An energy efficiency abnormality early warning method based on equipment running state recognition

Abstract

The application discloses an energy efficiency abnormality early warning method based on equipment running state recognition, comprising the following steps: collecting equipment running related time sequence state data through a multi-dimensional sensing array, extracting dynamic characteristics by using an equipment state time sequence difference sensing algorithm, completing mode classification through a time sequence mode self-adaptive clustering algorithm, inputting the classified feature data into a working condition adaptive abnormality discrimination model for abnormality screening, integrating and analyzing to determine the energy efficiency abnormality grade through an operation state intelligent research and judgment platform, and finally generating a targeted early warning signal. Hardware adopts a variety of special sensors and edge computing gateways, distributed architecture computing nodes, and software parameters are dynamically adjusted to adapt to different working conditions. The method captures equipment running abnormality characteristics through multi-algorithm cooperation and multi-dimensional data integration, dynamically adapts to complex working conditions, improves early warning accuracy and scene adaptability, and solves the problems of insufficient feature extraction precision and poor working condition adaptability in the prior art.

Description

Technical Field

[0001] This invention relates to the field of equipment operation anomaly early warning technology, and in particular to an energy efficiency anomaly early warning method based on equipment operation status identification. Background Technology

[0002] In the long-term operation of industrial equipment, abnormal energy efficiency can directly lead to increased energy consumption, accelerated equipment wear and tear, and even safety hazards. Therefore, timely early warning of abnormal equipment energy efficiency has become a key requirement for improving the quality and efficiency of industrial production. In current industrial scenarios, equipment operating environments are complex and operating conditions are highly variable. Traditional energy efficiency monitoring methods struggle to accurately capture subtle changes in equipment operating status. There is an urgent need for an intelligent early warning solution that combines multiple algorithms and covers the entire operating process. Energy efficiency anomaly early warning methods based on equipment operating status identification have emerged in this context. By integrating multi-dimensional data collection and intelligent algorithm analysis, they achieve accurate identification and early warning of abnormal equipment energy efficiency.

[0003] Currently, equipment energy efficiency anomaly monitoring typically involves first collecting equipment operating data using a single type of sensor, then extracting features using simple data comparison or basic statistical analysis methods, followed by determining whether energy efficiency anomalies exist based on fixed thresholds, and finally issuing a warning signal through a simple alarm device. Some technologies incorporate basic clustering algorithms or discriminant models, but lack in-depth mining of the dynamic features of time-series data and are not adapted and optimized for different operating conditions. They can only achieve coarse anomaly judgment under a single operating condition, making it difficult to adapt to the complex and ever-changing operating scenarios of equipment, thus limiting the accuracy and adaptability of warnings.

[0004] Existing technologies have two core drawbacks: First, the feature extraction accuracy is insufficient. They rely on a single data dimension or simple difference methods to obtain features, failing to fully capture the temporal correlation and subtle fluctuations in the equipment's operating status. This results in incomplete identification of abnormal features and a high risk of missed detections. Second, the adaptability to operating conditions is poor. The discrimination models used are mostly built based on fixed operating parameters and do not consider the differences in equipment energy efficiency standards under different operating conditions. They cannot dynamically adjust the discrimination logic according to real-time operating conditions, which increases the warning error when switching operating conditions and makes it difficult to meet the accurate warning needs in complex industrial scenarios. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides an energy efficiency anomaly early warning method based on equipment operating status identification.

[0006] The technical solution adopted in this invention is an energy efficiency anomaly early warning method based on equipment operating status identification, comprising the following steps: S1, collecting time-series status data related to energy efficiency during equipment operation through a multi-dimensional sensor array, wherein the time-series status data includes equipment load characteristic data, power output characteristic data, energy consumption characteristic data, and environmental impact characteristic data; S2, using an equipment status time-series differential sensing algorithm to dynamically extract features from the collected time-series status data, and capturing subtle fluctuation features and trend change features implicit in the data sequence by constructing a differential mapping relationship between adjacent time points of the time-series data; S3, using a time-series modal adaptive clustering algorithm to perform modal classification on the dynamic features extracted in S2. S4. Based on feature similarity and temporal correlation, an adaptive clustering criterion is constructed to classify feature data under different operating modes; S5. The modal feature data classified in S3 is input into the operating condition adaptation anomaly discrimination model, and a multi-dimensional discrimination boundary is constructed in combination with the equipment's preset operating condition parameters to screen for anomalies in the feature data under each mode; S6. The anomaly screening results of S4 are integrated through the intelligent operating status analysis platform, and the historical operating status data and real-time operating condition information of the equipment are combined for comprehensive analysis to determine the energy efficiency anomaly level corresponding to the anomaly feature; S7. Based on the energy efficiency anomaly level determined in S5, a targeted early warning signal is generated, and the early warning signal carries the source of the anomaly feature, the anomaly development trend, and related operating parameter information.

[0007] Furthermore, the expression for the device state timing differential sensing algorithm is as follows: ,in The device timing status data at time t. for Time-series status data of the device. The time interval for time-series data acquisition. This is the difference time lag factor. The dynamic weighting coefficients at time t are used; the feature enhancement formula is calculated as follows: ,in Let be the enhanced feature data at time t, and n be the number of feature fusion dimensions. Let be the fusion weight of the i-th feature. Let be the nonlinear mapping function for the i-th feature.

[0008] Furthermore, the expression for the temporal modality adaptive clustering algorithm is: ,in For the k-th cluster, For the feature data to be clustered, Let be the dynamic center of the k-th cluster at time t. Operators for Euclidean distance calculation Let be the adaptive adjustment coefficient for the k-th cluster; the cluster center update formula is calculated as follows: ,in For the k-th cluster in The center is updated in real time. The update rate coefficient is centered. Let be the amount of feature data included in the k-th cluster at time t. t is the clustering adjustment factor at time t.

[0009] Furthermore, the expression for the working condition adaptation anomaly discrimination model is: ,in For anomaly detection results, Here, M is the Sigmoid activation function, and M is the number of discriminant feature dimensions. Let be the weight coefficient of the m-th discriminative feature. The working condition adaptation mapping function is for the m-th discriminative feature. For the m-th modal feature data, is the preset standard parameter for the m-th feature, and b is the model bias term.

