Current sensor error compensation method and system

Abstract

The invention discloses a current sensor error compensation method and system, and the method comprises the steps: firstly obtaining a current sensor operation environment, dividing time windows, extracting the environment coding characteristics of each window, obtaining contribution degree coefficients through dynamic evaluation, and fusing the contribution degree coefficients into environment characteristics; and meanwhile, electric signal time sequence fragments are obtained and fused into joint features, and circuit state features are obtained through feature fusion network reconstruction. Environment and circuit state characteristics are aggregated, a compensation strategy module is matched according to the environment and circuit state characteristics, compensation parameter fusion operation is executed in the two modules to obtain a first compensation characteristic and a second compensation characteristic, finally, a current error compensation result is determined in combination with an error type identifier, and high-precision current sensor error compensation is achieved.

Description

Technical Field

[0001] The present invention relates to the field of automated detection, and more particularly, to a method and system for compensating the error of a current sensor. Background Art

[0002] Current sensors are widely used in many fields such as industrial production and power systems, and their measurement accuracy directly affects the stability and reliability of the system. However, the operating environment (such as temperature, electromagnetic interference, etc.) and the state of the circuit itself can cause errors in the current sensor. Existing error compensation methods often do not fully consider the dynamic changes and mutual influences of the environment and the circuit state, and the compensation accuracy is limited. Therefore, there is an urgent need for a method for compensating the error of a current sensor that can comprehensively consider the dynamic changes of multiple factors to improve the measurement accuracy. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for compensating the error of a current sensor.

[0004] In a first aspect, an embodiment of the present invention provides a method for compensating the error of a current sensor, including: Obtain the operating environment corresponding to the current sensor, and divide the operating environment into multiple time windows; Extract features from the multiple time windows to obtain the environmental coding features corresponding to the multiple time windows; Dynamically evaluate the environmental coding features corresponding to the multiple time windows to obtain the contribution coefficient corresponding to the multiple time windows; Dynamically fuse the environmental coding features corresponding to the multiple time windows and the contribution coefficients corresponding to the multiple time windows to obtain the environmental features corresponding to the operating environment; Obtain multiple electro-signal time series segments corresponding to the current sensor, and fuse the multiple electro-signal time series segments into an electro-signal joint feature; Input the electro-signal joint feature into a feature fusion network, and perform a feature reconstruction operation on the electro-signal joint feature based on the feature fusion network to obtain the circuit state feature corresponding to the current sensor; Obtain an error type identifier, and determine the current error compensation result corresponding to the current sensor based on the error type identifier, the environmental feature, and the circuit state feature.

[0005] Further, the obtaining an error type identifier, and determining the current error compensation result corresponding to the current sensor based on the error type identifier, the environmental feature, and the circuit state feature includes: Aggregate the environmental feature and the circuit state feature within the same feature domain to obtain a first aggregated feature. Based on the first aggregated feature, match a first compensation strategy module for the environmental feature among multiple compensation strategy modules; the multiple compensation strategy modules are configured with different compensation strategies to adjust the features of multiple data sources. In the first compensation strategy module, perform a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature. Aggregate the circuit state feature and the first compensation feature within the same feature domain to obtain a second aggregated feature. Based on the second aggregated feature, match a second compensation strategy module for the circuit state feature among the multiple compensation strategy modules. In the second compensation strategy module, perform a compensation parameter fusion operation on the circuit state feature based on the first compensation feature to obtain a second compensation feature. Obtain an error type identifier, and based on the error type identifier, the first compensation feature, and the second compensation feature, determine the current error compensation result corresponding to the current sensor.

[0006] Further, the aggregating the environmental feature and the circuit state feature within the same feature domain to obtain a first aggregated feature includes: Input the environmental feature and the circuit state feature into a dynamic fusion module, and based on a first feature projection unit in the dynamic fusion module, perform feature alignment on the environmental feature to obtain an environmental compensation feature. Based on a second feature projection unit in the dynamic fusion module, perform feature alignment on the circuit state feature to obtain a circuit compensation feature; the environmental compensation feature and the circuit compensation feature belong to the same feature domain. Obtain a first effective value corresponding to the environmental compensation feature, and based on the first effective value, perform calibration processing on the environmental compensation feature to obtain an environmental calibration feature. Perform dynamic gain adjustment on the environmental calibration feature to obtain an environmental gain feature, and perform smoothing processing on the environmental gain feature to obtain an environmental smoothing feature. Obtain a second effective value corresponding to the circuit compensation feature, and based on the second effective value, perform calibration processing on the circuit compensation feature to obtain a circuit calibration feature. Perform dynamic gain adjustment on the circuit calibration feature to obtain a circuit gain feature, and perform smoothing processing on the circuit gain feature to obtain a circuit smoothing feature. Merge the environmental smoothing feature and the circuit smoothing feature to obtain a comprehensive compensation feature, and perform a feature reconstruction operation on the comprehensive compensation feature based on the compensation parameter matrix corresponding to the parameter fusion module in the dynamic fusion module to obtain the first aggregated feature.

[0007] Further, matching the first compensation strategy module for the environmental feature among multiple compensation strategy modules based on the first aggregated feature includes: Perform a strategy evaluation on the first aggregated feature based on the strategy decision module in the dynamic fusion module to obtain the fitness coefficients corresponding to multiple compensation strategy modules; Determine the compensation strategy module corresponding to the maximum fitness coefficient as the first compensation strategy module matched for the environmental feature.

[0008] Further, the first compensation strategy module includes a dynamic weight fuser; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature includes: Perform channel combination on the environmental feature and the circuit state feature to obtain a first combined feature; Based on the transformation compensation parameter matrix corresponding to the dynamic weight fuser, map the first combined feature to a first environmental correlation matrix, a first circuit correlation matrix, and a first compensation parameter matrix; Based on the multiplication result between the first environmental correlation matrix and the environmental feature, and the multiplication result between the first environmental correlation matrix and the circuit state feature, construct a first coupling parameter matrix; Based on the dimension conversion operation result of the multiplication result between the first circuit correlation matrix and the environmental feature, and the dimension conversion operation result of the multiplication result between the first circuit correlation matrix and the circuit state feature, construct a second coupling parameter matrix; Based on the multiplication result between the first compensation parameter matrix and the environmental feature, and the multiplication result between the first compensation parameter matrix and the circuit state feature, construct a third coupling parameter matrix; Determine the multiplication result between the first coupling parameter matrix and the second coupling parameter matrix as a fourth coupling parameter matrix, and obtain the total number of channels of the environmental feature; Perform a normalization ratio calculation on the fourth coupling parameter matrix and the normalization scaling factor of the total number of channels to obtain a cross-source dynamic weight matrix, and determine the multiplication result between the cross-source dynamic weight matrix and the third coupling parameter matrix as a first dynamic correlation feature; Determine the first X groups of channel features in the first dynamic association feature as the first compensation feature; X is the number of main channels corresponding to the environmental feature.

[0009] Furthermore, the first compensation strategy module includes a dynamic weight integrator; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature includes: Based on the dynamic weight integrator, map the environmental feature to a second environmental association matrix, map the circuit state feature to a second circuit association matrix and a second compensation parameter matrix; Perform a matrix multiplication operation on the dimensionality conversion result of the multiplication result between the second environmental association matrix and the environmental feature and the multiplication result between the second circuit association matrix and the circuit state feature to obtain a fifth coupling parameter matrix, and obtain the total number of channels of the environmental feature; Perform a normalization ratio calculation on the fifth coupling parameter matrix and the normalization scaling factor of the total number of channels to obtain a multi-source contribution coefficient matrix, and perform a matrix multiplication operation on the multi-source contribution coefficient matrix and the multiplication result between the second compensation parameter matrix and the circuit state feature to obtain a multi-source compensation feature; Determine the first X groups of channel features in the multi-source compensation feature as the first compensation feature; X is the number of main channels corresponding to the environmental feature.

[0010] Furthermore, the first compensation strategy module includes a multi-source dynamic integrator; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature includes: Obtain a fusion compression feature, and perform channel combination on the fusion compression feature and the circuit state feature to obtain a second combined feature; the fusion compression feature is used to establish the connection between multi-source features; Based on the first dynamic weight assignment branch in the multi-source dynamic integrator, perform cross-source dynamic association analysis on the second combined feature to obtain a second dynamic association feature, and determine the first Y groups of main channel features in the second dynamic association feature as the target compensation feature; Y is the number of main channels corresponding to the fusion compression feature; Perform channel combination on the environmental feature and the target compensation feature to obtain a third combined feature, and based on the second dynamic weight assignment branch in the multi-source dynamic integrator, perform cross-source dynamic association analysis on the third combined feature to obtain a third dynamic association feature; Determine the first X groups of channel features in the third dynamic association feature as the first compensation feature; X is the number of main channels corresponding to the environmental feature; The first compensation strategy module further includes a main channel fusion unit; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature includes: Obtain the environmental calibration feature corresponding to the environmental feature and the circuit calibration feature corresponding to the circuit state feature; Based on the first dynamic time-series processor in the main channel fusion unit, perform time-series signal modulation on the environmental feature to obtain a first time-series response feature; Based on the second dynamic time-series processor in the main channel fusion unit, perform time-series signal modulation on the circuit state feature to obtain a second time-series response feature; Perform smoothing processing on the second time-series response feature to obtain a time-series smoothing feature, and merge the first time-series response feature and the time-series smoothing feature to obtain a first compensation feature.

