Data Fusion Method for Industrial Data Quality Consistency Testing Based on Artificial Intelligence

Abstract

This invention discloses a data fusion method for industrial data quality consistency testing based on artificial intelligence, belonging to the field of industrial test data fusion technology. The method includes the following steps: performing cross-comparison operations on spatial positioning parameters and evolutionary correlation parameters in the residual multi-domain mapping sequence to determine whether there is asynchronous offset in the feature space of the residuals after AI reconstruction of multi-source industrial data; if asynchronous offset exists, forming a residual offset characterization parameter; jointly calculating the residual offset characterization parameter and the evolutionary correlation parameter to construct offset decoupling parameters and consistency correlation parameters, and determining the degree of residual offset based on the joint calculation results. This invention solves the problem of inconsistent judgment and inaccurate alignment processing caused by asynchronous offset of residuals after AI reconstruction of multi-source industrial data, achieving offset and correlation decoupling, accurate judgment, and highly reliable data fusion effects.

Description

Technical Field

[0001] This invention relates to the field of industrial test data fusion technology, and more specifically to a data fusion method for industrial data quality consistency testing based on artificial intelligence. Background Technology

[0002] Data fusion for industrial data quality consistency testing based on artificial intelligence refers to a data processing method that comprehensively evaluates the integrity, accuracy, temporal consistency, and reliability of multi-source heterogeneous data from different devices, systems, or platforms in industrial scenarios by introducing artificial intelligence models. Based on this evaluation, consistency detection and conflict identification are performed, and the fusion strategy is dynamically adjusted according to the quality assessment results to obtain a highly reliable unified data view. Existing technologies typically first collect and preprocess industrial data, including data cleaning, missing value imputation, and format standardization, and then use machine learning or deep learning models (such as clustering models, anomaly detection models, or regression models) to model and evaluate data quality. The process involves estimating and generating quality scores or confidence indices, followed by alignment and comparison of multi-source data through a consistency test to identify conflicts, inconsistencies, or anomalies. Based on this, a fusion decision-making mechanism is constructed, such as a fusion strategy based on weight allocation, confidence screening, or a combination of rules and models to achieve data integration. Finally, the fusion result is output, and the model can be iteratively optimized through a feedback mechanism to enable the system to adapt. The entire process typically includes key steps such as data acquisition and preprocessing, feature extraction and quality assessment modeling, consistency detection and conflict identification, fusion strategy generation and execution, and result feedback optimization. These steps are closely linked to jointly achieve high-quality, consistent fusion of industrial data in complex environments.

[0003] The existing technology has the following shortcomings:

[0004] In the data fusion process for industrial data quality consistency testing based on artificial intelligence, when an AI model is used to predict and reconstruct multi-source industrial data during the consistency test calculation stage, and the residual between the original data and the reconstruction result is used as the basis for consistency testing, the residuals corresponding to each data source exhibit asynchronous shifts in the feature space due to the different fitting capabilities of the AI ​​model to different data sources. This results in the residuals simultaneously containing data differences and model biases. In this case, existing technologies cannot complete the consistency determination in industrial data quality consistency testing and the alignment processing in data fusion based on the degree of residual shift in the feature space when the residuals of multi-source industrial data are reconstructed by AI and exhibit asynchronous shifts. This leads to distorted consistency test results, causing model biases and data inconsistencies to be confused, resulting in unreasonable weight allocation or misjudgment and rejection of valid data during the data fusion process, ultimately reducing the accuracy and reliability of the fusion results.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a data fusion method for industrial data quality consistency testing based on artificial intelligence, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a data fusion method for industrial data quality consistency testing based on artificial intelligence, specifically including the following steps: S1. Perform artificial intelligence reconstruction processing on multi-source industrial data and obtain residual sequences. Construct residual multi-domain mapping sequences containing spatial positioning parameters and evolutionary correlation parameters based on the residual sequences. S2. Perform cross-comparison operation on the spatial positioning parameters and evolution correlation parameters in the residual multi-domain mapping sequence to determine whether there is asynchronous offset in the feature space of the residual after the multi-source industrial data is reconstructed by artificial intelligence. If there is asynchronous offset, form residual offset characterization parameters. S3. Jointly calculate the residual migration characterization parameter and the evolution correlation parameter to construct the migration decoupling parameter and the consistency correlation parameter, and determine the degree of residual migration based on the joint calculation results; S4. Couple the residual offset degree with the consistency correlation parameter to construct the consistency evaluation parameter. Based on the consistency evaluation parameter, complete the consistency judgment in the industrial data quality consistency test, and perform the alignment processing in data fusion in combination with the residual offset degree. S5. Based on the data fusion results, feedback correction parameters are generated, and iterative updates are performed on the residual offset characterization parameters, consistency correlation parameters, and residual offset degree.