[0010] Furthermore, the multi-dimensional sensing array includes: a fiber optic strain sensor, a piezoelectric vibration sensor, an infrared temperature sensor, an ultrasonic flow sensor, and a Hall current sensor. Each sensor establishes a communication connection with an edge computing gateway via an industrial Ethernet bus. The edge computing gateway has a built-in FPGA chip for real-time reception and forwarding of time-series data. In the device status time-series differential sensing algorithm, the time lag factor ranges from 1 to 5, and the dynamic weight coefficient is dynamically adjusted according to the device operating condition level. The initial number of clusters in the time-series modality adaptive clustering algorithm is set to 3 to 8 based on the statistical results of the device's historical operating modes. The number of discriminant feature dimensions in the operating condition adaptation anomaly discrimination model is set to 6 to 12, and the weight coefficient is obtained through iterative optimization using the gradient descent method. The intelligent judgment platform for operating status adopts a distributed computing architecture, including 4 computing nodes, each configured with a 16-core processor and 64GB of memory.

[0011] Further, step S2 includes the following sub-steps: S21, selecting a continuous data segment of a preset length from the time-series state data, determining the specific value of the differential time lag factor, and establishing a calculation rule for the dynamic weight coefficient based on the device operating rate and data acquisition frequency; S22, pairing adjacent time-time data of the selected data segment, calculating the time-series differential data of each time-time according to the differential expression, and retaining the data sequence after differential calculation and the corresponding timestamp information; S23, dynamically identifying abnormal fluctuation points in the differential data sequence, determining the fluctuation threshold through local data distribution characteristics, and filtering out differential data exceeding the threshold range; S24, associating and mapping the filtered differential data with the original time-series data to construct a composite feature set including the original features and differential features, providing data support for subsequent modal clustering.

[0012] Further, S3 includes the following sub-steps: S31, based on the feature dimension and data distribution characteristics of the composite feature set, initialize the number of clusters and the initial center of each cluster, and set the number of clustering iterations and convergence criteria; S32, calculate the distance between each composite feature data and the center of each cluster, and determine the cluster to which the data belongs by combining the adaptive adjustment coefficient, thus completing the first round of clustering; S33, calculate the internal compactness and inter-cluster separation of each cluster based on the results of the first round of clustering, and dynamically adjust the cluster center and adaptive adjustment coefficient based on the balance between the two; S34, repeat the clustering and parameter adjustment process until the convergence criteria are met, and output the final modality classification result to ensure the effective separation of feature data under different operating modalities.

[0013] Further, step S4 includes the following sub-steps: S41, extracting feature parameters from the classification results of each modality, and determining the operating condition adaptation range and discrimination threshold corresponding to each feature parameter by combining the operating standards of the equipment under different operating conditions; S42, inputting the feature parameters into the operating condition adaptation anomaly discrimination model, and completing the adaptation calculation between the feature parameters and the operating condition standards through a preset operating condition adaptation mapping function; S43, comparing and analyzing the model output results with the discrimination threshold to identify abnormal feature parameters that exceed the adaptation range, and recording the specific category and occurrence time of the abnormal parameters; S44, performing correlation analysis on the identified abnormal feature parameters to determine whether the anomaly is caused by a single parameter anomaly or a multi-parameter collaborative anomaly, and forming a complete anomaly screening report.

[0014] Further, S5 includes the following sub-steps: S51, calling the historical data storage module of the intelligent judgment platform for operating status, extracting historical operating status data and energy efficiency anomaly records of the equipment within the past three months, and establishing a comparison database of historical data and real-time data; S52, matching and analyzing the anomaly screening report output by S4 with the historical data in the comparison database, finding similar anomaly cases and corresponding processing results; S53, combining the equipment load rate, ambient temperature, and power supply voltage parameters in the real-time operating condition information, predicting the development trend of anomaly characteristics, and assessing the degree of energy efficiency impact that anomalies may cause; S54, comprehensively considering the anomaly matching results and trend prediction conclusions, using a hierarchical evaluation method to determine the energy efficiency anomaly level, and storing the level classification results in association with the anomaly characteristic data.

[0015] An energy efficiency anomaly early warning method based on equipment operating status identification is disclosed. This method is implemented through an energy efficiency anomaly early warning system based on equipment operating status identification, comprising: a multi-dimensional data acquisition unit for equipment operating status, a time-series state data differential feature extraction unit, a composite feature time-series modality adaptive clustering unit, a working condition-adaptive multi-dimensional anomaly discrimination unit, an intelligent comprehensive judgment unit for operating status, and an energy efficiency anomaly graded early warning signal generation unit. The multi-dimensional data acquisition unit collects various time-series state data through a multi-dimensional sensor array and transmits the data to the time-series state data differential feature extraction unit. The time-series state data differential feature extraction unit processes the data using an equipment state time-series differential sensing algorithm. Subsequently, the composite feature set is sent to the composite feature temporal modality adaptive clustering unit; the composite feature temporal modality adaptive clustering unit completes modality classification through the temporal modality adaptive clustering algorithm, and transmits the classification results to the operating condition adaptation multi-dimensional anomaly discrimination unit; the operating condition adaptation multi-dimensional anomaly discrimination unit uses the operating condition adaptation anomaly discrimination model to perform anomaly screening, and sends the screening report to the operating status intelligent comprehensive judgment unit; the operating status intelligent comprehensive judgment unit integrates the data and determines the anomaly level, and transmits the level information to the energy efficiency anomaly classification early warning signal generation unit; the energy efficiency anomaly classification early warning signal generation unit generates an early warning signal carrying detailed information according to the anomaly level, and provides early warning of equipment energy efficiency anomalies.