[0011] Further, the first compensation strategy module includes a reference signal suppressor; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature includes: Clear the circuit state feature, and use the environmental feature as the first compensation feature output by the reference signal suppressor.

[0012] Further, determining the current error compensation result corresponding to the current sensor based on the error type identifier, the first compensation feature, and the second compensation feature includes: Perform channel combination on the first compensation feature and the second compensation feature to obtain a fourth combined feature, and add the error type identifier at the front end of the fourth combined feature to obtain a combined compensation feature; Perform dynamic evaluation on the combined compensation feature based on the cross-source dynamic weight distribution branch to obtain a dynamic weight fusion parameter, and perform a feature reconstruction operation on the dynamic weight fusion parameter to obtain the current error compensation result corresponding to the current sensor.

[0013] In a second aspect, an embodiment of the present invention provides a server system, including a server, and the server is used to execute the method described in the first aspect.

[0014] Compared with the prior art, the beneficial effects provided by the present invention include: By using a current sensor error compensation method and system disclosed in the present invention, the operating environment of the current sensor is obtained and time windows are divided, the environmental coding features of each window are extracted, and after dynamic evaluation to obtain the contribution coefficient, they are fused into environmental features; at the same time, the time series segments of the electrical signals are obtained and fused into joint features, and the circuit state features are reconstructed through the feature fusion network. The environmental and circuit state features are aggregated, and the compensation strategy module is matched accordingly. The compensation parameter fusion operations are respectively performed in the two modules to obtain the first and second compensation features, and finally, the current error compensation result is determined in combination with the error type identifier, realizing high-precision current sensor error compensation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a schematic flowchart of the steps of the current sensor error compensation method provided by the embodiment of the present invention; Figure 2 It is a schematic block diagram of the structure of the computer device provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0018] The following will describe in detail the specific embodiments of the present invention with reference to the drawings.

[0019] To solve the technical problems in the foregoing background art, Figure 1 It is a flowchart of the current sensor error compensation method provided by the embodiment of the present disclosure. The following will introduce this current sensor error compensation method in detail.

[0020] Step S201, obtain the operating environment corresponding to the current sensor, and divide the operating environment into multiple time windows; Step S202, perform feature extraction on the multiple time windows to obtain the environmental coding features corresponding to the multiple time windows; Step S203: Dynamically evaluate the environmental encoding features corresponding to the multiple time windows to obtain the contribution coefficient corresponding to each of the multiple time windows; Step S204: Dynamically fuse the environmental encoding features corresponding to the multiple time windows and the contribution coefficients corresponding to the multiple time windows to obtain the environmental features corresponding to the operating environment; Step S205: Obtain multiple time-series segments of electrical signals corresponding to the current sensor, and fuse the multiple time-series segments of electrical signals into a combined electrical signal feature; Step S206: Input the combined electrical signal feature into a feature fusion network, and perform a feature reconstruction operation on the combined electrical signal feature based on the feature fusion network to obtain the circuit state feature corresponding to the current sensor; Step S207: Aggregate the environmental features and the circuit state features in the same feature domain to obtain a first aggregated feature, and based on the first aggregated feature, match a first compensation strategy module for the environmental features from multiple compensation strategy modules; the multiple compensation strategy modules are configured with different compensation strategies to adjust the features of multiple data sources; Step S208: In the first compensation strategy module, perform a compensation parameter fusion operation on the environmental features based on the circuit state features to obtain a first compensated feature; Step S209: Aggregate the circuit state features and the first compensated feature in the same feature domain to obtain a second aggregated feature, and based on the second aggregated feature, match a second compensation strategy module for the circuit state features from the multiple compensation strategy modules; Step S210: In the second compensation strategy module, perform a compensation parameter fusion operation on the circuit state features based on the first compensated feature to obtain a second compensated feature; Step S211: Obtain an error type identifier, and based on the error type identifier, the first compensated feature, and the second compensated feature, determine the current error compensation result corresponding to the current sensor.