[0008] Preferably, S1 is as follows: The multi-source industrial data is reconstructed using artificial intelligence. The multi-source industrial data is input into the corresponding artificial intelligence reconstruction model for mapping calculation, generating a reconstructed data sequence corresponding to the multi-source industrial data. The difference between the multi-source industrial data and the reconstructed data sequence is calculated for each data point to obtain the residual sequence. For the residual sequence, position mapping calculation is performed in a unified feature space to extract the distribution coordinates of each data point in the residual sequence. Then, cluster center calculation and relative position offset calculation are performed on the distribution coordinates to construct spatial positioning parameters based on the distribution center position and offset vector. After completing the construction of spatial positioning parameters, differential calculation and trend extraction are performed on adjacent data points of the residual sequence in continuous time series to form a time series change sequence. Correlation calculation is then performed on the time series change sequence to construct evolutionary correlation parameters. At the same time, according to the correspondence of data points, the spatial positioning parameters and evolutionary correlation parameters are combined and mapped at the sequence level to form a residual multi-domain mapping sequence.

[0009] Preferably, S2 specifically includes the following steps: S201. The spatial positioning parameters and evolutionary correlation parameters in the residual multi-domain mapping sequence are synchronously paired according to the correspondence of data points, and numerical difference calculation is performed for each paired data to obtain the numerical difference between the spatial positioning parameters and the evolutionary correlation parameters. At the same time, the direction of change of the spatial positioning parameters corresponding to adjacent data points and the direction of change of the evolutionary correlation parameters are determined to be consistent. Based on the numerical difference and direction consistency results, a cross-comparison result sequence containing spatial distribution difference and temporal evolution difference is constructed. S202. Based on the cross-comparison result sequence, perform joint discrimination calculation on the spatial distribution difference and the temporal evolution difference. By judging the consistency of the change trend and change direction of the difference in continuous data points, determine whether there is asynchronous shift in the feature space of the residual after the reconstruction of multi-source industrial data by artificial intelligence. S203. In the case of asynchronous offset, the spatial distribution difference and temporal evolution difference in the cross-comparison result sequence are weighted and combined, and the residual offset characterization parameter is constructed by combining the consistency result of the change direction. The residual offset characterization parameter is serialized and output according to the data point order.

[0010] Preferably, S202 specifically refers to: Based on the cross-comparison result sequence, the spatial distribution difference and the temporal evolution difference are arranged continuously according to the data point order. The spatial distribution difference corresponding to adjacent data points is differentially calculated to form a spatial difference change sequence, and the temporal evolution difference corresponding to adjacent data points is differentially calculated to form an evolution difference change sequence. For spatial difference change sequences and evolutionary difference change sequences, the change trend corresponding to each data point is marked with a direction. For the same data point, the spatial difference change direction and the evolutionary difference change direction are compared and calculated to determine the consistency of the signs. When the signs of the two directions are the same, they are marked as consistent. When the signs of the two directions are opposite, they are marked as inconsistent. The consistent and inconsistent states are then sequentially arranged according to the order of the data points to form a sequence of the degree of consistency of the change direction. Based on the sequence of consistency of change direction, interval statistical processing is performed on the distribution of consistent and inconsistent states in continuous data points. Combined with the continuity of the change trend of spatial difference change sequence and evolutionary difference change sequence, joint discrimination calculation is performed. When there are continuous inconsistent state intervals in the sequence of consistency of change direction and the corresponding change trend does not have synchronous change characteristics, it is determined that the residual of multi-source industrial data after artificial intelligence reconstruction has asynchronous offset in the feature space.