[0016] Beneficial Effects: This invention proposes an energy efficiency anomaly early warning method based on equipment operating status identification. It comprehensively collects equipment operating data through a multi-dimensional sensor array, deeply mines subtle fluctuations and trend changes in time-series data using a time-series state differential sensing algorithm, and accurately classifies feature data under different operating modes using a time-series modality adaptive clustering algorithm. A judgment boundary is dynamically constructed by combining a condition-adaptive anomaly discrimination model with real-time operating conditions. Finally, a comprehensive analysis of historical and real-time data is conducted through an intelligent operating status judgment platform to determine the anomaly level and generate targeted early warnings. This invention abandons single data dimensions and simple analysis methods, and achieves comprehensive capture of anomaly features by deeply mining the temporal correlation and implicit features of data through multi-algorithm collaboration, effectively avoiding missed detections. Addressing the problem of poor condition adaptability, its discrimination model dynamically adjusts the adaptation logic based on real-time operating parameters, fully considering the differences in energy efficiency standards under different operating conditions, significantly reducing early warning errors during operating condition switching. Simultaneously, the comprehensive analysis capabilities of multi-dimensional data acquisition and the intelligent judgment platform improve the accuracy and adaptability of early warnings, meeting the needs for accurate early warning of equipment energy efficiency anomalies in complex industrial scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0018] Figure 2 This is a flowchart of method step S2 of the present invention;

[0019] Figure 3 This is a flowchart of method step S3 of the present invention;

[0020] Figure 4 This is a flowchart of method step S4 of the present invention;

[0021] Figure 5 This is a flowchart of step S5 of the method of the present invention. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1As shown, an energy efficiency anomaly early warning method based on equipment operating status identification includes the following steps: S1, collecting time-series status data related to energy efficiency during equipment operation through a multi-dimensional sensor array, wherein the time-series status data includes equipment load characteristic data, power output characteristic data, energy consumption characteristic data, and environmental impact characteristic data; S2, using an equipment status time-series differential sensing algorithm to dynamically extract features from the collected time-series status data, and capturing subtle fluctuation features and trend change features hidden in the data sequence by constructing a differential mapping relationship between adjacent time moments of the time-series data; S3, using a time-series modal adaptive clustering algorithm to perform modal classification on the dynamic features extracted in S2, based on the features... Similarity and temporal correlation are used to construct an adaptive clustering criterion for classifying feature data under different operating modes; S4, the modal feature data classified in S3 is input into the operating condition adaptation anomaly discrimination model, and multi-dimensional discrimination boundaries are constructed in combination with the equipment's preset operating condition parameters to screen for anomalies in feature data under each mode; S5, the anomaly screening results of S4 are integrated through the intelligent operating status analysis platform, and the historical operating status data and real-time operating condition information of the equipment are combined for comprehensive analysis to determine the energy efficiency anomaly level corresponding to the anomaly features; S6, a targeted early warning signal is generated based on the energy efficiency anomaly level determined in S5, and the early warning signal carries the source of the anomaly features, the anomaly development trend, and related operating parameter information.

[0024] Step S1 involves using a multi-dimensional sensor array composed of fiber optic strain sensors, piezoelectric vibration sensors, infrared temperature sensors, ultrasonic flow sensors, and Hall current sensors to continuously collect time-series status data related to energy efficiency during equipment operation at a sampling frequency of 20 times per second. Each sensor establishes a communication connection with an edge computing gateway via an industrial Ethernet bus. The edge computing gateway, with its built-in FPGA chip, enables real-time data reception and forwarding. The collected time-series status data specifically includes equipment load characteristic data, power output characteristic data, energy consumption characteristic data, and environmental impact characteristic data. The load characteristic data includes equipment... During operation, data on stress distribution and load-bearing strength are collected. Power output characteristic data includes output power and operating speed. Energy consumption characteristic data includes the consumption rate and cumulative total consumption of various energy sources. Environmental impact characteristic data includes parameters related to temperature, humidity, and air pressure in the equipment's operating environment. Data collection is conducted continuously without interruption, with each data point carrying a corresponding timestamp. This provides comprehensive, accurate, and time-series correlated raw data support for subsequent feature extraction and analysis. This step, through collaborative collection from multiple types of sensors, breaks the limitations of a single data dimension and ensures that the data covers the core energy efficiency influencing factors of equipment operation.

[0025] Step S2 involves using a device status time-series differential sensing algorithm to dynamically extract features from the collected time-series status data. First, 100 consecutive data points from the time-series status data are selected as a data segment. The differential time lag factor is set to 3. A calculation rule for the dynamic weight coefficient is established based on the device operating rate and data acquisition frequency. The dynamic weight coefficient ranges from 0.1 to 0.9 and is adjusted in real-time according to the device operating condition level. Then, adjacent time points are paired within the selected data segment, and the time-series differential data for each time point is calculated according to the set differential logic. The differential data is then retained. The data sequence and corresponding timestamp information are used to dynamically identify abnormal fluctuation points in the differential data sequence. The fluctuation threshold is determined by the local data distribution characteristics, and differential data exceeding the threshold range is filtered out. The filtered differential data is associated and mapped with the original time series data to construct a composite feature set including the original features and differential features. In this process, the subtle fluctuation features and trend change features in the data are enhanced by differential operation, which effectively captures the dynamic change pattern of the equipment's operating status and provides rich and highly identifiable feature data support for subsequent modal clustering, ensuring the comprehensiveness and accuracy of feature extraction.

[0026] Step S3 involves using a temporal modal adaptive clustering algorithm to perform modal classification on the dynamic features extracted in S2. First, based on the feature dimension and data distribution characteristics of the composite feature set, the number of clusters is initialized to 5, and the initial center of each cluster is set. The number of clustering iterations is set to 50, and the convergence criterion is that the change in the cluster center between two adjacent iterations is less than 0.001. Then, the distance between each composite feature data point and the center of each cluster is calculated. Combined with an adaptive adjustment coefficient, the cluster to which the data belongs is determined. The adaptive adjustment coefficient ranges from 0.8 to 1.2. The cluster is dynamically classified according to the data distribution density within the cluster. The process involves adjusting and completing the first round of clustering. Based on the results of the first round of clustering, the internal compactness and inter-cluster separation of each cluster are calculated. The cluster centers and adaptive adjustment coefficients are dynamically adjusted based on the balance between the two. This clustering and parameter adjustment process is repeated until the convergence criteria are met or the maximum number of iterations is reached. The final modality classification result is then output. This process, through the construction and dynamic adjustment of adaptive clustering criteria, achieves accurate segmentation of feature data under different operating modalities, ensuring that data under the same modality have high similarity and data under different modalities have significant differences, thus providing a targeted classification data foundation for anomaly detection.