[0021] In an embodiment of the present invention, exemplarily, in this industrial production park, different production devices may experience different operating environments at different time periods. For example, during the peak production period in the daytime, the ambient temperature around some devices may rise due to the intensive operation of the devices, and at the same time, the electromagnetic interference around will also increase due to the simultaneous operation of many devices; while during the low period at night, the temperature may decrease and the electromagnetic interference will also decrease accordingly. The server obtains the operating environment information corresponding to a certain current sensor, and this information may include various environmental parameters such as temperature, humidity, and electromagnetic intensity. The server divides the operating environment according to the time dimension, for example, taking every 10 minutes as a time window. This is because in the experience of this production park, a 10-minute time interval can not only capture the dynamic changes of environmental parameters but also not make the data volume too large to increase the processing burden. For example, from 8 am to 9 am, the server divides this period of operating environment into 6 time windows of 10 minutes each, and each time window contains the complete environmental parameter information within that time period. For each 10-minute time window, the server will use specific algorithms and models for feature extraction. For example, for the temperature parameter, the server may extract features such as the mean value, maximum value, minimum value of the temperature, and the slope of the temperature change; for the electromagnetic intensity parameter, it may extract features such as the peak value, valley value of the electromagnetic intensity, and the fluctuation frequency within that time period. Taking one of the time windows as an example, the mean value of the temperature within this time window is 30 °C, the maximum value is 32 °C, the minimum value is 28 °C, and the slope of the temperature change is an increase of 1 °C every 10 minutes; the peak value of the electromagnetic intensity is 50 microteslas, the valley value is 30 microteslas, and the fluctuation frequency is 5 times per minute. The server encodes these extracted features to form a unique environmental encoding feature. This encoding feature is like the "fingerprint" of the operating environment of this time window, containing the key information of the environment within that time period for subsequent further analysis and processing. The influence degrees of the environments of different time windows on the current sensor error may be different. The server will dynamically evaluate the environmental encoding features of each time window to determine their contribution degrees to the overall operating environment characteristics. For example, during the peak production period, temperature and electromagnetic interference have a greater impact on the current sensor error, while during the low period, humidity may have a relatively greater impact on the error. The server judges the importance of the environmental encoding features of each time window by analyzing historical data and the current sensor error situation monitored in real time. For example, within the time period from 9 am to 10 am, through analysis, it is found that within the time window from 9:00 to 9:10, due to the startup of a large motor nearby, the electromagnetic interference changes violently, having a greater impact on the current sensor error. Therefore, the contribution degree coefficient of the environmental encoding feature of this time window is relatively high, assumed to be 0.8; while within the time window from 9:20 to 9:30, the environmental parameters are relatively stable, having a smaller impact on the current sensor error, and the contribution degree coefficient may be 0.3.In this way, the server determines a corresponding contribution coefficient for each time window, reflecting its importance to the characteristics of the overall operating environment. The server fuses the environmental encoding features of each time window with their corresponding contribution coefficients. For example, the environmental encoding features of the time window from 9:00 to 9:10 are [30, 32, 28, 1, 50, 30, 5], corresponding to the mean temperature, maximum value, minimum value, temperature change slope, peak electromagnetic intensity, valley value, and fluctuation frequency respectively, and the contribution coefficient is 0.8; the environmental encoding features of the time window from 9:20 to 9:30 are [29, 30, 28, 0.5, 40, 35, 3], and the contribution coefficient is 0.3. The server fuses this information through specific fusion algorithms, such as the weighted average method. The fusion formula is: , where is the th environmental feature value after fusion, such as the mean temperature, peak electromagnetic intensity, etc.; is the serial number of the time window, taking values from 1 to , is the total number of time windows, is the th original feature value of the th The contribution coefficient of each time window, obtained through dynamic evaluation, reflects the importance of this time window to the overall environmental characteristics. Based on this, the fused average temperature feature is: [(30×0.8 + 29×0.3) / (0.8 + 0.3)] ≈ 29.73°C. By performing such fusion operations on all features, the server obtains the environmental characteristics corresponding to the entire operating environment. This environmental characteristic comprehensively considers the influence of the environments of different time windows on the current sensor error. The current sensor will collect electrical signals in real time, and these electrical signals exist in the form of time-series segments. In industrial production, different production processes and equipment operating states will cause changes in electrical signals. For example, when a motor starts, operates stably, and stops, the electrical signals will exhibit different characteristics. The server obtains multiple time-series segments of electrical signals of a certain current sensor over a period of time, and each segment may last from a few milliseconds to several seconds. For example, at the moment when the motor starts, there will be a large peak in the electrical signal, and then it gradually stabilizes. The server fuses these different time-series segments of electrical signals to form a combined electrical signal feature. The fusion method may be to splice these time-series segments in chronological order, or through some signal processing algorithms, such as Fourier transform, etc., to integrate the frequency, phase and other characteristics of different segments to obtain a combined electrical signal feature containing the comprehensive information of the electrical signals in this time period. The server inputs the combined electrical signal feature into a pre-trained feature fusion network. This feature fusion network is similar to an intelligent information processing center. It can identify various patterns and features in the combined electrical signal feature and perform reconstruction. For example, the feature fusion network may identify periodic change features, abnormal fluctuation features, etc. in the electrical signals, and reconstruct features that can reflect the current circuit state of the current sensor based on these features. For example, when there are abnormal high-frequency fluctuations in the electrical signals, the feature fusion network may reconstruct this fluctuation feature into a circuit state feature indicating potential interference in the circuit. Through the feature reconstruction operation of the combined electrical signal feature, the server obtains the circuit state feature that can accurately reflect the current circuit state of the current sensor, providing a key basis for subsequent error compensation. The server first needs to ensure that the environmental feature and the circuit state feature are in the same feature domain for aggregation processing. For example, the server may adjust the environmental feature and the circuit state feature to the same numerical range and feature representation method through some feature mapping and normalization methods. Taking the temperature feature - the value range may be 20 - 40°C and the voltage fluctuation feature in the circuit state feature - the value range may be 100 - 200V as an example, the server will perform normalization processing on the temperature feature so that its value range also becomes 0 - 1, being in the same feature domain as the voltage fluctuation feature after similar processing. Then, the server inputs the environmental feature and the circuit state feature into the dynamic fusion module.In the dynamic fusion module, the first feature projection unit performs feature alignment on the environmental features to obtain the environmental compensation features; the second feature projection unit performs feature alignment on the circuit state features to obtain the circuit compensation features, and the two compensation features are in the same feature domain. Next, the server obtains the first effective value corresponding to the environmental compensation feature, for example, by calculating the effective value of the environmental compensation feature in each dimension, assuming that the first effective value obtained is 0.8. Based on this first effective value, the environmental compensation feature is calibrated, that is, each dimension value of the environmental compensation feature is multiplied by 0.8 to obtain the environmental calibration feature. Then the environmental calibration feature is dynamically gain adjusted, for example, according to a pre-set gain rule, some dimensional values of the environmental calibration feature are enlarged or reduced to obtain the environmental gain feature. Then the environmental gain feature is smoothed to remove possible noise and mutations to obtain the environmental smoothing feature. In the same way, the server obtains the second effective value corresponding to the circuit compensation feature, assuming it is 0.6, based on which the circuit compensation feature is calibrated, dynamically gain adjusted and smoothed to obtain the circuit smoothing feature. Finally, the server merges the environmental smoothing feature and the circuit smoothing feature to obtain the comprehensive compensation feature. Based on the compensation parameter matrix corresponding to the parameter fusion module in the dynamic fusion module, a feature reconstruction operation is performed on the comprehensive compensation feature to obtain a first aggregate feature. Based on this first aggregate feature, the server uses the policy decision module in the dynamic fusion module to perform a policy evaluation on it. The policy decision module analyzes the dimensions and feature patterns of the first aggregate feature, compares them with the fitness of multiple compensation strategy modules, and obtains the fitness coefficients corresponding to multiple compensation strategy modules. For example, for a compensation strategy module with a dynamic weight fuser, the policy decision module finds that its fitness coefficient with the first aggregate feature is 0.7; for another compensation strategy module with a main channel fusion unit, the fitness coefficient is 0.5. The server determines the compensation strategy module corresponding to the largest fitness coefficient as the first compensation strategy module matching the environmental feature. Assume that the first compensation strategy module includes a dynamic weight fuser. The server first performs channel union on the environmental feature and the circuit state feature, that is, splices the two features on the channel dimension to obtain the first joint feature. Based on the transformation compensation parameter matrix corresponding to the dynamic weight fuser, the server maps the first joint feature into a first environmental association matrix, a first circuit association matrix, and a first compensation parameter matrix. For example, the first joint feature is converted into these three matrices through matrix multiplication and specific mapping rules. Then, based on the multiplication results between the first environment association matrix and the environment feature, and the multiplication results between the first environment association matrix and the circuit state feature, a first coupling parameter matrix is constructed. Specifically, each row of the first environment association matrix is multiplied by the corresponding elements of the environment feature and the circuit state feature, respectively, and then these product results are combined into the first coupling parameter matrix according to certain rules.In the same way, based on the dimension conversion operation results of the multiplication results between the first circuit correlation matrix and the environmental features, and the dimension conversion operation results of the multiplication results between the first circuit correlation matrix and the circuit state features, a second coupling parameter matrix is constructed. Here, the dimension conversion operation may be to adjust the matrix dimension of the multiplication result to adapt to subsequent calculations. Based on the multiplication results between the first compensation parameter matrix and the environmental features, and the multiplication results between the first compensation parameter matrix and the circuit state features, a third coupling parameter matrix is constructed. The multiplication result between the first coupling parameter matrix and the second coupling parameter matrix is determined as the fourth coupling parameter matrix. The server obtains the total number of channels of the environmental features, assumed to be 10. The fourth coupling parameter matrix is subjected to a normalization ratio calculation with a normalization scaling factor of the total number of channels, such as 1 / 10, to obtain a cross-source dynamic weight matrix. The multiplication result between the cross-source dynamic weight matrix and the third coupling parameter matrix is determined as the first dynamic correlation feature. Finally, the first X groups of channel features in the first dynamic correlation feature are determined as the first compensation feature. Assuming X is the number of main channels corresponding to the environmental features, such as 5, then the first 5 groups of channel features of the first dynamic correlation feature are selected as the first compensation feature. Similar to the previous aggregation process of environmental features and circuit state features, the server again ensures that the circuit state features and the first compensation feature are in the same feature domain. Through operations such as feature alignment, calibration, gain adjustment, and smoothing, the circuit state features and the first compensation feature are aggregated to obtain a second aggregated feature. For example, after obtaining the respective compensation features of the circuit and the first compensation feature through feature alignment in the same way, and then performing calibration, gain adjustment, and smoothing processing respectively, after merging to obtain a comprehensive feature, based on the compensation parameter matrix of the parameter fusion module, feature reconstruction is performed to obtain the second aggregated feature. Then, based on the second aggregated feature, the server uses the policy decision module to evaluate the fitness of multiple compensation policy modules, and obtains the fitness coefficients corresponding to the multiple compensation policy modules. For example, for a compensation policy module with a multi-source dynamic fusion device, the fitness coefficient is 0.6; for another compensation policy module with a reference signal suppressor, the fitness coefficient is 0.4. The server determines the compensation policy module corresponding to the maximum fitness coefficient as the second compensation policy module that matches the circuit state features. Assume that the second compensation policy module is a multi-source dynamic fusion device. The server obtains a fusion compression feature, which may be obtained by compressing and fusing some intermediate features in the previous processing process and is used to establish the connection between multi-source features. The fusion compression feature and the circuit state features are subjected to channel combination to obtain a second combined feature. Based on the first dynamic weight assignment branch in the multi-source dynamic fusion device, cross-source dynamic correlation analysis is performed on the second combined feature to obtain a second dynamic correlation feature. For example, the first dynamic weight assignment branch will assign different weights according to the correlation and importance between different channel features in the second combined feature, so as to obtain the second dynamic correlation feature.Determine the first Y sets of main channel features in the second dynamic association feature as the target compensation feature. Assume that Y is the number of main channels corresponding to the fusion compression feature. For example, if it is 3, then select the first 3 sets of main channel features as the target compensation feature. Perform channel combination on the environmental feature and the target compensation feature to obtain the third combined feature. Based on the second dynamic weight assignment branch in the multi-source dynamic fusion device, perform cross-source dynamic association analysis on the third combined feature to obtain the third dynamic association feature. Finally, determine the first X sets of channel features in the third dynamic association feature as the second compensation feature. Assume that X is the number of main channels corresponding to the circuit state feature. For example, if it is 4, then select the first 4 sets of channel features of the third dynamic association feature as the second compensation feature. The server obtains the error type identifier from the current sensor or relevant monitoring system. For example, the error types may include temperature drift error, electromagnetic interference error, circuit component aging error, etc., and each error type has a specific identifier. Assume that the currently obtained error type identifier is "temperature drift error". The server performs channel combination on the first compensation feature and the second compensation feature to obtain the fourth combined feature. Add the error type identifier to the front end marker of the fourth combined feature to obtain the combined compensation feature. For example, encode the identifier of "temperature drift error" as [1, 0, 0]. Assume that there are a total of three error types and add it to the front end of the fourth combined feature to form the combined compensation feature. Perform dynamic evaluation on the combined compensation feature based on the cross-source dynamic weight assignment branch, and assign different weights according to the importance of each feature in the combined compensation feature and its correlation with the error type to obtain the dynamic weight fusion parameter. Perform a feature reconstruction operation on the dynamic weight fusion parameter. For example, through specific matrix operations and transformations, convert the dynamic weight fusion parameter into a value that can be directly used to compensate the current sensor error, and finally obtain the current error compensation result corresponding to the current sensor. This compensation result can be used to adjust the output data of the current sensor to make it more accurately reflect the actual current value, thereby ensuring the stable operation of industrial production equipment. Through the above detailed steps and scenario examples, the server can effectively compensate for the error of the current sensor, improve the accuracy of current data, and provide strong support for the stable operation of fields such as industrial production.