[0011] Preferably, S203 is as follows: In cases where asynchronous offset is determined, spatial distribution difference and temporal evolution difference are extracted from the cross-comparison result sequence. The spatial distribution difference and temporal evolution difference are then paired and arranged according to the data point order. At the same time, the direction consistency result is combined to identify the corresponding direction consistency state for each pair of data. For the spatial distribution difference and temporal evolution difference after pairing and arrangement, a weighted combination calculation is performed on each data point. The corresponding weights are determined according to the numerical ratio between the spatial distribution difference and the temporal evolution difference, and the weights are adjusted in combination with the consistency results of the change direction to form a combined difference sequence containing the weighted spatial distribution difference and the weighted temporal evolution difference. Based on the combined difference sequence, the weighted spatial distribution difference and weighted temporal evolution difference corresponding to each data point are fused and calculated to construct residual offset characterization parameters, and the residual offset characterization parameters are serialized and output according to the order of data points.

[0012] Preferably, S3 specifically includes the following steps: S301. The residual migration characterization parameters and evolution correlation parameters are paired according to the data point order, and the difference calculation of adjacent data points is performed for each paired data point. The change of the residual migration characterization parameters in continuous data points is extracted as the residual migration change information. At the same time, the change of the evolution correlation parameters in continuous data points is extracted as the evolution correlation change information. A joint calculation sequence is formed based on the two types of changes. S302. For the joint calculation sequence, perform separation mapping calculation on the residual offset change information and the evolution association change information. By comparing the difference and synchronization relationship of the two types of change information in adjacent data points, construct the offset decoupling parameter to characterize the independent change characteristics of the residual offset and the consistency association parameter to characterize the consistent relationship between the two types of change, and arrange them accordingly according to the order of data points. S303. Based on the offset decoupling parameter and the consistency association parameter, perform a combined metric calculation on each data point. By performing hierarchical mapping processing on the numerical ratio relationship between the offset decoupling parameter and the consistency association parameter, and combining the degree of consistency of their changes in continuous data points, perform a weighted superposition calculation. The combined result is used as the residual offset degree of the corresponding data point, and is serialized and output according to the order of the data points.

[0013] Preferably, S302 is as follows: For the residual offset change information and evolution association change information in the joint calculation sequence, corresponding pairing processing is performed according to the data point order, and the difference comparison calculation is performed on the residual offset change information of adjacent data points to form an offset difference sequence, and the difference comparison calculation is performed on the evolution association change information of adjacent data points to form an association difference sequence. Based on the offset difference sequence and the associated difference sequence, the synchronization relationship is determined for the residual offset change information and the evolutionary associated change information corresponding to each data point. By performing point-by-point comparison calculation on the change direction and change magnitude of the offset difference sequence and the associated difference sequence at the same data point position, a change difference relationship identification sequence is formed, and interval division is performed according to the data point order. For the change difference relationship identifier sequence, a separation mapping calculation is performed on the residual offset change information and the evolution association change information. The residual offset change information in the data point interval corresponding to the non-consistent state in the change difference relationship identifier sequence is extracted to form an offset decoupling parameter. At the same time, the residual offset change information in the data point interval corresponding to the consistent state in the change difference relationship identifier sequence is combined with the evolution association change information to form a consistent association parameter, and then arranged in the corresponding order according to the data point order.

[0014] Preferably, S4 is as follows: The residual offset degree and the consistency correlation parameter are paired according to the data point order, and a coupling calculation is performed on each pair of data. After the residual offset degree and the consistency correlation parameter are processed by a unified scale, a numerical ratio division is performed, and the ratio division result is mapped to the corresponding level weight. At the same time, the weight is adjusted according to the stability of the change of the consistency correlation parameter in continuous data points, and the adjusted weight is weighted and calculated with the residual offset degree to construct the consistency evaluation parameter and arrange it according to the data point order. Based on the consistency assessment parameters, consistency determination calculations are performed on each data point. The consistency assessment parameters are divided into intervals, and the interval to which the consistency assessment parameters belong is determined for each data point. The corresponding interval is marked as a consistent state or a non-consistent state. At the same time, the continuity of the consistency state distribution in continuous data points is verified, and the consistency determination in the industrial data quality consistency test is completed. By combining the consistency judgment result and the residual offset degree, alignment processing calculation is performed on each data point. The position of the data point in the feature space is offset correction according to the residual offset degree, and the correction magnitude is controlled in stages according to the consistency judgment result. The corrected data is arranged according to the order of the data points to achieve alignment processing in data fusion.