[0027] Step S4 involves inputting the modal feature data classified in S3 into the operating condition adaptation anomaly discrimination model. First, key feature parameters are extracted from each modal classification result. The number of key feature parameters is set to eight, including core dimensions such as load, energy consumption, and power output. Combined with the operating standards of the equipment under different operating conditions, the operating condition adaptation range and discrimination threshold corresponding to each feature parameter are determined. The discrimination threshold is obtained through statistical analysis of the equipment's historical normal operating data. Then, the key feature parameters are input into the operating condition adaptation anomaly discrimination model. The adaptation calculation between the feature parameters and the operating condition standards is completed through a preset operating condition adaptation mapping function. Based on the model output results and the discrimination threshold, a comparative analysis is performed to identify abnormal feature parameters that exceed the adaptation range. The specific category and occurrence time of the abnormal parameters are recorded. Correlation analysis is conducted on the identified abnormal feature parameters to determine whether the anomaly is caused by a single parameter anomaly or a multi-parameter collaborative anomaly, forming a complete anomaly screening report. During this process, the model dynamically constructs the discrimination boundary in conjunction with real-time operating conditions, ensuring that the anomaly discrimination is highly adapted to the actual operating conditions of the equipment, improving the accuracy and targeting of anomaly identification, and effectively distinguishing between normal fluctuations and real anomalies.

[0028] Step S5 involves integrating the anomaly screening results from S4 through an intelligent operational status assessment platform. This platform employs a distributed computing architecture, comprising four computing nodes, each equipped with a 16-core processor and 64GB of memory, providing efficient data processing and analysis capabilities. First, it utilizes the platform's historical data storage module to extract historical operational status data and energy efficiency anomaly records from the past three months, establishing a comparison database of historical and real-time data. The anomaly screening report output by S4 is then matched and analyzed against the historical data in the comparison database to identify similar anomaly cases and their corresponding handling results. Combined with parameters such as equipment load rate, ambient temperature, and power supply voltage from real-time operating conditions, the development trend of anomaly characteristics is predicted, and the potential energy efficiency impact of the anomaly is assessed. A hierarchical assessment method is used to determine the energy efficiency anomaly level, categorized into three levels: minor, moderate, and severe anomalies. Each level corresponds to specific characteristic judgment criteria. The level classification results are stored in association with the anomaly characteristic data. This process, through the fusion analysis of historical and real-time data, achieves in-depth anomaly assessment, ensuring the scientific and rational nature of the anomaly level determination and providing accurate basis for subsequent early warning signal generation.

[0029] Step S6 involves generating targeted early warning signals based on the energy efficiency anomaly levels determined in S5. First, the priorities of the early warning signals are set according to the anomaly level: minor anomalies correspond to Level 1 priority, moderate anomalies to Level 2 priority, and severe anomalies to Level 3 priority. Early warning signals of different priorities differ in transmission rate and response mechanism. The information carried by the early warning signal includes the source of the anomaly characteristics, the anomaly development trend, and related operating parameter information. The source of the anomaly characteristics clearly points to the specific anomaly parameter category and the corresponding equipment operating part. The anomaly development trend, based on the prediction results of S5, details the possible direction and speed of the anomaly's evolution. The related operating parameter information includes real-time values ​​of key parameters related to the anomaly, such as load, energy consumption, and environment. The early warning signal is transmitted to the equipment control system and monitoring center via the industrial Ethernet bus and is also prominently displayed on the monitoring center's interface to ensure that relevant personnel can quickly obtain anomaly information. This process, by generating targeted early warning signals carrying comprehensive information, enables timely notification of equipment energy efficiency anomalies, providing accurate guidance for subsequent anomaly handling and equipment adjustments, and ensuring the safety and energy efficiency stability of equipment operation.

[0030] Preferably, the expression for the device state timing differential sensing algorithm is: ,in The device timing status data at time t. for Time-series status data of the device. The time interval for time-series data acquisition. This is the difference time lag factor. The dynamic weighting coefficients at time t are used; the feature enhancement formula is calculated as follows: ,in Let be the enhanced feature data at time t, and n be the number of feature fusion dimensions. Let be the fusion weight of the i-th feature. Let be the nonlinear mapping function for the i-th feature.

[0031] Specifically, the equipment status time-series differential sensing algorithm, based on the dynamic change characteristics and feature enhancement requirements of time-series data, addresses the difficulty in capturing subtle fluctuations in the original time-series data. It establishes a correlation between the difference between adjacent time-time data and the time interval, introducing dynamic weight coefficients to balance the influence weights of data from different time-time points, forming a basic differential formula. This formula is based on the continuity of equipment operating status; the difference between adjacent time-time data directly reflects the rate of status change. The introduction of dynamic weight coefficients is to adapt to the sensitivity requirements of different operating stages of the equipment. Building upon this, to further enhance feature recognition, the differential data is enhanced by fusing multi-dimensional nonlinear mapping results, resulting in a feature enhancement formula. This calculation logic stems from the problem of insufficient single-dimensional differential features; multi-dimensional fusion enriches feature expression. The time lag factor ranges from 1 to 5, adjusted according to the equipment operating rate, using a smaller value for higher operating rates and a larger value for lower rates. The time interval is fixed at the data acquisition cycle. The dynamic weight coefficient ranges from 0.1 to 0.9, adjusted in real-time according to the operating condition level. The number of feature fusion dimensions is set from 6 to 12, and the fusion weights are determined through data training. The nonlinear mapping function adopts a commonly used nonlinear transformation form. During implementation, continuous time-series data is first acquired, the specific values ​​of each parameter are determined, and the difference data is calculated according to the basic difference formula. Then, the multi-dimensional mapping results are fused through the feature enhancement formula to output the enhanced feature data. This process, through the formula formed by two steps, not only ensures the accurate capture of time-series changes but also achieves in-depth feature enhancement, providing high-quality feature support for subsequent modality clustering and effectively improving the comprehensiveness and accuracy of feature extraction.

[0032] Preferably, the expression for the temporal modality adaptive clustering algorithm is: ,in For the k-th cluster, For the feature data to be clustered, Let be the dynamic center of the k-th cluster at time t. Operators for Euclidean distance calculation Let be the adaptive adjustment coefficient for the k-th cluster; the cluster center update formula is calculated as follows: ,in For the k-th cluster in The center is updated in real time. The update rate coefficient is centered. Let be the amount of feature data included in the k-th cluster at time t. t is the clustering adjustment factor at time t.