[0022] In an embodiment of the present invention, the aggregating process of the environmental feature and the circuit state feature in the same feature domain to obtain the first aggregated feature can be implemented through the following examples.

[0023] Input the environmental feature and the circuit state feature into the dynamic fusion module, and based on the first feature projection unit in the dynamic fusion module, perform feature alignment on the environmental feature to obtain the environmental compensation feature; Based on the second feature projection unit in the dynamic fusion module, feature alignment is performed on the circuit state feature to obtain a circuit compensation feature; the environmental compensation feature and the circuit compensation feature belong to the same feature domain; The environmental compensation feature and the circuit compensation feature are aggregated to obtain a first aggregated feature.

[0024] In an embodiment of the present invention, exemplarily, the server inputs the previously obtained environmental feature and circuit state feature into the dynamic fusion module together. The environmental feature covers comprehensive features such as temperature, humidity, and electromagnetic intensity after processing, and the circuit state feature reflects the current circuit condition of the current sensor, such as features like the fluctuation and frequency of the electrical signal. The first feature projection unit in the dynamic fusion module starts to work and performs feature alignment on the environmental feature. For example, the original range of temperature in the environmental feature is 20 - 40 °C, while the unified feature range set in the module is 0 - 1. The first feature projection unit maps the temperature feature to the 0 - 1 range through linear transformation and other methods, and at the same time processes other environmental features such as humidity and electromagnetic intensity according to corresponding rules to obtain an environmental compensation feature. Then, the second feature projection unit performs feature alignment on the circuit state feature. For example, the amplitude of the electrical signal fluctuation in the circuit state feature may originally be 10 - 100 mV. The second feature projection unit adjusts it to the same range as the environmental compensation feature, such as 0 - 1, according to the set rules, and other circuit state features such as the electrical signal frequency are also processed similarly to obtain a circuit compensation feature. At this time, the environmental compensation feature and the circuit compensation feature belong to the same feature domain. Finally, the server aggregates the environmental compensation feature and the circuit compensation feature. Suppose the environmental compensation feature is represented as a vector [0.3, 0.5, 0.4], representing the processed temperature, humidity, and electromagnetic intensity feature values respectively; the circuit compensation feature is represented as a vector [0.6, 0.2, 0.7], representing the processed electrical signal fluctuation, frequency, and phase feature values. The server uses a specific algorithm, such as adding the elements at the corresponding positions and normalizing: [(0.3 + 0.6) / 2, (0.5 + 0.2) / 2, (0.4 + 0.7) / 2] = [0.45, 0.35, 0.55]. The new vector obtained is the first aggregated feature. This first aggregated feature synthesizes environmental and circuit state information and provides a basis for the subsequent matching compensation strategy module.

[0025] In an embodiment of the present invention, the aggregating the environmental compensation feature and the circuit compensation feature to obtain a first aggregated feature can be implemented through the following example.

[0026] Obtain a first effective value corresponding to the environmental compensation feature, and perform calibration processing on the environmental compensation feature based on the first effective value to obtain an environmental calibration feature; Perform dynamic gain adjustment on the environment calibration feature to obtain an environment gain feature, and perform smoothing processing on the environment gain feature to obtain an environment smoothing feature; Obtain a second effective value corresponding to the circuit compensation feature, and perform calibration processing on the circuit compensation feature based on the second effective value to obtain a circuit calibration feature; Perform dynamic gain adjustment on the circuit calibration feature to obtain a circuit gain feature, and perform smoothing processing on the circuit gain feature to obtain a circuit smoothing feature; Merge the environment smoothing feature and the circuit smoothing feature to obtain a comprehensive compensation feature, and perform a feature reconstruction operation on the comprehensive compensation feature based on the compensation parameter matrix corresponding to the parameter fusion module in the dynamic fusion module to obtain the first aggregation feature.

[0027] In an embodiment of the present invention, exemplarily, the server first obtains a first valid value corresponding to an environmental compensation feature. For example, the environmental compensation feature is composed of multiple features such as temperature, humidity, and electromagnetic intensity, and its first valid value obtained through calculation is 0.75. Based on this first valid value, the server performs a calibration process on the environmental compensation feature. Suppose the temperature feature value in the environmental compensation feature is 0.6, then after calibration, the temperature feature value becomes 0.6×0.75 = 0.45, and other features such as humidity and electromagnetic intensity are also processed in this way to obtain an environmentally calibrated feature. Next, the server performs a dynamic gain adjustment on the environmentally calibrated feature. According to the empirical model of the influence of the environment in different areas of the park on the current sensor, for the temperature feature, if the equipment in this area is sensitive to temperature, the server may set the gain coefficient of the temperature feature to 1.2, and the adjusted temperature feature value becomes 0.45×1.2 = 0.54, and other features are adjusted according to the corresponding gain rules to obtain an environmental gain feature. Subsequently, the server performs a smoothing process on the environmental gain feature to remove the small fluctuations that may be introduced due to the gain adjustment, and obtains an environmental smoothing feature. After that, the server obtains a second valid value corresponding to the circuit compensation feature, which is assumed to be 0.8. Based on this, a calibration process is performed on the circuit compensation feature. For example, if the electrical signal fluctuation feature value in the circuit compensation feature is 0.5, it becomes 0.5×0.8 = 0.4 after calibration, and other features such as electrical signal frequency and phase are also processed in the same way to obtain a circuit calibrated feature. The server then performs a dynamic gain adjustment on the circuit calibrated feature. For example, according to the circuit characteristics, the gain coefficient of the electrical signal frequency feature is set to 0.9, and the adjusted electrical signal frequency feature value becomes 0.4×0.9 = 0.36, and other features are adjusted according to the rules to obtain a circuit gain feature. A smoothing process is performed on the circuit gain feature to obtain a circuit smoothing feature. Finally, the server merges the environmental smoothing feature and the circuit smoothing feature to obtain a comprehensive compensation feature. Suppose the environmental smoothing feature is [0.54, 0.3, 0.4], and the circuit smoothing feature is [0.36, 0.4, 0.5], then the merged comprehensive compensation feature is [0.54, 0.3, 0.4, 0.36, 0.4, 0.5]. Based on the compensation parameter matrix corresponding to the parameter fusion module in the dynamic fusion module, a feature reconstruction operation is performed on the comprehensive compensation feature. The compensation parameter matrix performs a matrix operation with the comprehensive compensation feature to recombine the features, and finally obtains a first aggregated feature, laying a foundation for the subsequent matching compensation strategy module.

[0028] In an embodiment of the present invention, based on the first aggregated feature, to match a first compensation strategy module for the environmental feature among multiple compensation strategy modules, the following example can be used for implementation.

[0029] Based on the strategy decision module in the dynamic fusion module, a strategy evaluation is performed on the first aggregated feature to obtain the fitness coefficients corresponding to multiple compensation strategy modules; Determine the compensation strategy module corresponding to the maximum adaptation coefficient as the first compensation strategy module for matching the environmental features.

[0030] In an embodiment of the present invention, exemplarily, the server uses the policy decision module in the dynamic fusion module to perform a policy evaluation on the previously obtained first aggregated feature. In this park, each of the multiple compensation strategy modules has its own characteristics. For example, the dynamic weight fusion module is good at handling the weight distribution of multi-source data to achieve compensation; the main channel fusion unit module focuses on the fusion compensation of key channel features; the reference signal suppressor module suppresses and compensates for specific interference reference signals, etc. The policy decision module starts to work. It deeply analyzes each dimension and feature pattern of the first aggregated feature. Taking the dynamic weight fusion module as an example, the policy decision module examines the degree of association and the trend of feature changes between the environmental features and the circuit state features in the first aggregated feature. If the first aggregated feature shows a close connection between the temperature and the electrical signal fluctuation in the environment, and this connection presents a certain non-linear law, the dynamic weight fusion module may better adapt to this situation through its unique weight distribution mechanism. The policy decision module, according to its own algorithm, gives this module an adaptation coefficient, such as 0.7. For the main channel fusion unit module, the policy decision module examines the importance and mutual relationship of the main channel features in the first aggregated feature. If the main channel features such as temperature and electrical signal frequency in the first aggregated feature have a significant impact on the error and their changes have a certain time series characteristic, and the main channel fusion unit module can effectively capture and fuse these features, the policy decision module will assign it an adaptation coefficient, assumed to be 0.5. Similarly, the policy decision module evaluates the reference signal suppressor module. If the first aggregated feature reflects that the electromagnetic interference in the environment affects the current sensor similar to the reference signal, and the reference signal suppressor module can effectively suppress this interference, the policy decision module will give the corresponding adaptation coefficient, such as 0.4. After evaluating all the compensation strategy modules, the server obtains the adaptation coefficients corresponding to multiple compensation strategy modules. At this time, the server determines the compensation strategy module corresponding to the maximum adaptation coefficient as the first compensation strategy module for environmental feature matching. In the above example, the adaptation coefficient of the dynamic weight fusion module is the largest at 0.7, so the server determines the dynamic weight fusion module as the first compensation strategy module for environmental feature matching, so as to perform more accurate error compensation operations on the environmental features based on this module later.