[0015] Preferably, S5 is as follows: Based on the data fusion results, the position difference calculation is performed on the corresponding data points in the feature space of the data before and after fusion. The difference results of each data point are decomposed and processed to extract the difference amount representing the offset change and the difference amount representing the correlation change. The offset change difference amount and the correlation change difference amount are combined and mapped to form feedback correction parameters that correspond one-to-one with the data points and are arranged in the order of the data points. Based on the feedback correction parameters, iterative update calculations are performed on the residual offset characterization parameters, the consistency correlation parameters, and the residual offset degree. By mapping the feedback correction parameters to the residual offset characterization parameters and the consistency correlation parameters respectively, numerical adjustment processing is performed on the two types of parameters. Based on the updated residual offset characterization parameters and the consistency correlation parameters, a joint correction calculation is performed on the residual offset degree, resulting in updated residual offset characterization parameters, consistency correlation parameters, and residual offset degree, which are arranged in the order of data points.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention achieves in-depth analysis and hierarchical representation of residual structures in multi-source industrial data by constructing a complete processing chain of "residual multi-domain mapping—asynchronous offset identification—offset decoupling—consistency determination—alignment processing—feedback correction." Compared with existing technologies that only rely on residual values ​​for consistency judgment, this invention can distinguish between model biases and data differences mixed in the residuals. By introducing spatial positioning parameters and evolutionary correlation parameters, the residuals are expanded from single numerical values ​​to a multi-dimensional structure with spatial distribution and temporal evolution characteristics. Furthermore, asynchronous offsets are identified through cross-comparison operations and trend analysis, enabling the system to maintain accurate judgment of data consistency even when complex offset behaviors exist in the residuals. Based on this, by constructing offset decoupling parameters and consistency correlation parameters, the structural decomposition of offset sources is achieved, thereby avoiding misjudging model fitting errors as data inconsistencies and improving the authenticity and reliability of consistency test results.

[0017] 2. This invention constructs a consistency evaluation parameter by coupling the residual offset degree with a consistency correlation parameter. Based on this parameter, it performs hierarchical consistency judgment and differential alignment processing, enabling the data fusion process to have targeted adjustment capabilities. While ensuring the stability of highly consistent data, it performs fine-grained position correction on low-consistency data, thereby improving the overall accuracy and structural rationality of the fusion results. Simultaneously, by introducing a feedback correction parameter based on the fusion results, the residual offset characterization parameter, the consistency correlation parameter, and the residual offset degree are iteratively updated, giving the system dynamic adaptive capabilities. This allows the system to continuously optimize the internal parameter expression as the data distribution changes, avoiding the cumulative error problem caused by static models. Overall, this technical solution not only improves the accuracy of industrial data quality consistency testing but also enhances the robustness and stability of the data fusion process, making it suitable for the high-reliability fusion requirements of multi-source heterogeneous data in complex industrial scenarios. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a schematic diagram of the process of the present invention.

[0020] Figure 2 This is a schematic diagram of the multi-source industrial data reconstruction and residual analysis process of the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the changes in key indicators of the multi-source industrial data fusion results of the present invention.