[0033] Specifically, the temporal modality adaptive clustering algorithm achieves effective grouping of feature data based on the accuracy of cluster partitioning and the need for dynamic adaptation. It constructs a cluster partitioning formula with the distance between feature data and cluster centers as the core, and introduces an adaptive adjustment coefficient to balance intra-cluster compactness and inter-cluster separation. This formula takes advantage of the characteristic that feature data of the same type are closer to the cluster centers. The adaptive adjustment coefficient is used to solve the problem that fixed clustering criteria are difficult to adapt to dynamic data distribution. To achieve dynamic optimization of cluster centers, based on the error feedback of the first round of clustering results, an update rate coefficient and an adjustment factor are introduced to calculate the cluster center update formula. This calculation logic stems from the problem that fixed cluster centers lead to insufficient classification accuracy. Dynamic updates enable cluster centers to continuously fit the data distribution. The initial number of clusters is set to 3 to 8, determined based on the statistical results of the equipment's historical operating modes. The adaptive adjustment coefficient ranges from 0.8 to 1.2, dynamically adjusted according to the data distribution density within each cluster. The center update rate coefficient ranges from 0.01 to 0.1 to ensure stable convergence of the update process. The clustering adjustment factor ranges from 0.9 to 1.1, dynamically changing according to the number of iterations. During implementation, the clustering parameters are first initialized, and the first round of classification is completed according to the clustering partitioning formula. Then, the error is calculated based on the classification results, and the center position is iteratively optimized through the cluster center update formula until the convergence condition is met. This process, through the formula formed by two steps, achieves dynamic adaptation between the clustering criteria and the cluster centers, ensuring accurate partitioning of feature data under different operating modes and providing targeted classification data for anomaly detection.

[0034] Preferably, the expression for the working condition adaptation anomaly discrimination model is: ,in For anomaly detection results, Here, M is the Sigmoid activation function, and M is the number of discriminant feature dimensions. Let be the weight coefficient of the m-th discriminative feature. The working condition adaptation mapping function is for the m-th discriminative feature. For the m-th modal feature data, is the preset standard parameter for the m-th feature, and b is the model bias term.

[0035] Specifically, the operating condition adaptation anomaly discrimination model is based on the adaptation requirements between anomaly features and operating condition standards. Addressing the issue of differing anomaly judgment standards under different operating conditions, it constructs a weighted fusion of multi-dimensional features and an operating condition adaptation mapping relationship. An activation function is introduced to normalize the anomaly results. Based on the differences in energy efficiency standards under different equipment operating conditions, the weighted fusion highlights the impact of key features. The operating condition adaptation mapping function is used to convert feature data into judgment indicators corresponding to the operating conditions. The introduction of the activation function is to limit the output results to a reasonable range, facilitating anomaly judgment. The number of discriminative feature dimensions is set to 6 to 12, including core dimensions such as load, energy consumption, and power output. The weight coefficients are obtained through iterative optimization using the gradient descent method, with the number of iterations set to 1000 to 5000 to ensure convergence of weight optimization. The bias term ranges from -0.5 to 0.5 to correct for model system errors. During implementation, key feature parameters under each modality are first extracted, standard parameters for corresponding working conditions are determined, and specific values ​​for the weights and biases of each parameter are set. The feature data and working condition standards are adapted and converted through a working condition adaptation mapping function. After weighted fusion and activation function processing, the anomaly discrimination result is output. This formula, by integrating multi-dimensional working condition adaptation logic, achieves accurate identification of abnormal features under different working conditions, effectively distinguishes between normal fluctuations and real anomalies, provides accurate anomaly screening results for subsequent intelligent judgment, and improves the working condition adaptation capability and discrimination accuracy of the early warning method.

[0036] Preferably, the multi-dimensional sensing array includes: a fiber optic strain sensor, a piezoelectric vibration sensor, an infrared temperature sensor, an ultrasonic flow sensor, and a Hall current sensor. Each sensor establishes a communication connection with an edge computing gateway via an industrial Ethernet bus. The edge computing gateway has a built-in FPGA chip for real-time reception and forwarding of time-series data. In the device status time-series differential sensing algorithm, the time lag factor ranges from 1 to 5, and the dynamic weight coefficient is dynamically adjusted according to the device operating condition level. The initial number of clusters in the time-series modality adaptive clustering algorithm is set to 3 to 8 based on the statistical results of the device's historical operating modes. The number of discriminant feature dimensions in the operating condition adaptation anomaly discrimination model is set to 6 to 12, and the weight coefficient is obtained through iterative optimization using the gradient descent method. The intelligent judgment platform for operating status adopts a distributed computing architecture, including 4 computing nodes, each configured with a 16-core processor and 64GB of memory.

[0037] Preferred, such as Figure 2As shown, step S2 includes the following sub-steps: S21, selecting a continuous data segment of a preset length from the time-series state data, determining the specific value of the differential time lag factor, and establishing a calculation rule for the dynamic weight coefficient based on the device operating rate and data acquisition frequency; S22, pairing adjacent time-time data of the selected data segment, calculating the time-series differential data of each time-time according to the differential expression, and retaining the data sequence after differential calculation and the corresponding timestamp information; S23, dynamically identifying abnormal fluctuation points in the differential data sequence, determining the fluctuation threshold through local data distribution characteristics, and filtering out differential data exceeding the threshold range; S24, associating and mapping the filtered differential data with the original time-series data to construct a composite feature set including the original features and differential features, providing data support for subsequent modal clustering.

[0038] Specifically, step S2 includes: S21 First, select 100 consecutive data points from the collected time-series status data to form a data segment. Combined with the device's acquisition frequency of 20 times per second and operating rate, determine the specific value of the differential time lag factor to be 3. Simultaneously, based on the device's operating condition level classification standard, establish a dynamic weighting coefficient calculation rule to ensure that the weighting coefficient can be adjusted in real time according to changes in operating conditions; S22 Pair the selected data segment with adjacent time points according to timestamp order, follow the differential operation logic to complete the calculation of the time-series differential data at each time point, and synchronously record the timestamp information corresponding to each differential data point to ensure the temporal correlation of the data; S23 Statistically analyze local data... Distribution characteristics are analyzed, and a fluctuation threshold is calculated. This threshold is set at three times the standard deviation of the local data. Based on this threshold, abnormal fluctuation points that exceed the range in the differential data sequence are screened out, accurately locating potential feature mutations. In step S24, the screened abnormal differential data are mapped one-to-one with the original time series data according to timestamps, and integrated to form a composite feature set including the original basic features and differential change features. Each feature dimension is sorted by importance, and the top 12 core dimensions are retained. This step-by-step process ensures the comprehensiveness and accuracy of dynamic feature extraction through a complete link from data selection, differential calculation, anomaly screening to feature integration, providing a high-quality data foundation for subsequent modality clustering.