[0031] In an embodiment of the present invention, the first compensation strategy module includes a dynamic weight fusion unit; In the first compensation strategy module, based on the circuit state features, perform a compensation parameter fusion operation on the environmental features to obtain a first compensation feature, which can be implemented through the following examples.

[0032] Perform channel combination on the environmental feature and the circuit state feature to obtain a first combined feature; Based on the dynamic weight fusion device, perform cross-source dynamic correlation analysis on the first combined feature to obtain a first dynamic correlation feature; Determine the first X groups of channel features in the first dynamic correlation feature as the first compensation feature; X is the number of main channels corresponding to the environmental feature.

[0033] In the embodiment of the present invention, for example, in the scenario of a large industrial production park, the server uses the dynamic weight fusion device in the first compensation strategy module to process the environmental feature and the circuit state feature to obtain the first compensation feature. The server first performs channel combination on the environmental feature and the circuit state feature. The environmental feature includes processed feature data such as temperature, humidity, and electromagnetic intensity, which is assumed to be represented as a vector [0.6, 0.4, 0.5], corresponding to temperature, humidity, and electromagnetic intensity respectively; the circuit state feature such as feature data of electrical signal fluctuation, frequency, phase, etc., is assumed to be represented as a vector [0.7, 0.3, 0.8]. The server splices these two vectors in the channel dimension to obtain the first combined feature [0.6, 0.4, 0.5, 0.7, 0.3, 0.8]. Then, the server performs cross-source dynamic correlation analysis on the first combined feature based on the dynamic weight fusion device. The dynamic weight fusion device will analyze the correlation between different sources in the first combined feature - environmental and circuit features. For example, it finds that there is a certain correlation between temperature and electrical signal fluctuation, and there is also a potential connection between humidity and electrical signal frequency. Based on these correlations, the dynamic weight fusion device assigns different weights to each feature. For the temperature feature, since it is closely related to the electrical signal fluctuation, it may be assigned a weight of 0.8; the correlation between humidity and electrical signal frequency is relatively weak, and a weight of 0.6 is assigned, etc. Through the weight assignment and calculation of all features, the first dynamic correlation feature is obtained. Suppose the first dynamic correlation feature after calculation is [0.48, 0.24, 0.3, 0.56, 0.18, 0.64]. Finally, determine the number of main channels X corresponding to the environmental feature. Suppose through the previous analysis of the current sensor data in the park, it is determined that temperature and humidity are the main channels corresponding to the environmental feature, that is, X = 2. The server determines the first 2 groups of channel features [0.48, 0.24] in the first dynamic correlation feature as the first compensation feature. This first compensation feature comprehensively considers the dynamic correlation between the environmental feature and the circuit state feature, providing an important data basis for subsequent further processing to compensate for the current sensor error.

[0034] In the embodiment of the present invention, the step of performing cross-source dynamic correlation analysis on the first combined feature based on the dynamic weight fusion device to obtain the first dynamic correlation feature can be implemented through the following example.

[0035] Based on the transformation compensation parameter matrix corresponding to the dynamic weight fuser, map the first joint feature into a first environment correlation matrix, a first circuit correlation matrix, and a first compensation parameter matrix; Based on the multiplication result between the first environment correlation matrix and the environment feature, and the multiplication result between the first environment correlation matrix and the circuit state feature, construct a first coupling parameter matrix; Based on the dimension conversion operation result of the multiplication result between the first circuit correlation matrix and the environment feature, and the dimension conversion operation result of the multiplication result between the first circuit correlation matrix and the circuit state feature, construct a second coupling parameter matrix; Based on the multiplication result between the first compensation parameter matrix and the environment feature, and the multiplication result between the first compensation parameter matrix and the circuit state feature, construct a third coupling parameter matrix; Determine the multiplication result between the first coupling parameter matrix and the second coupling parameter matrix as the fourth coupling parameter matrix, and obtain the total number of channels of the environment feature; Perform a normalization ratio calculation on the fourth coupling parameter matrix and the normalization scaling factor of the total number of channels to obtain a cross-source dynamic weight matrix, and determine the multiplication result between the cross-source dynamic weight matrix and the third coupling parameter matrix as the first dynamic correlation feature.

[0036] In an embodiment of the present invention, exemplarily, in a large industrial production park, the server continues to use a dynamic weight integrator to perform a more detailed cross-source dynamic correlation analysis on the first combined feature to obtain a first dynamic correlation feature. The server processes the first combined feature according to the transformation compensation parameter matrix corresponding to the dynamic weight integrator. Assume that the first combined feature is [0.6, 0.4, 0.5, 0.7, 0.3, 0.8]. Through specific matrix operation rules, it is mapped into a first environment correlation matrix, a first circuit correlation matrix, and a first compensation parameter matrix. For example, after complex linear transformation and other operations, the first environment correlation matrix is [[0.2, 0.3], [0.4, 0.1]], the first circuit correlation matrix is [[0.5, 0.2], [0.3, 0.4]], and the first compensation parameter matrix is [[0.6, 0.1], [0.2, 0.5]]. Next, a first coupling parameter matrix is constructed. Given that the environment feature is [0.6, 0.4] and the circuit state feature is [0.7, 0.3]. The first environment correlation matrix is multiplied by the environment feature, that is, [0.2×0.6 + 0.3×0.4, 0.4×0.6 + 0.1×0.4] = [0.24, 0.28]; the first environment correlation matrix is multiplied by the circuit state feature, that is, [0.2×0.7 + 0.3×0.3, 0.4×0.7 + 0.1×0.3] = [0.23, 0.31]. Combining these two results to construct the first coupling parameter matrix as [[0.24, 0.23], [0.28, 0.31]]. When constructing the second coupling parameter matrix, first perform a dimension conversion operation on the results of multiplying the first circuit correlation matrix by the environment feature and the circuit state feature. The product of the first circuit correlation matrix and the environment feature is [0.5×0.6 + 0.2×0.4, 0.3×0.6 + 0.4×0.4] = [0.38, 0.34], and after dimension conversion, it is assumed to become [[0.38], [0.34]]; the product with the circuit state feature is [0.5×0.7 + 0.2×0.3, 0.3×0.7 + 0.4×0.3] = [0.41, 0.33], and after dimension conversion, it is assumed to become [[0.41], [0.33]]. Combining these two dimension-converted results to construct the second coupling parameter matrix as [[0.38, 0.41], [0.34, 0.33]]. To construct the third coupling parameter matrix, the product of the first compensation parameter matrix and the environment feature is [0.6×0.6 + 0.1×0.4, 0.2×0.6 + 0.5×0.4] = [0.4, 0.32]; the product with the circuit state feature is [0.6×0.7 + 0.1×0.3, 0.2×0.7 + 0.5×0.3] = [0.45, 0.29]. Thus, the third coupling parameter matrix is constructed as [[0.4, 0.45], [0.32, 0.29]]. Then, multiply the first coupling parameter matrix and the second coupling parameter matrix to obtain a fourth coupling parameter matrix.That is, [[0.24×0.38 + 0.23×0.34, 0.24×0.41 + 0.23×0.33], [0.28×0.38 + 0.31×0.34, 0.28×0.41 + 0.31×0.33]] = [[0.17, 0.18], [0.20, 0.21]]. The server obtains the total number of channels of the environmental features. Here, the environmental features are [0.6, 0.4] and the total number of channels is 2. The fourth coupling parameter matrix is calculated with the normalization scaling factor (1 / 2) of the total number of channels to obtain the cross-source dynamic weight matrix. For example, [[0.17×0.5, 0.18×0.5], [0.20×0.5, 0.21×0.5]] = [[0.085, 0.09], [0.1, 0.105]]. Finally, the cross-source dynamic weight matrix is multiplied by the third coupling parameter matrix to obtain the first dynamic correlation feature. That is, [[0.085×0.4 + 0.09×0.32, 0.085×0.45 + 0.09×0.29], [0.1×0.4 + 0.105×0.32, 0.1×0.45 + 0.105×0.29]] = [[0.0628, 0.0621], [0.0736, 0.0755]], and this is the first dynamic correlation feature.

[0037] In an embodiment of the present invention, the first compensation strategy module includes a dynamic weight integrator; In the first compensation strategy module, based on the circuit state features, a compensation parameter integration operation is performed on the environmental features to obtain the first compensation feature, which can be implemented through the following example.