[0022] Figure 4 This is a schematic diagram illustrating the time-series distribution of parameters characterized by the residual offset in this invention. Detailed Implementation

[0023] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0024] This invention provides, for example Figures 1-4 The data fusion method for industrial data quality consistency testing based on artificial intelligence, as shown, specifically includes the following steps: S1. Perform artificial intelligence reconstruction processing on multi-source industrial data and obtain residual sequences. Construct residual multi-domain mapping sequences containing spatial positioning parameters and evolutionary correlation parameters based on the residual sequences. In this embodiment, S1 specifically refers to: The multi-source industrial data is reconstructed using artificial intelligence. The multi-source industrial data is input into the corresponding artificial intelligence reconstruction model for mapping calculation, generating a reconstructed data sequence corresponding to the multi-source industrial data. The difference between the multi-source industrial data and the reconstructed data sequence is calculated for each data point to obtain the residual sequence. When performing AI-based reconstruction on multi-source industrial data, data from different sources can be aligned and organized according to timestamps or sampling order, and then input into corresponding AI reconstruction models for mapping calculations. These AI reconstruction models can be trained on historical industrial data to learn the inherent patterns of change and structural characteristics of the data. For example, a neural network based on sequence learning can be used to predict and reconstruct time-series data, or a network based on encoding and decoding structures can be used to reconstruct the data. In the specific implementation process, each set of input data generates a reconstructed data sequence that corresponds one-to-one with the original input data after model processing. Then, the original data and the reconstructed data are compared point by point at the same position. The residual value is obtained by performing difference calculations on the data at each corresponding position, and all residual values ​​are arranged in the original order to form a residual sequence. For example, when there is a deviation between the original collected value and the model output value at a certain moment, this deviation constitutes the residual at that position. In this way, the data differences that the model cannot fully fit can be explicitly extracted, providing a basis for subsequent analysis.

[0025] Multi-source industrial data refers to data sets from different devices, acquisition channels, or systems, which may differ in time distribution, numerical range, and variation patterns. Artificial intelligence (AI) reconstruction processing refers to the process of regenerating or predicting input data using a trained model, making the output data structurally approximate the input data. An AI reconstruction model is a computational model used to perform this reconstruction processing, containing the ability to learn data features. Mapping computation represents the process of converting input data into output data through the model's internal computation. The reconstructed data sequence corresponding to multi-source industrial data indicates that the model's output data maintains a consistent relationship with the input data in terms of order and position. Point-by-point difference calculation represents comparing the original data and reconstructed data one by one at the same position and calculating the difference. The residual sequence is a data sequence formed by arranging all differences in the order of the original data, used to characterize the distribution of differences between the model's reconstruction results and the original data; these differences can reflect information in the data that was not fully expressed by the model.

[0026] For the residual sequence, position mapping calculation is performed in a unified feature space to extract the distribution coordinates of each data point in the residual sequence. Then, cluster center calculation and relative position offset calculation are performed on the distribution coordinates to construct spatial positioning parameters based on the distribution center position and offset vector. When spatially representing the residual sequence, each data point in the residual sequence can be projected into the same feature space according to a unified mapping rule. For example, each residual data point can be mapped to a coordinate point with multidimensional attributes based on the magnitude, trend, or temporal adjacency of the residual value. In practice, a multidimensional vector representation can be constructed to combine the residual value with its preceding and following changes to form a coordinate description. Then, a distribution analysis is performed on all coordinate points. By traversing the value range of all coordinate points, the overall distribution center position is calculated, and the positional difference of each coordinate point relative to the distribution center is further calculated to obtain the offset direction and degree, thus forming the offset vector of each data point. For example, when a certain segment of residual data is generally large, its mapped coordinates will be concentrated in a certain region of the feature space. By calculating the center of this region and analyzing the distance between each point and the center, the spatial distribution characteristics of the residual can be determined. The purpose of this processing is to transform the original one-dimensional residual information into spatial distribution information, so that the overall offset and local fluctuations can be distinguished in the future.

[0027] A unified feature space refers to establishing a common coordinate representation system for residual data from different sources, enabling all residual data to be compared at the same scale and dimension. Position mapping calculation refers to transforming residual data into coordinate points in the feature space through a pre-defined mapping relationship. The distribution coordinates of each data point in the residual sequence represent the position of each residual data point in the feature space. Cluster center calculation involves performing overall statistical analysis on all distribution coordinates to determine their clustering location, reflecting the overall distribution of the residuals. Relative position offset calculation calculates the positional difference of each distribution coordinate relative to the cluster center, describing the degree of deviation of a single data point. A spatial positioning parameter based on the distribution center position and offset vector is a descriptive result formed by combining the cluster center position with the offset relationship of each data point. This parameter can simultaneously reflect the overall distribution and individual offset characteristics, used for subsequent residual offset analysis and consistency determination.