[0039] Preferred, such as Figure 3As shown, step S3 includes the following sub-steps: S31, based on the feature dimension and data distribution characteristics of the composite feature set, initialize the number of clusters and the initial center of each cluster, and set the number of clustering iterations and convergence criteria; S32, calculate the distance between each composite feature data and the center of each cluster, and determine the cluster to which the data belongs by combining the adaptive adjustment coefficient, thus completing the first round of clustering; S33, calculate the internal compactness and inter-cluster separation of each cluster based on the results of the first round of clustering, and dynamically adjust the cluster center and adaptive adjustment coefficient based on the balance between the two; S34, repeat the clustering and parameter adjustment process until the convergence criteria are met, and output the final modality classification result to ensure the effective separation of feature data under different operating modalities.

[0040] Specifically, step S3 includes S31, first analyzing the feature dimensions (12 dimensions in total) and data distribution characteristics of the composite feature set; based on the statistical results of the equipment's historical operating modes (historical data shows that the equipment has 5 typical operating modes), initializing the number of clusters to 5; determining the initial center of each cluster using the K-means++ algorithm; simultaneously setting the number of clustering iterations to 50; and determining the convergence criterion as the change in the cluster center between two adjacent iterations being less than 0.001, thus defining a clear boundary for the clustering process; S32, using the Euclidean distance calculation method, calculating the distance value between each composite feature data and each cluster center one by one; and combining an adaptive adjustment coefficient with a value range of 0.8 to 1.2, determining the distance of each data point according to the principle of minimum distance. Based on the clusters they belong to, the first round of clustering is completed; S33 calculates the internal compactness (average distance between data within a cluster and the center) and inter-cluster separation (average distance between different cluster centers) of each cluster, and dynamically adjusts the cluster centers and adaptive adjustment coefficients based on the ratio of the two (the target ratio is 3:1) to make the clustering results more consistent with the data distribution; S34 repeats the clustering division in S32 and the parameter adjustment process in S33 until the convergence judgment condition is met or the maximum number of iterations is reached, and finally outputs the classification results of 5 modalities. This step-by-step process achieves accurate separation of feature data under different operating modalities through the closed-loop execution of initialization, first round of clustering, parameter adjustment, and iterative convergence, providing targeted classification data for anomaly detection.

[0041] Preferred, such as Figure 4As shown, step S4 includes the following sub-steps: S41, extracting feature parameters from the classification results of each modality, and determining the operating condition adaptation range and discrimination threshold corresponding to each feature parameter by combining the operating standards of the equipment under different operating conditions; S42, inputting the feature parameters into the operating condition adaptation anomaly discrimination model, and completing the adaptation calculation between the feature parameters and the operating condition standards through a preset operating condition adaptation mapping function; S43, comparing and analyzing the model output results with the discrimination threshold to identify abnormal feature parameters that exceed the adaptation range, and recording the specific category and occurrence time of the abnormal parameters; S44, performing correlation analysis on the identified abnormal feature parameters to determine whether the anomaly is caused by a single parameter anomaly or a multi-parameter collaborative anomaly, and forming a complete anomaly screening report.

[0042] Specifically, step S4 includes S41 extracting eight key feature parameters (including core dimensions such as load, energy consumption, and power output) from the modal classification results. Combining this with standard operating data under different equipment operating conditions (three conditions in total: light load, medium load, and heavy load), and by statistically analyzing the 95% confidence interval of historical normal operating data, the adaptation range and discrimination threshold of each feature parameter under different operating conditions are determined to ensure the scientific validity of the threshold setting. S42 inputting the extracted key feature parameters one by one into the operating condition adaptation anomaly discrimination model according to modal category, the model calls a preset operating condition adaptation mapping function. This function sets specific mapping rules for different operating conditions, completes the adaptation calculation between the feature parameters and the corresponding operating condition standards, and converts it into a unified dimension judgment. Indicators; S43 compares and analyzes the judgment indicators output by the model with the preset judgment thresholds one by one to identify abnormal feature parameters that exceed the adaptation range, and records the specific category (such as load abnormality, energy consumption abnormality, etc.) and occurrence time (accurate to the second) of each abnormal parameter to form a preliminary abnormal list; S44 uses the association rule mining method to analyze the occurrence time and change trend correlation of each abnormal parameter in the preliminary abnormal list, and determines whether the abnormality is caused by a single parameter independent abnormality or multiple parameter collaborative abnormality, and finally forms a complete abnormal screening report including abnormality type, associated parameters, and occurrence time. This step-by-step process achieves accurate identification and classification of abnormal features through feature extraction, working condition adaptation, threshold comparison, and association analysis.

[0043] Preferred, such as Figure 5As shown, S5 includes the following sub-steps: S51, calling the historical data storage module of the intelligent judgment platform for operating status, extracting historical operating status data and energy efficiency anomaly records of the equipment within the past three months, and establishing a comparison database of historical data and real-time data; S52, matching and analyzing the anomaly screening report output by S4 with the historical data in the comparison database, and finding similar anomaly cases and corresponding processing results; S53, combining the equipment load rate, ambient temperature, and power supply voltage parameters in the real-time operating condition information, predicting the development trend of anomaly characteristics, and assessing the degree of energy efficiency impact that anomalies may cause; S54, comprehensively considering the anomaly matching results and trend prediction conclusions, using a hierarchical evaluation method to determine the energy efficiency anomaly level, and storing the level classification results in association with the anomaly characteristic data.