[0038] Based on the dynamic weight integrator, the environmental features are mapped to a second environmental correlation matrix, and the circuit state features are mapped to a second circuit correlation matrix and a second compensation parameter matrix; A matrix multiplication operation is performed on the dimension conversion operation results of the multiplication results between the second environmental correlation matrix and the environmental features, and between the second circuit correlation matrix and the circuit state features to obtain a fifth coupling parameter matrix, and the total number of channels of the environmental features is obtained; The fifth coupling parameter matrix is calculated with the normalization scaling factor of the total number of channels to obtain a multi-source contribution coefficient matrix, and a matrix multiplication operation is performed on the multi-source contribution coefficient matrix, and the multiplication results between the second compensation parameter matrix and the circuit state features to obtain a multi-source compensation feature; The first X groups of channel features in the multi-source compensation feature are determined as the first compensation feature; X is the number of main channels corresponding to the environmental features.

[0039] In an embodiment of the present invention, exemplarily, in a large industrial production park, the server utilizes a dynamic weight fuser in the first compensation strategy module to perform a compensation parameter fusion operation on environmental features based on circuit state features to obtain first compensation features. The server uses the dynamic weight fuser to map environmental features and circuit state features. Assume that the environmental features are represented as a vector [0.7, 0.6, 0.5], corresponding to temperature, humidity, and electromagnetic intensity respectively; the circuit state features are represented as a vector [0.8, 0.4, 0.9], corresponding to electrical signal fluctuations, frequency, and phase respectively. Through a specific operation rule of the dynamic weight fuser, the environmental features are mapped to a second environmental correlation matrix [[0.3, 0.2, 0.1], [0.4, 0.3, 0.2]]; the circuit state features are mapped to a second circuit correlation matrix [[0.5, 0.3, 0.2], [0.4, 0.3, 0.3]] and a second compensation parameter matrix [[0.6, 0.4, 0.5], [0.3, 0.5, 0.4]]. Next, the server performs operations on the relevant results to obtain a fifth coupling parameter matrix. The second environmental correlation matrix is multiplied by the environmental features, that is, [0.3 * 0.7 + 0.2×0.6 + 0.1×0.5, 0.4×0.7 + 0.3×0.6 + 0.2×0.5] = [0.38, 0.52]. The second circuit correlation matrix multiplied by the circuit state features is [0.5×0.8 + 0.3×0.4 + 0.2×0.9, 0.4×0.8 + 0.3×0.4 + 0.3×0.9] = [0.7, 0.67], and after performing a dimension conversion operation, it is assumed to become [[0.7], [0.67]]. Performing a matrix multiplication operation on these two results, the fifth coupling parameter matrix is [[0.38×0.7], [0.52×0.67]] = [[0.266], [0.3484]]. The server obtains that the total number of channels of the environmental features is 3. Then, the server performs a normalization ratio calculation on the fifth coupling parameter matrix and the normalization scaling factor 1 / 3 of the total number of channels to obtain a multi-source contribution coefficient matrix. That is, [[0.266×1 / 3], [0.3484×1 / 3]] = [[0.0887], [0.1161]]. The second compensation parameter matrix multiplied by the circuit state features is [0.6×0.8 + 0.4×0.4 + 0.5×0.9, 0.3×0.8 + 0.5×0.4 + 0.4×0.9] = [1.21, 1.0]. Performing a matrix multiplication operation on the multi-source contribution coefficient matrix and this result, a multi-source compensation feature is obtained. That is, [[0.0887×1.21], [0.1161×1.0]] = [[0.1073], [0.1161]]. Finally, assume that through long-term analysis of the park current sensor data, it is determined that temperature and humidity are the main channels corresponding to the environmental features, that is, X = 2. The server determines the first 2 groups of channel features [0.1073, 0.1161] in the multi-source compensation feature as the first compensation features.This first compensation feature comprehensively considers the relationship between the environmental feature and the circuit state feature, providing key data for subsequent current sensor error compensation.

[0040] In the embodiment of the present invention, the first compensation strategy module includes a multi-source dynamic fusion device. In the first compensation strategy module, a compensation parameter fusion operation is performed on the environmental feature based on the circuit state feature to obtain a first compensation feature, which can be implemented through the following examples.

[0041] Obtain a fusion compression feature, perform channel combination on the fusion compression feature and the circuit state feature to obtain a second combined feature; the fusion compression feature is used to establish the connection between multi-source features. Based on the first dynamic weight assignment branch in the multi-source dynamic fusion device, perform cross-source dynamic association analysis on the second combined feature to obtain a second dynamic association feature, and determine the first Y groups of main channel features in the second dynamic association feature as the target compensation feature; Y is the number of main channels corresponding to the fusion compression feature. Perform channel combination on the environmental feature and the target compensation feature to obtain a third combined feature, and perform cross-source dynamic association analysis on the third combined feature based on the second dynamic weight assignment branch in the multi-source dynamic fusion device to obtain a third dynamic association feature. Determine the first X groups of channel features in the third dynamic association feature as the first compensation feature; X is the number of main channels corresponding to the environmental feature.

[0042] In an embodiment of the present invention, exemplarily, in a large industrial production park, a server utilizes a first compensation strategy module including a multi-source dynamic fuser to perform compensation parameter fusion operation on environmental features based on circuit state features, generating a first compensation feature. The server first obtains a fused and compressed feature. This feature is obtained by deeply analyzing and fusing various data related to current sensors in the park, such as historical environmental data, circuit performance indicators, etc., and using a specific compression algorithm. For example, by analyzing data such as temperature, humidity, electromagnetic intensity at different times in the past week, and the electrical signal fluctuations and frequencies of current sensors at corresponding times, and using algorithms such as principal component analysis, a set of key information is extracted as the fused and compressed feature. Suppose this feature is represented as a vector [0.4, 0.3, 0.2]. The server performs channel combination on the fused and compressed feature and the circuit state feature. Suppose the circuit state feature is [0.7, 0.5, 0.6, 0.8], representing the fluctuations, frequencies, phases, and amplitudes of electrical signals respectively. After combination, the second combined feature [0.4, 0.3, 0.2, 0.7, 0.5, 0.6, 0.8] is obtained. Then, the server uses the first dynamic weight allocation branch in the multi-source dynamic fuser to perform cross-source dynamic correlation analysis on the second combined feature. The first dynamic weight allocation branch will consider the potential relationships between different features, such as the correlation between electrical signal fluctuations and temperature, humidity, and the correlation between frequency and electromagnetic intensity, etc. Through complex algorithm analysis, weights are assigned to each feature. For example, it is found that the electrical signal fluctuations are closely related to temperature, and a higher weight of 0.8 is assigned to the corresponding feature, and other features also obtain corresponding weights accordingly. After weighted calculation and processing, a second dynamic correlation feature is obtained, suppose it is [0.32, 0.18, 0.12, 0.56, 0.25, 0.36, 0.64]. Suppose the number of main channels Y corresponding to the fused and compressed feature is 2, and the server determines the first 2 groups of main channel features [0.32, 0.18] in the second dynamic correlation feature as the target compensation feature. Then, the server performs channel combination on the environmental feature and the target compensation feature. Suppose the environmental feature is [0.6, 0.5, 0.4], and after combination, the third combined feature [0.6, 0.5, 0.4, 0.32, 0.18] is obtained. The server then uses the second dynamic weight allocation branch in the multi-source dynamic fuser to perform cross-source dynamic correlation analysis on the third combined feature. The second dynamic weight allocation branch re-examines the relationships between features, reassigns weights, performs weighted calculation and processing, and obtains a third dynamic correlation feature, such as [0.48, 0.35, 0.24, 0.16, 0.09]. Suppose the number of main channels X corresponding to the environmental feature is 3, and the server determines the first 3 groups of channel features [0.48, 0.35, 0.24] in the third dynamic correlation feature as the first compensation feature. This first compensation feature fully integrates the relevant information of environmental features and circuit state features, laying a foundation for accurately compensating the errors of current sensors in the follow-up.

[0043] In an embodiment of the present invention, the first compensation strategy module includes a main channel fusion unit; In the first compensation strategy module, based on the circuit state characteristics, a compensation parameter fusion operation is performed on the environmental characteristics to obtain a first compensation characteristic, which can be implemented through the following examples.

[0044] Obtain the environmental calibration characteristics corresponding to the environmental characteristics and the circuit calibration characteristics corresponding to the circuit state characteristics; Based on the first dynamic timing processor in the main channel fusion unit, perform timing signal modulation on the environmental characteristics to obtain a first timing response characteristic; Based on the second dynamic timing processor in the main channel fusion unit, perform timing signal modulation on the circuit state characteristics to obtain a second timing response characteristic; Perform smoothing processing on the second timing response characteristic to obtain a timing smoothing characteristic, and merge the first timing response characteristic and the timing smoothing characteristic to obtain a first compensation characteristic.