[0028] After completing the construction of spatial positioning parameters, differential calculation and trend extraction are performed on adjacent data points of the residual sequence in continuous time series to form a time series change sequence. Correlation calculation is then performed on the time series change sequence to construct evolutionary correlation parameters. At the same time, according to the correspondence of data points, the spatial positioning parameters and evolutionary correlation parameters are combined and mapped at the sequence level to form a residual multi-domain mapping sequence.

[0029] After constructing the spatial positioning parameters, the residual sequence can be continuously arranged according to the sampling order, and differential calculations can be performed on the numerical changes between adjacent data points. By comparing the changes in adjacent residuals point by point, the direction and magnitude of the residual changes can be characterized. Based on this, continuous analysis is performed on the differential results to extract the trend of the residual changes over a period of time. For example, when a certain residual shows a continuous increase or continuous fluctuation, the trend characteristics can be reflected by the differential results. Then, all differential results are arranged in the original order to form a time-series change sequence. Further comparative analysis is performed on the change relationships between different positions in the time-series change sequence. The correlation degree is calculated by calculating the similarity or synchronicity between the change trends in different time periods, thereby obtaining evolutionary correlation parameters that can describe the relationship of residual changes over time. Subsequently, according to the position of each data point in the original residual sequence, the corresponding spatial positioning parameters and evolutionary correlation parameters are matched one by one and combined and arranged according to the data order. Finally, a residual multi-domain mapping sequence containing both spatial distribution information and temporal evolution information is formed. This expands the original single-dimensional residual information into multi-dimensional features, providing a more comprehensive data foundation for subsequent consistency analysis.

[0030] Continuous time series refers to a set of residual data arranged in chronological order, reflecting the changes in data over time. Difference calculation involves comparing adjacent data points to obtain the amount of change, used to characterize the local changes in the data. Trend extraction involves identifying the direction and pattern of data change over a period of time based on continuous difference results. A time series change sequence is a set of data formed by arranging all difference results in chronological order, used to represent the overall change process. Correlation calculation involves comparing change patterns between different time periods to quantify their similarity or synchronicity. Evolutionary correlation parameters are parameters obtained through correlation calculation, used to describe the evolutionary relationship of residuals over time. Spatial positioning parameters and evolutionary correlation parameters represent the feature information of residuals in the spatial and temporal dimensions, respectively. Sequence-level combination mapping involves pairing two types of parameters point by point according to the correspondence between data points to form a new sequence structure. Residual multi-domain mapping sequence is a data sequence that integrates spatial distribution characteristics and temporal evolution characteristics, used to comprehensively describe the multidimensional attributes of residuals.

[0031] Taking the consistency detection scenario of multi-source industrial data in the coating production line of lithium battery electrodes for new energy vehicles as an example, the collected industrial data includes coating temperature sensor data, roller pressure sensor data, and equipment vibration monitoring data. The temperature data sampling period is 1 second, the pressure data sampling period is 500 ms, and the vibration data sampling frequency is 100 Hz. Data was continuously collected for 72 hours, yielding approximately 268,000 sets of valid industrial data. During data processing, data from different acquisition channels were first time-aligned according to a unified timestamp, and then input into a pre-trained artificial intelligence reconstruction model for reconstruction. In this embodiment, the artificial intelligence reconstruction model uses a temporal reconstruction network based on an LSTM-AutoEncoder structure. The encoding end contains two LSTM hidden layers, each with 128 hidden units, and the decoding end uses a symmetric LSTM decoding structure. During model training, historical industrial data from the production line under normal operating conditions for 15 consecutive days was used as training samples. The loss function was the mean squared error function, the training epochs were set to 100, and the learning rate was set to 0.001. The processing size is set to 64. Through model training, the reconstructed output can learn the normal variation patterns of industrial data. After the model completes training, the real-time collected multi-source industrial data is input into the artificial intelligence reconstruction model to generate the corresponding reconstructed data sequence. The original industrial data and the reconstructed data sequence are then subjected to point-by-point difference calculation to obtain the residual sequence. The residual value Rt is calculated according to Rt=Xt-X^t, where Xt represents the original collected value and X^t represents the model reconstruction value at the corresponding time. For example, when the original collected temperature is 86.3℃ and the model reconstruction value is 84.9℃ at a certain time, the corresponding residual value is 1.4. A continuous residual sequence is formed by performing point-by-point difference calculation on all sampling points.