[0044] Specifically, step S5 includes S51 calling the platform's historical data storage module. This module adopts a distributed storage architecture, extracting historical operating status data (cumulative data volume of no less than 1 million records) and past energy efficiency anomaly records (a total of 120 valid records) from the equipment within the past three months. These records are categorized and stored according to modal type and operating condition, establishing a comparison database of historical and real-time data. The database supports rapid retrieval based on feature parameters. S52 uses the anomaly feature parameters and modal type information from the anomaly screening report output in S4 as search criteria to find historical anomaly cases with a similarity higher than 85% in the comparison database. The processing results and anomaly development patterns of these historical cases are extracted to form a case reference library. S53 collects key parameters from real-time operating condition information, such as equipment load rate, ambient temperature, and power supply voltage, and combines this with the equipment's physical operating model. The system predicts the development trend of abnormal characteristics and assesses the potential energy efficiency decline (accurate to a percentage) and equipment damage caused by the abnormality within the next hour. S54 adopts a hierarchical assessment method, setting three levels of assessment indicators based on the scope of the abnormality's impact, its development speed, and the degree of energy efficiency loss. Each indicator is divided into five scoring levels according to its severity. A comprehensive score is calculated by weighted summation (with weights of 0.4, 0.3, and 0.3, respectively). Based on the score, the energy efficiency abnormality level is divided into minor abnormality (score 1-2 points), moderate abnormality (score 2-4 points), and severe abnormality (score 4-5 points). The level classification results are stored in association with abnormal characteristic data and historical case reference information. This step-by-step process achieves in-depth analysis of abnormalities through historical data retrieval, case matching, trend prediction, and level assessment, providing accurate basis for the generation of early warning signals.

[0045] An energy efficiency anomaly early warning method based on equipment operating status identification is disclosed. This method is implemented through an energy efficiency anomaly early warning system based on equipment operating status identification, comprising: a multi-dimensional data acquisition unit for equipment operating status, a time-series state data differential feature extraction unit, a composite feature time-series modality adaptive clustering unit, a working condition-adaptive multi-dimensional anomaly discrimination unit, an intelligent comprehensive judgment unit for operating status, and an energy efficiency anomaly graded early warning signal generation unit. The multi-dimensional data acquisition unit collects various time-series state data through a multi-dimensional sensor array and transmits the data to the time-series state data differential feature extraction unit. The time-series state data differential feature extraction unit processes the data using an equipment state time-series differential sensing algorithm. Subsequently, the composite feature set is sent to the composite feature temporal modality adaptive clustering unit; the composite feature temporal modality adaptive clustering unit completes modality classification through the temporal modality adaptive clustering algorithm, and transmits the classification results to the operating condition adaptation multi-dimensional anomaly discrimination unit; the operating condition adaptation multi-dimensional anomaly discrimination unit uses the operating condition adaptation anomaly discrimination model to perform anomaly screening, and sends the screening report to the operating status intelligent comprehensive judgment unit; the operating status intelligent comprehensive judgment unit integrates the data and determines the anomaly level, and transmits the level information to the energy efficiency anomaly classification early warning signal generation unit; the energy efficiency anomaly classification early warning signal generation unit generates an early warning signal carrying detailed information according to the anomaly level, and provides early warning of equipment energy efficiency anomalies.

[0046] An energy efficiency anomaly early warning method based on equipment operation status identification is proposed. This method collects comprehensive data on equipment operation from multiple dimensions using a multi-dimensional sensor array, overcoming the limitations of single data collection and providing comprehensive data support for accurate analysis. Through a multi-algorithm collaborative working mode, it forms a complete technical chain covering data processing, feature analysis, and anomaly identification, from time-series data differential feature extraction to modal adaptive clustering and then to operating condition-adaptive anomaly discrimination. By integrating historical and real-time data through an intelligent operation status assessment platform, it achieves accurate anomaly level classification and targeted early warning signal generation, ensuring the comprehensiveness of anomaly identification and improving the accuracy and practicality of early warnings. Furthermore, the innovative design of hardware configuration and software parameters further enhances the method's adaptability and operational stability in complex industrial scenarios.

[0047] This method addresses the issue of insufficient feature extraction accuracy by abandoning single-dimensional and simplistic analysis methods. It deeply mines the temporal correlation and subtle fluctuation features of data through a device status temporal differential perception algorithm, and combines it with a temporal modality adaptive clustering algorithm to achieve accurate feature classification, ensuring that no abnormal features are missed and effectively avoiding false negatives. Addressing the deficiency of poor adaptability to operating conditions, the operating condition-adaptive anomaly discrimination model no longer relies on fixed parameters but dynamically adjusts the discrimination logic based on real-time operating conditions. This fully adapts to the differences in energy efficiency standards under different operating conditions, significantly reducing warning errors during operating condition switching. Furthermore, the collaborative analysis of multiple algorithms and the comprehensive analytical capabilities of the intelligent judgment platform further enhance the method's adaptability to complex operating scenarios, completely solving the problems of insufficient warning accuracy and scenario adaptability in existing technologies.

[0048] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0049] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of energy efficiency anomalies based on equipment operating status identification, characterized in that, Includes the following steps: S1. Collect time-series status data related to energy efficiency during equipment operation using a multi-dimensional sensor array. This time-series status data includes equipment load characteristic data, power output characteristic data, energy consumption characteristic data, and environmental impact characteristic data. S2. Use a time-series differential sensing algorithm to dynamically extract features from the collected time-series status data. By constructing a differential mapping relationship between adjacent time points of the time-series data, subtle fluctuations and trend changes hidden in the data sequence are captured. S3. Use a time-series modal adaptive clustering algorithm to classify the dynamic features extracted in S2. Based on feature similarity and temporal correlation, construct adaptive clustering criteria to divide feature data under different operating modes. S4. Input the modal feature data classified in S3 into a condition-adaptive anomaly discrimination model. Combine this with preset operating condition parameters of the equipment to construct a multi-dimensional discrimination boundary and perform anomaly screening on the feature data under each mode. S5. Integrate the anomaly screening results from S4 through an intelligent operating status analysis platform. Combine historical operating status data and real-time operating condition information for comprehensive analysis to determine the energy efficiency anomaly level corresponding to the abnormal features. S6. Generate a targeted early warning signal based on the energy efficiency anomaly level determined in S5. The early warning signal carries information on the source of the anomaly, the trend of the anomaly, and related operating parameters.

2. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, The expression for the device status timing differential sensing algorithm is: ,in The device timing status data at time t. for Time-series status data of the device. The time interval for time-series data acquisition. This is the difference time lag factor. The dynamic weighting coefficients at time t are used; the feature enhancement formula is calculated as follows: ,in Let n be the enhanced feature data at time t, and n be the number of feature fusion dimensions. Let be the fusion weight of the i-th feature. Let be the nonlinear mapping function for the i-th feature.

3. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, The expression for the temporal modality adaptive clustering algorithm is: ,in For the k-th cluster, For the feature data to be clustered, Let be the dynamic center of the k-th cluster at time t. Operators for Euclidean distance calculation Let be the adaptive adjustment coefficient for the k-th cluster; the cluster center update formula is calculated as follows: ,in For the k-th cluster in The center is updated in real time. The update rate coefficient is centered. Let be the amount of feature data included in the k-th cluster at time t. t is the clustering adjustment factor at time t.

4. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, The expression for the working condition adaptation anomaly discrimination model is: ,in This is the result of the anomaly detection. Here, M is the Sigmoid activation function, and M is the number of discriminant feature dimensions. Let be the weight coefficient of the m-th discriminative feature. The working condition adaptation mapping function is for the m-th discriminative feature. For the m-th modal feature data, is the preset standard parameter for the m-th feature, and b is the model bias term.

5. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, The multi-dimensional sensing array includes: a fiber optic strain sensor, a piezoelectric vibration sensor, an infrared temperature sensor, an ultrasonic flow sensor, and a Hall current sensor. Each sensor establishes a communication connection with an edge computing gateway via an industrial Ethernet bus. The edge computing gateway has a built-in FPGA chip for real-time reception and forwarding of time-series data. In the device status time-series differential sensing algorithm, the time lag factor ranges from 1 to 5, and the dynamic weight coefficient is dynamically adjusted according to the device operating condition level. The initial number of clusters in the time-series modality adaptive clustering algorithm is set to 3 to 8 based on the statistical results of the device's historical operating modes. The number of discriminant feature dimensions in the operating condition adaptation anomaly discrimination model is set to 6 to 12, and the weight coefficient is obtained through iterative optimization using the gradient descent method. The intelligent judgment platform for operating status adopts a distributed computing architecture, including 4 computing nodes, each configured with a 16-core processor and 64GB of memory.

6. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, S2 includes the following sub-steps: S21, selecting a continuous data segment of a preset length from the time-series state data, determining the specific value of the differential time lag factor, and establishing a calculation rule for the dynamic weight coefficient based on the equipment operating rate and data acquisition frequency; S22, pairing adjacent time-time data of the selected data segment, calculating the time-series differential data of each time-time according to the differential expression, and retaining the data sequence after differential calculation and the corresponding timestamp information; S23, dynamically identifying abnormal fluctuation points in the differential data sequence, determining the fluctuation threshold through local data distribution characteristics, and filtering out differential data exceeding the threshold range; S24, associating and mapping the filtered differential data with the original time-series data to construct a composite feature set including the original features and differential features, providing data support for subsequent modality clustering.

7. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, S3 includes the following steps: S31, based on the feature dimension and data distribution characteristics of the composite feature set, initialize the number of clusters and the initial center of each cluster, and set the number of clustering iterations and convergence criteria; S32, calculate the distance between each composite feature data and the center of each cluster, and determine the cluster to which the data belongs by combining the adaptive adjustment coefficient, thus completing the first round of clustering; S33, calculate the internal compactness and inter-cluster separation of each cluster based on the results of the first round of clustering, and dynamically adjust the cluster center and adaptive adjustment coefficient based on the balance between the two; S34, repeat the clustering and parameter adjustment process until the convergence criteria are met, and output the final modality classification result to ensure the separation of feature data under different operating modalities.

8. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, The S4 includes the following sub-steps: S41, extract the feature parameters from each modality classification result, and determine the working condition adaptation range and discrimination threshold corresponding to each feature parameter by combining the operating standards of the equipment under different working conditions. S42, Input the feature parameters into the working condition adaptation anomaly discrimination model, and complete the adaptation calculation between the feature parameters and the working condition standard through the preset working condition adaptation mapping function; S43, Compare and analyze the model output results with the discrimination threshold to identify abnormal feature parameters that exceed the adaptation range, and record the specific category and occurrence time of the abnormal parameters; S44, Perform correlation analysis on the identified abnormal feature parameters to determine whether the anomaly is caused by a single parameter anomaly or a multi-parameter collaborative anomaly, and form a complete anomaly screening report.

9. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, S5 includes the following steps: S51, calling the historical data storage module of the intelligent judgment platform for operating status, extracting historical operating status data and energy efficiency anomaly records of the equipment within the past three months, and establishing a comparison database of historical data and real-time data; S52, matching and analyzing the anomaly screening report output by S4 with the historical data in the comparison database, finding similar anomaly cases and corresponding processing results; S53, combining the equipment load rate, ambient temperature, and power supply voltage parameters in the real-time operating condition information, predicting the development trend of anomaly characteristics, and assessing the degree of energy efficiency impact that anomalies may cause; S54. Based on the comprehensive anomaly matching results and trend prediction conclusions, a hierarchical evaluation method is used to determine the energy efficiency anomaly level, and the level classification results are stored in association with the anomaly characteristic data.

10. The energy efficiency anomaly early warning method based on equipment operating status identification according to claim 1, characterized in that, This method is implemented through an energy efficiency anomaly early warning system based on equipment operating status recognition, comprising: a multi-dimensional data acquisition unit for equipment operating status, a time-series state data differential feature extraction unit, a composite feature time-series modal adaptive clustering unit, a working condition-adaptive multi-dimensional anomaly discrimination unit, an intelligent comprehensive judgment unit for operating status, and an energy efficiency anomaly graded early warning signal generation unit; the multi-dimensional data acquisition unit for equipment operating status collects various time-series state data through a multi-dimensional sensor array and transmits the data to the time-series state data differential feature extraction unit; the time-series state data differential feature extraction unit processes the data using an equipment state time-series differential perception algorithm and sends the composite feature set... The data is sent to the composite feature temporal modality adaptive clustering unit; the composite feature temporal modality adaptive clustering unit completes modality classification through the temporal modality adaptive clustering algorithm and transmits the classification results to the operating condition adaptation multi-dimensional anomaly discrimination unit; the operating condition adaptation multi-dimensional anomaly discrimination unit uses the operating condition adaptation anomaly discrimination model to perform anomaly screening and sends the screening report to the intelligent comprehensive judgment unit of operating status; the intelligent comprehensive judgment unit of operating status integrates the data and determines the anomaly level, and transmits the level information to the energy efficiency anomaly classification early warning signal generation unit; the energy efficiency anomaly classification early warning signal generation unit generates an early warning signal with detailed information according to the anomaly level to provide early warning of equipment energy efficiency anomalies.

Images