Incremental updating method of power efficiency fingerprint library for industrial equipment

Abstract

The application relates to an incremental updating method of an industrial equipment electric energy efficiency fingerprint library, wherein through real-time monitoring and updating early warning of fingerprint library core feature index changes by a sliding window monitoring module, active sensing and timely response of feature drift can be realized, so that the timeliness of the fingerprint library is guaranteed; through a Kalman filtering algorithm combined with an incremental learning strategy, dynamic incremental optimization correction of historical feature data in the fingerprint library can be realized; through input of standardized multi-dimensional fingerprint features into a graph convolution network, a feature correlation graph is constructed, so that semantic correlation and deep representation between electrical features and non-electrical features can be realized, the fingerprint library not only stores isolated feature values, but also retains correlation structure information between features, and the richness and accuracy of feature expression are improved; and the application has the advantages of high accuracy, high efficiency and strong applicability.

Description

Technical Field

[0001] This invention belongs to the field of energy efficiency monitoring technology for industrial equipment, and specifically relates to an incremental update method for an energy efficiency fingerprint database for industrial equipment. Background Technology

[0002] An industrial equipment energy efficiency fingerprint database is a feature database constructed by collecting electrical and non-electrical characteristics during equipment operation. This database is used for equipment energy efficiency assessment, fault diagnosis, and operational status identification. With the rapid development of the Industrial Internet and intelligent manufacturing, the demand for refined monitoring and management of industrial equipment energy efficiency is increasingly urgent, making the construction of a high-precision, high-time-efficiency energy efficiency fingerprint database a key focus in the industry. Currently, existing methods for constructing and updating industrial equipment fingerprint databases have the following shortcomings: First, insufficient fusion of multi-source heterogeneous data, neglecting the correlation between non-electrical characteristics such as vibration and temperature and electrical characteristics, leading to one-sided fingerprint representation; second, the fingerprint database lacks a dynamic update mechanism, making it difficult to adapt to changes in equipment operating conditions and the addition of new equipment, resulting in significant feature drift; third, existing fingerprint database update methods mostly adopt a full update approach, which is costly, inefficient, and renders the fingerprint database unusable during the update process, failing to meet the real-time requirements of industrial scenarios. Therefore, to solve these problems, it is necessary to develop an accurate, efficient, and widely applicable incremental update method for an industrial equipment energy efficiency fingerprint database. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide an incremental update method for an accurate, efficient and widely applicable energy efficiency fingerprint database for industrial equipment.

[0004] The objective of this invention is achieved as follows: a method for incrementally updating an energy efficiency fingerprint database for industrial equipment, comprising the following steps: Step 1: Collect multi-source heterogeneous data of industrial equipment in multi-physics scenarios, and then preprocess the collected data. The preprocessing specifically includes: using wavelet denoising algorithm to filter out noise from the collected data, using event detection algorithm to identify the time of equipment operation state switching, using mutual information entropy method to filter high-dimensional features, remove redundant features, and extract effective feature vectors. Step 2: Extract features from the preprocessed multi-source heterogeneous data to construct a multi-dimensional fingerprint feature system containing steady-state and dynamic features. Then, standardize the extracted multi-dimensional fingerprint features, specifically: normalize feature data from different sources and with different dimensions; align electrical and non-electrical features in time and space; thereby establishing a unified feature data format specification to ensure the comparability and consistency of features across devices and categories. Step 3: Input the standardized multidimensional fingerprint features into a graph convolutional network to construct a feature association graph. Then, extract spatial features from the feature association graph through the graph convolutional layer in the graph convolutional network. This achieves high-dimensional effective feature screening and deep association modeling, while simultaneously building an energy efficiency fingerprint database for industrial equipment. Step 4: Deploy the sliding window monitoring module to monitor changes in the core feature indicators of the fingerprint database in real time. When the core feature indicators deviate from the preset threshold range, an update warning will be triggered. Step 5: Based on the Kalman filter algorithm combined with the incremental learning strategy, the historical feature data in the fingerprint database is dynamically and incrementally optimized and corrected, and the latest modeled and optimized feature vector is entered into the fingerprint database to complete the incremental update; Step 6: Incrementally train and optimize the recognition and evaluation models associated with the fingerprint database to keep the model parameters synchronized with the updated fingerprint features, thereby continuously improving the accuracy of energy efficiency assessment and fault diagnosis.

[0005] Furthermore, the multi-source heterogeneous data in step 1 includes steady-state electrical characteristics, dynamic transient electrical characteristics, and equipment non-electrical condition monitoring characteristics.

[0006] Furthermore, the core feature indicators of the fingerprint database in step 4 include: current harmonic distortion rate, reactive power fluctuation coefficient, voltage sag frequency, and power factor change rate.

