constructed.

4. The method according to claim 1, characterized in that, Based on a two-layer time-series feature representation, the strength of causal influence and transmission delay between devices are calculated. An operation and maintenance coupling matrix and an energy consumption transmission matrix are constructed, generating device group collaborative features including: Multi-scale state decomposition is performed on the two-layer time series feature representation. State change features are extracted through fluctuation feature analysis and amplitude reconstruction. The state transition law is calculated by combining the change features to obtain the intensity and depth of causal influence between devices and form a causal feature sequence. Based on the causal characteristic sequence analysis of the state change process, combined with the state transition law, the equipment state response time and response amplitude are calculated, the transmission delay and transmission link between equipment are determined, and the transmission characteristic matrix is ​​constructed. The transmission feature matrix is ​​mapped and matched with the device connection order to analyze the direct impact transmission caused by state changes between adjacent devices, as well as the interval impact transmission caused by intermediate devices. The coupling degree between devices is calculated, and the operation and maintenance coupling relationship matrix is ​​constructed by combining the causal impact strength and impact depth. Based on the operation and maintenance coupling relationship matrix, energy transfer paths are identified, energy transfer efficiency is calculated by combining transfer delay and transfer link, energy transfer direction and transfer gain are determined by coupling degree analysis, and energy consumption transfer relationship matrix is ​​generated. The operation and maintenance coupling relationship matrix and the energy consumption transfer relationship matrix are fused to generate the collaborative features of the equipment group.

5. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 4.

6. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.

Images