[0032] like Figure 2 The diagram illustrates the multi-source industrial data reconstruction and residual analysis process of this invention. After multi-source industrial data is input into the artificial intelligence reconstruction model, a time-series reconstruction process is performed on the industrial data using an LSTM-AutoEncoder structure, generating corresponding reconstructed data. Subsequently, point-by-point difference calculations are performed between the original industrial data and the reconstructed data to form a residual sequence, thereby enabling the analysis of abnormal offset characteristics in the industrial data.

[0033] For the aforementioned residual sequence, this embodiment constructs a unified feature space to represent the spatial distribution of the residual data. Specifically, each residual data point is mapped to a three-dimensional feature vector containing the residual magnitude, local rate of change, and temporal adjacency difference. For example, the feature vector corresponding to a certain residual point is (1.4, 0.32, -0.15). The K-means clustering algorithm is used to calculate the cluster center of all feature vectors to obtain the coordinates of the residual distribution center (0.27, 0.04, 0.01). The position offset distance and offset direction vector of each residual data point relative to the cluster center are further calculated to form spatial positioning parameters. In the temporal evolution analysis, the residual sequence is subjected to sliding window differencing calculation in chronological order. The sliding window length is set to 10 sampling periods. The change in adjacent residuals is calculated using ΔRt = Rt - R(t-1), and the residual change trend is extracted based on the continuous differencing results to form a temporal change sequence. Furthermore, the Pearson correlation coefficient is used to calculate the correlation between the change trends in different time windows. For example, when the correlation coefficient between two time periods reaches 0.93, it indicates that the corresponding residual change trends have high synchronicity, thus generating evolutionary correlation parameters. Subsequently, according to the correspondence of residual data points, the spatial positioning parameters and evolutionary correlation parameters are combined and mapped point by point to form a residual multi-domain mapping sequence that simultaneously contains spatial distribution information and temporal evolution information.

[0034] Sampling time drift and local data misalignment anomalies were introduced into the experiment. The time drift range was set to 2 to 5 seconds, and the local data misalignment ratio was set to 5%. The traditional mean fluctuation detection method and the method described in this embodiment were used to identify and analyze the anomalies. Experimental results show that the traditional mean fluctuation detection method has an accuracy rate of 81.6% in identifying anomalies, while the method described in this embodiment improves the accuracy rate to 95.2%. Furthermore, after the fusion of multi-source industrial data, the Pearson consistency coefficient between the data increased from 0.78 to 0.94, indicating that this embodiment can effectively identify asynchronous offset problems in multi-source industrial data and improve the consistency analysis capability and data fusion alignment accuracy between different industrial data.