[0045] In an embodiment of the present invention, exemplarily, in a large industrial production park, a server uses a first compensation strategy module including a main channel fusion unit to process environmental characteristics and circuit state characteristics to obtain first compensation characteristics. The server first obtains environmental calibration characteristics corresponding to the environmental characteristics and circuit calibration characteristics corresponding to the circuit state characteristics. Suppose the environmental characteristics are [0.7, 0.6, 0.5], representing temperature, humidity, and electromagnetic intensity respectively. According to the preset calibration rules, for example, multiplying each characteristic value by a calibration coefficient of 0.8, the environmental calibration characteristics [0.56, 0.48, 0.4] are obtained. For the circuit state characteristics, suppose they are [0.8, 0.4, 0.9], representing electrical signal fluctuation, frequency, and phase. Similarly, according to the corresponding calibration rules, such as multiplying by 0.9, the circuit calibration characteristics [0.72, 0.36, 0.81] are obtained. Then, the server uses the first dynamic time series processor in the main channel fusion unit to perform time series signal modulation on the environmental characteristics. In an industrial production scenario, environmental factors change regularly over time. For example, the temperature gradually rises during the daytime production period, and the humidity fluctuates in areas with intensive equipment operation. The first dynamic time series processor modulates the environmental characteristics based on these time series change characteristics. It may add a time-related weight to the current environmental characteristic values according to the change trends of temperature, humidity, and electromagnetic intensity at different times in historical data. Suppose the current time is during the peak production period in the morning, and the weight of the temperature characteristic increases. After modulation, the first time series response characteristics, such as [0.6, 0.5, 0.45], are obtained. At the same time, the server uses the second dynamic time series processor in the main channel fusion unit to perform time series signal modulation on the circuit state characteristics. The circuit state also changes over time. For example, the electrical signals have different manifestations during the startup, operation, and shutdown processes of equipment. The second dynamic time series processor modulates the circuit calibration characteristics considering the changes in electrical signal fluctuation, frequency, and phase at different production stages. For example, when the equipment just starts, the electrical signal fluctuation is large, and the corresponding characteristic weight increases. After modulation, the second time series response characteristics, such as [0.85, 0.4, 0.9], are obtained. After that, the server performs smoothing processing on the second time series response characteristics. Since the electrical signals may be affected by some instantaneous interferences during the operation of the equipment, resulting in fluctuations in characteristic values, the smoothing processing can remove these unnecessary fluctuations. Through a specific smoothing algorithm, such as the moving average method, [0.85, 0.4, 0.9] is processed to obtain the time series smoothing characteristics, suppose it is [0.83, 0.4, 0.88]. Finally, the server combines the first time series response characteristics [0.6, 0.5, 0.45] and the time series smoothing characteristics [0.83, 0.4, 0.88]. By concatenating in order or through a specific combination algorithm, the first compensation characteristics, such as [0.6, 0.5, 0.45, 0.83, 0.4, 0.88], are obtained.This first compensation feature integrates the variation information of environmental features and circuit state features in terms of time sequence, providing more comprehensive data support for subsequent current sensor error compensation.

[0046] In an embodiment of the present invention, the first compensation strategy module includes a reference signal suppressor; In the first compensation strategy module, a compensation parameter fusion operation is performed on the environmental features based on the circuit state features to obtain a first compensation feature, which can be implemented through the following example.

[0047] Clear the circuit state features, and use the environmental features as the first compensation feature output by the reference signal suppressor.

[0048] In an embodiment of the present invention, exemplarily, in a complex electromagnetic environment of a large industrial production park, the server adopts a first compensation strategy module including a reference signal suppressor to process the environmental characteristics and circuit state characteristics of the current sensor to obtain a first compensation feature. In this park, there are various interference sources around the current sensor, which will generate interference similar to the reference signal, affecting the measurement accuracy. When the server calls this module, it will first analyze the currently obtained environmental characteristics and circuit state characteristics. Suppose the environmental characteristics are represented as the vector [0.6, 0.5, 0.4], representing temperature, humidity, and electromagnetic intensity respectively; the circuit state characteristics are represented as [0.7, 0.3, 0.8], representing electrical signal fluctuation, frequency, and phase. Based on its design principle, the reference signal suppressor here believes that under the current compensation requirement, the signals contained in the circuit state characteristics may be the main source of interference, similar to the reference signal interfering with the accurate analysis of the environmental characteristics, thereby affecting the accuracy of the current sensor error compensation. Therefore, the server directly clears the circuit state characteristics [0.7, 0.3, 0.8] according to the function setting of the reference signal suppressor. Subsequently, the server uses the environmental characteristics [0.6, 0.5, 0.4] as the first compensation feature output by the reference signal suppressor. This is because in this case, the reference signal suppressor determines that after removing the circuit state characteristics, the environmental characteristics themselves can more accurately reflect the information related to the current sensor error and can be used for subsequent error compensation calculations. For example, in a specific area of the park, the strong electromagnetic interference generated by the startup of a large motor causes abnormal changes in parameters such as electrical signal fluctuation and frequency in the circuit state characteristics. These abnormally changed signals are like reference signals, interfering with the judgment of the real current situation. And environmental factors such as temperature, humidity, and electromagnetic intensity are relatively stable and can be more directly related to the possible errors of the current sensor. By clearing the disturbed circuit state characteristics and retaining the environmental characteristics as the first compensation feature, the server takes a crucial step towards accurately compensating the current sensor error in the future, ensuring that in a complex electromagnetic environment, the data of the current sensor can more accurately reflect the actual current situation and providing strong support for the stable operation of industrial production equipment.

[0049] In an embodiment of the present invention, based on the error type identifier, the first compensation feature, and the second compensation feature, determining the current error compensation result corresponding to the current sensor can be implemented through the following examples.

[0050] Perform channel combination on the first compensation feature and the second compensation feature to obtain a fourth combined feature, and add the error type identifier at the front of the fourth combined feature to obtain a combined compensation feature; Based on the cross-source dynamic weight allocation branch, dynamically evaluate the combined compensation feature to obtain a dynamic weight fusion parameter, and perform a feature reconstruction operation on the dynamic weight fusion parameter to obtain the current error compensation result corresponding to the current sensor.

[0051] In an embodiment of the present invention, exemplarily, in a large industrial production park, a server needs to determine the current error compensation result corresponding to a current sensor based on an error type identifier, a first compensation feature, and a second compensation feature. Suppose the first compensation feature is represented as a vector [0.5, 0.3, 0.4], which is obtained by processing environmental features through a specific compensation strategy module before, reflecting the information related to the impact of environmental factors on the current error; the second compensation feature is represented as a vector [0.6, 0.2, 0.7], which is obtained by performing a compensation parameter fusion operation on the circuit state features, reflecting the effect of the circuit state on the current error. The server first performs channel combination on the first compensation feature and the second compensation feature. The two vectors are concatenated in sequence to obtain a fourth combined feature [0.5, 0.3, 0.4, 0.6, 0.2, 0.7]. At this time, the server obtains the error type identifier. For example, after in-depth analysis of the current sensor data, it is determined that the current error type is "electromagnetic interference error", and the system encodes this error type. Suppose the encoding is [1, 0, 0]. If there are multiple error types, they are represented in one-hot code form. The server adds this error type identifier at the front end of the fourth combined feature to obtain a combined compensation feature [1, 0, 0, 0.5, 0.3, 0.4, 0.6, 0.2, 0.7]. Next, the server performs dynamic evaluation on the combined compensation feature based on a cross-source dynamic weight allocation branch. The cross-source dynamic weight allocation branch will analyze the correlation degree between each element in the combined compensation feature and the error type, as well as the mutual relationship between different feature sources. For example, it is found that the "1" in the error type identifier represents that the electromagnetic interference error is closely related to the electromagnetic intensity-related element in the first compensation feature, here supposed to be 0.4, and the electrical signal fluctuation-related element in the second compensation feature, here supposed to be 0.6. Based on these analyses, the cross-source dynamic weight allocation branch assigns different weights to each element of the combined compensation feature. After a series of complex calculations, a dynamic weight fusion parameter is obtained, supposed to be [0.2, 0.1, 0.1, 0.3, 0.1, 0.2]. Finally, the server performs a feature reconstruction operation on the dynamic weight fusion parameter. Through a specific algorithm and a pre-trained model, the dynamic weight fusion parameter is converted into a value that can be directly used to compensate the current sensor error. For example, through operations such as matrix operations and non-linear transformations, the dynamic weight fusion parameter is mapped to the numerical range of current error compensation, and finally the current error compensation result corresponding to the current sensor is obtained. This result may be a specific value used to adjust the current value output by the current sensor to correct the error caused by factors such as electromagnetic interference, ensuring that the data provided by the current sensor more accurately reflects the actual current situation and guaranteeing the stable operation of industrial production equipment.

[0052] An embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the foregoing current sensor error compensation method. As Figure 2 shown, Figure 2 is a structural block diagram of the computer device 100 provided by an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To achieve data transmission or interaction, the elements of the memory 111, the processor 112, and the communication unit 113 are electrically connected to each other directly or indirectly. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines.

[0053] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. The embodiments were chosen and described in order to best illustrate the principles of the disclosure and its practical applications, to thereby enable those skilled in the art to best utilize the disclosure and to utilize various embodiments with various modifications as are suited to the particular application contemplated.