[0007] Furthermore, in step 2, the steady-state characteristics include the effective value of voltage, the effective value of current, active power, reactive power, power factor, and current harmonic distortion rate; the dynamic characteristics include the equipment starting inrush current, transient response time, and power change rate.

[0008] Furthermore, in step 3, within the feature association graph: nodes represent individual fingerprint features, and edges represent semantic associations between features.

[0009] The beneficial effects of this invention are as follows: This invention achieves proactive perception and timely response to feature drift by using a sliding window monitoring module to monitor and update the changes in the core feature indicators of the fingerprint database in real time, thus effectively ensuring the timeliness of the fingerprint database. This invention uses a Kalman filter algorithm combined with an incremental learning strategy to dynamically and incrementally optimize and correct historical feature data in the fingerprint database. In practical use, only the data in the database needs to be corrected based on the latest modeled feature values, without the need to recollect all data and reconstruct the entire fingerprint database. This reduces update costs and time overhead, and the update process does not affect the normal use of the fingerprint database. This invention constructs a feature association graph by inputting standardized multidimensional fingerprint features into a graph convolutional network. This enables semantic association and deep representation between electrical and non-electrical features, allowing the fingerprint database to not only store isolated feature values ​​but also retain the structural information of the association between features, thus improving the richness and accuracy of feature representation. Furthermore, by incrementally training and optimizing the association model after incremental updates, this invention allows the model and the fingerprint database to evolve synchronously, thereby continuously improving the accuracy of energy efficiency assessment and fault diagnosis. In summary, this invention has the advantages of being accurate, efficient, and highly applicable. Detailed Implementation

[0010] The present invention will now be further described.

[0011] Example: An incremental update method for an energy efficiency fingerprint database for industrial equipment, comprising the following steps: Step 1: Collect multi-source heterogeneous data of industrial equipment in a multi-physics scenario, and then preprocess the collected data. The preprocessing specifically includes: using a wavelet denoising algorithm to filter out noise from the collected data, using an event detection algorithm to identify the switching time of equipment operating status, and using the mutual information entropy method to filter high-dimensional features, remove redundant features, and extract effective feature vectors. The multi-source heterogeneous data includes steady-state electrical features, dynamic transient electrical features, and equipment non-electrical condition monitoring features. Step 2: Extract features from the preprocessed multi-source heterogeneous data to construct a multi-dimensional fingerprint feature system containing steady-state and dynamic features. Steady-state features include RMS voltage, RMS current, active power, reactive power, power factor, and current harmonic distortion rate; dynamic features include equipment startup inrush current, transient response time, and power change rate. Then, standardize the extracted multi-dimensional fingerprint features, specifically by normalizing feature data from different sources and with different dimensions; and aligning electrical and non-electrical features in time and space. This establishes a unified feature data format specification, ensuring comparability and consistency of features across devices and categories. Step 3: Input the standardized multidimensional fingerprint features into a graph convolutional network to construct a feature association graph. Then, extract spatial features from the feature association graph through the graph convolutional layer in the graph convolutional network. This achieves high-dimensional effective feature selection and deep association modeling, while simultaneously building an energy efficiency fingerprint database for industrial equipment. In the feature association graph, nodes represent individual fingerprint features, and edges represent semantic relationships between features. Step 4: Deploy the sliding window monitoring module to monitor changes in the core feature indicators of the fingerprint database in real time. When the core feature indicators deviate from the preset threshold range, an update warning is triggered. The core feature indicators of the fingerprint database include: current harmonic distortion rate, reactive power fluctuation coefficient, voltage sag frequency, and power factor change rate. Step 5: Based on the Kalman filter algorithm combined with the incremental learning strategy, the historical feature data in the fingerprint database is dynamically and incrementally optimized and corrected, and the latest modeled and optimized feature vector is entered into the fingerprint database to complete the incremental update; Step 6: Incrementally train and optimize the recognition and evaluation models associated with the fingerprint database to keep the model parameters synchronized with the updated fingerprint features, thereby continuously improving the accuracy of energy efficiency assessment and fault diagnosis.