Claims

1. A method for compensating the error of a current sensor, characterized in that, Including: Obtain the operating environment corresponding to the current sensor, and divide the operating environment into multiple time windows; Extract features from the multiple time windows to obtain environment coding features corresponding to the multiple time windows; Dynamically evaluate the environment coding features corresponding to the multiple time windows to obtain contribution coefficient corresponding to the multiple time windows; Dynamically fuse the environment coding features corresponding to the multiple time windows and the contribution coefficient corresponding to the multiple time windows to obtain the environment feature corresponding to the operating environment; Obtain multiple electrical signal time series segments corresponding to the current sensor, and fuse the multiple electrical signal time series segments into an electrical signal joint feature; Input the electrical signal joint feature into a feature fusion network, and perform a feature reconstruction operation on the electrical signal joint feature based on the feature fusion network to obtain a circuit state feature corresponding to the current sensor; Obtain an error type identifier, and determine a current error compensation result corresponding to the current sensor based on the error type identifier, the environment feature, and the circuit state feature.

2. The error compensation method of a current sensor according to claim 1, characterized in that The obtaining an error type identifier, and determining a current error compensation result corresponding to the current sensor based on the error type identifier, the environment feature, and the circuit state feature includes: Perform an aggregation process on the environment feature and the circuit state feature within the same feature domain to obtain a first aggregation feature, and based on the first aggregation feature, match a first compensation strategy module for the environment feature among multiple compensation strategy modules; the multiple compensation strategy modules are configured with different compensation strategies to adjust features of multiple data sources; In the first compensation strategy module, perform a compensation parameter fusion operation on the environment feature based on the circuit state feature to obtain a first compensation feature; Perform an aggregation process on the circuit state feature and the first compensation feature within the same feature domain to obtain a second aggregation feature, and based on the second aggregation feature, match a second compensation strategy module for the circuit state feature among the multiple compensation strategy modules; In the second compensation strategy module, perform a compensation parameter fusion operation on the circuit state feature based on the first compensation feature to obtain a second compensation feature; Obtain an error type identifier, and determine a current error compensation result corresponding to the current sensor based on the error type identifier, the first compensation feature, and the second compensation feature.

3. The error compensation method for a current sensor according to claim 2, wherein, The performing an aggregation process on the environment feature and the circuit state feature within the same feature domain to obtain a first aggregation feature includes: Input the environment feature and the circuit state feature into a dynamic fusion module, and perform feature alignment on the environment feature based on a first feature projection unit in the dynamic fusion module to obtain an environment compensation feature; Perform feature alignment on the circuit state feature based on a second feature projection unit in the dynamic fusion module to obtain a circuit compensation feature; the environment compensation feature and the circuit compensation feature belong to the same feature domain; Obtain a first effective value corresponding to the environment compensation feature, and perform a calibration process on the environment compensation feature based on the first effective value to obtain an environment calibration feature; Dynamically adjust the gain of the environmental calibration feature to obtain an environmental gain feature, and smooth the environmental gain feature to obtain an environmental smoothing feature; Acquire a second effective value corresponding to the circuit compensation feature, and perform calibration processing on the circuit compensation feature based on the second effective value to obtain a circuit calibration feature; Dynamically adjust the circuit calibration characteristics to obtain circuit gain characteristics, and smooth the circuit gain characteristics to obtain circuit smoothing characteristics; The environmental smoothing feature and the circuit smoothing feature are combined to obtain a comprehensive compensation feature, and based on the compensation parameter matrix corresponding to the parameter fusion module in the dynamic fusion module, a feature reconstruction operation is performed on the comprehensive compensation feature to obtain the first aggregated feature.

4. A method for compensating the error of a current sensor according to claim 3, characterized in that, The matching of a first compensation strategy module for the environmental feature among multiple compensation strategy modules based on the first aggregate feature includes: Performing a strategy evaluation on the first aggregated feature based on the strategy decision module in the dynamic fusion module to obtain fitness coefficients corresponding to multiple compensation strategy modules; The compensation strategy module corresponding to the largest fitness coefficient is determined as the first compensation strategy module matching the environmental features.

5. A method for compensating the error of a current sensor according to claim 2, characterized in that, The first compensation strategy module includes a dynamic weight fuser; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature includes: Perform channel combination on the environment feature and the circuit state feature to obtain a first combination feature; Based on the transformation compensation parameter matrix corresponding to the dynamic weight fuser, mapping the first joint feature into a first environment association matrix, a first circuit association matrix and a first compensation parameter matrix; constructing a first coupling parameter matrix based on a multiplication result between the first environment association matrix and the environment feature, and a multiplication result between the first environment association matrix and the circuit state feature; constructing a second coupling parameter matrix based on a dimension conversion operation result of a multiplication result between the first circuit association matrix and the environmental feature, and a dimension conversion operation result of a multiplication result between the first circuit association matrix and the circuit state feature; constructing a third coupling parameter matrix based on a multiplication result between the first compensation parameter matrix and the environmental characteristics, and a multiplication result between the first compensation parameter matrix and the circuit state characteristics; Determine the multiplication result between the first coupling parameter matrix and the second coupling parameter matrix as a fourth coupling parameter matrix, and obtain the total number of channels of the environmental feature; Performing normalized proportional calculation on the fourth coupling parameter matrix and the normalized scaling factor of the total number of channels to obtain a cross-source dynamic weight matrix, and determining a multiplication result between the cross-source dynamic weight matrix and the third coupling parameter matrix as a first dynamic association feature; The first X groups of channel features in the first dynamic association feature are determined as first compensation features, where X is the number of main channels corresponding to the environmental feature.

6. A method for compensating the error of a current sensor according to claim 2, characterized in that, The first compensation strategy module includes a dynamic weight fuser; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature, including: Based on the dynamic weight fuser, mapping the environmental feature to a second environmental correlation matrix, mapping the circuit state feature to a second circuit correlation matrix and a second compensation parameter matrix; Performing a matrix multiplication operation on the dimensionality conversion operation results of the multiplication results between the second environmental correlation matrix and the environmental feature, and between the second circuit correlation matrix and the circuit state feature, to obtain a fifth coupling parameter matrix, and obtaining the total number of channels of the environmental feature; Performing a normalization ratio calculation on the fifth coupling parameter matrix and the normalization scaling factor of the total number of channels to obtain a multi-source contribution coefficient matrix, and performing a matrix multiplication operation on the multi-source contribution coefficient matrix and the multiplication result between the second compensation parameter matrix and the circuit state feature to obtain a multi-source compensation feature; Determining the first X groups of channel features in the multi-source compensation feature as the first compensation feature; X is the number of main channels corresponding to the environmental feature.

7. A method for compensating the error of a current sensor according to claim 2, characterized in that, The first compensation strategy module includes a multi-source dynamic fuser; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature, including: Obtaining a fusion compression feature, and performing channel combination on the fusion compression feature and the circuit state feature to obtain a second combined feature; the fusion compression feature is used to establish a connection between multi-source features; Based on the first dynamic weight allocation branch in the multi-source dynamic fuser, performing cross-source dynamic correlation analysis on the second combined feature to obtain a second dynamic correlation feature, and determining the first Y groups of main channel features in the second dynamic correlation feature as the target compensation feature; Y is the number of main channels corresponding to the fusion compression feature; Performing channel combination on the environmental feature and the target compensation feature to obtain a third combined feature, and based on the second dynamic weight allocation branch in the multi-source dynamic fuser, performing cross-source dynamic correlation analysis on the third combined feature to obtain a third dynamic correlation feature; Determining the first X groups of channel features in the third dynamic correlation feature as the first compensation feature; X is the number of main channels corresponding to the environmental feature; The first compensation strategy module further includes a main channel fusion unit; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature, including: Obtaining an environmental calibration feature corresponding to the environmental feature and a circuit calibration feature corresponding to the circuit state feature; Based on the first dynamic timing processor in the main channel fusion unit, performing timing signal modulation on the environmental feature to obtain a first timing response feature; Based on the second dynamic timing processor in the main channel fusion unit, performing timing signal modulation on the circuit state feature to obtain a second timing response feature; Smooth the second timing response feature to obtain a timing smooth feature, and merge the first timing response feature and the timing smooth feature to obtain a first compensation feature.

8. The error compensation method of a current sensor according to claim 2, characterized in that The first compensation strategy module includes a reference signal suppressor; In the first compensation strategy module, performing a compensation parameter fusion operation on the environmental feature based on the circuit state feature to obtain a first compensation feature includes: Clear the circuit state feature, and use the environmental feature as the first compensation feature output by the reference signal suppressor.

9. A method for compensating the error of a current sensor according to claim 2, characterized in that Determining the current error compensation result corresponding to the current sensor based on the error type identifier, the first compensation feature, and the second compensation feature includes: Perform channel combination on the first compensation feature and the second compensation feature to obtain a fourth combined feature, and add the error type identifier at the front end of the fourth combined feature to obtain a combined compensation feature; Perform dynamic evaluation on the combined compensation feature based on the cross-source dynamic weight allocation branch to obtain a dynamic weight fusion parameter, and perform a feature reconstruction operation on the dynamic weight fusion parameter to obtain the current error compensation result corresponding to the current sensor.

10. A server system, characterized in that, Including a server, the server is configured to execute the current sensor error compensation method according to any one of claims 1-9.

Images