[0012] In use, this invention first collects multi-source heterogeneous data from industrial equipment in a multi-physical environment. Then, it preprocesses and extracts features from the collected data to construct a multi-dimensional fingerprint feature system containing both steady-state and dynamic features. Next, it standardizes the extracted multi-dimensional fingerprint features, normalizing feature data from different sources and with different dimensions, and aligning electrical and non-electrical features temporally and spatially to establish a unified feature data format standard, ensuring comparability and consistency of features across devices and categories. Then, it inputs the standardized multi-dimensional fingerprint features into a graph convolutional network to construct a feature association graph. Spatial feature extraction is then performed on the feature association graph through graph convolutional layers in the graph convolutional network, achieving high-dimensional effective feature selection and deep association modeling while simultaneously building an industrial equipment energy efficiency fingerprint database. In this process, by inputting the standardized multi-dimensional fingerprint features into a graph convolutional network to construct the feature association graph, semantic association and deep representation between electrical and non-electrical features are achieved. This ensures that the fingerprint database not only stores isolated feature values ​​but also retains the structural information of the relationships between features, enhancing the richness of feature expression. The invention ensures the timeliness and accuracy of fingerprint databases. Finally, a sliding window monitoring module is deployed to monitor changes in core feature indicators of the fingerprint database in real time. When a core feature indicator deviates from a preset threshold range, an update warning is triggered. This method enables proactive perception and timely response to feature drift, ensuring the timeliness of the fingerprint database. After completing the above operations, based on the Kalman filter algorithm combined with an incremental learning strategy, the historical feature data in the fingerprint database is dynamically and incrementally optimized and corrected. The latest optimized feature vector is then entered into the fingerprint database, completing the incremental update. Finally, after the incremental update, the associated model is simultaneously trained and optimized, ensuring that the model and the fingerprint database evolve synchronously, thereby continuously improving the accuracy of energy efficiency assessment and fault diagnosis. This invention uses this method, combining the Kalman filter algorithm with an incremental learning strategy, to dynamically and incrementally optimize and correct historical feature data in the fingerprint database. In specific implementation, only the data in the database needs to be corrected based on the latest modeled feature values, without needing to re-collect all data and fully reconstruct the fingerprint database, reducing update costs and time overhead. Furthermore, the update process does not affect the normal use of the fingerprint database. In summary, this invention has the advantages of accuracy, efficiency, and strong applicability.

[0013] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for incrementally updating an energy efficiency fingerprint database for industrial equipment, characterized in that, Includes the following steps: Step 1: Collect multi-source heterogeneous data of industrial equipment in multi-physics scenarios, and then preprocess the collected data. The preprocessing specifically includes: using wavelet denoising algorithm to filter out noise from the collected data, using event detection algorithm to identify the time of equipment operation state switching, using mutual information entropy method to filter high-dimensional features, remove redundant features, and extract effective feature vectors. Step 2: Extract features from the preprocessed multi-source heterogeneous data to construct a multi-dimensional fingerprint feature system containing steady-state and dynamic features. Then, standardize the extracted multi-dimensional fingerprint features, specifically: normalize feature data from different sources and with different dimensions; align electrical and non-electrical features in time and space; thereby establishing a unified feature data format specification to ensure the comparability and consistency of features across devices and categories. Step 3: Input the standardized multidimensional fingerprint features into a graph convolutional network to construct a feature association graph. Then, extract spatial features from the feature association graph through the graph convolutional layer in the graph convolutional network. This achieves high-dimensional effective feature screening and deep association modeling, while simultaneously building an energy efficiency fingerprint database for industrial equipment. Step 4: Deploy the sliding window monitoring module to monitor changes in the core feature indicators of the fingerprint database in real time. When the core feature indicators deviate from the preset threshold range, an update warning will be triggered. Step 5: Based on the Kalman filter algorithm combined with the incremental learning strategy, the historical feature data in the fingerprint database is dynamically and incrementally optimized and corrected, and the latest modeled and optimized feature vector is entered into the fingerprint database to complete the incremental update; Step 6: Incrementally train and optimize the recognition and evaluation models associated with the fingerprint database to keep the model parameters synchronized with the updated fingerprint features, thereby continuously improving the accuracy of energy efficiency assessment and fault diagnosis.

2. The incremental update method for an energy efficiency fingerprint database for industrial equipment as described in claim 1, characterized in that: The multi-source heterogeneous data in step 1 includes steady-state electrical characteristics, dynamic transient electrical characteristics, and equipment non-electrical condition monitoring characteristics.

3. The incremental update method for an energy efficiency fingerprint database for industrial equipment as described in claim 1, characterized in that: The core feature indicators of the fingerprint database in step 4 include: current harmonic distortion rate, reactive power fluctuation coefficient, voltage sag frequency, and power factor change rate.

4. The incremental update method for an energy efficiency fingerprint database for industrial equipment as described in claim 1, characterized in that: In step 2, the steady-state characteristics include the effective value of voltage, the effective value of current, active power, reactive power, power factor, and current harmonic distortion rate; the dynamic characteristics include the equipment starting inrush current, transient response time, and power change rate.

5. The incremental update method for an industrial equipment energy efficiency fingerprint database as described in claim 1, characterized in that: In step 3, within the feature association graph: nodes represent individual fingerprint features, and edges represent semantic associations between features.