[0035] S2. Perform cross-comparison operation on the spatial positioning parameters and evolution correlation parameters in the residual multi-domain mapping sequence to determine whether there is asynchronous offset in the feature space of the residual after the multi-source industrial data is reconstructed by artificial intelligence. If there is asynchronous offset, form residual offset characterization parameters. In this embodiment, S2 specifically includes the following steps: S201. The spatial positioning parameters and evolutionary correlation parameters in the residual multi-domain mapping sequence are synchronously paired according to the correspondence of data points, and numerical difference calculation is performed for each paired data to obtain the numerical difference between the spatial positioning parameters and the evolutionary correlation parameters. At the same time, the direction of change of the spatial positioning parameters corresponding to adjacent data points and the direction of change of the evolutionary correlation parameters are determined to be consistent. Based on the numerical difference and direction consistency results, a cross-comparison result sequence containing spatial distribution difference and temporal evolution difference is constructed. When processing residual multi-domain mapping sequences, spatial positioning parameters and evolutionary correlation parameters can be matched one-to-one according to the positional order of data points in the original sequence. This ensures that each data point possesses both spatial distribution information and temporal evolution information. Then, numerical difference calculation is performed on each pair of matched data. By subtracting the values ​​of spatial positioning parameters and evolutionary correlation parameters point by point, the numerical difference reflecting the degree of difference between the two is obtained. Simultaneously, the changing direction of spatial positioning parameters and the changing direction of evolutionary correlation parameters are calculated between adjacent data points. For example, the changing trend direction is determined by comparing the numerical magnitude of the current data point with that of the previous data point. Then, the consistency of the two directions is judged. If the changing trend directions are the same, they are judged as consistent; if the changing trend directions are opposite, they are judged as inconsistent. Finally, the numerical difference and directional consistency results corresponding to each data point are combined and arranged in the original order to form a cross-comparison result sequence. For example, when the spatial positioning parameter continuously increases while the evolutionary correlation parameter continuously decreases in a certain data segment, it can be identified as directional inconsistency. The degree of deviation can be further quantified by numerical difference, thereby achieving a unified characterization of the difference features of data from different sources.

[0036] In the residual multi-domain mapping sequence, spatial positioning parameters and evolutionary correlation parameters are used to represent the feature information of the residual in the spatial and temporal dimensions, respectively; the data point correspondence represents the matching relationship between spatial positioning parameters and evolutionary correlation parameters at the same location; synchronous pairing processing refers to combining the two types of parameters one-to-one according to the order of data points; numerical difference calculation refers to calculating the difference value point by point for the paired two types of parameters to obtain the difference value; the numerical difference between spatial positioning parameters and evolutionary correlation parameters is used to reflect the degree of difference between spatial distribution and temporal evolution; the change direction of spatial positioning parameters corresponding to adjacent data points and the change direction of evolutionary correlation parameters are used to describe the change trend of the two types of parameters on continuous data points; directional consistency judgment is to compare the two types of change trends to determine whether they are consistent; the cross-comparison result sequence containing spatial distribution difference and temporal evolution difference is a sequence structure formed by combining numerical difference and directional consistency results, used to simultaneously express the degree of difference and change relationship.

[0037] S202. Based on the cross-comparison result sequence, perform joint discrimination calculation on the spatial distribution difference and the temporal evolution difference. By judging the consistency of the change trend and change direction of the difference in continuous data points, determine whether there is asynchronous shift in the feature space of the residual after the reconstruction of multi-source industrial data by artificial intelligence. S203. In the case of asynchronous offset, the spatial distribution difference and temporal evolution difference in the cross-comparison result sequence are weighted and combined, and the residual offset characterization parameter is constructed by combining the consistency result of the change direction. The residual offset characterization parameter is serialized and output according to the data point order.

[0038] In this embodiment, S202 specifically refers to: Based on the cross-comparison result sequence, the spatial distribution difference and the temporal evolution difference are arranged continuously according to the data point order. The spatial distribution difference corresponding to adjacent data points is differentially calculated to form a spatial difference change sequence, and the temporal evolution difference corresponding to adjacent data points is differentially calculated to form an evolution difference change sequence. When processing the cross-comparison result sequence, we can first arrange the spatial distribution difference and temporal evolution difference point by point according to the position order of the data points in the original sequence, so that the two types of differences maintain a consistent arrangement structure in the time dimension. Then, we perform difference calculation on the spatial distribution difference between adjacent data points, and obtain the magnitude and direction of change of the spatial distribution difference by comparing the change of the difference between the current data point and the previous data point. We then arrange all the calculation results in order to form a spatial difference change sequence. At the same time, we perform the same form of difference calculation on the temporal evolution difference between adjacent data points to extract the magnitude and trend of change of the temporal evolution difference, and form an evolution difference change sequence. For example, when the spatial distribution difference gradually increases while the temporal evolution difference changes relatively slowly in a certain data segment, we can form an evolution difference change sequence. When the situation is relatively stable, differential calculation can be used to identify the different rhythms of the two types of differences. The purpose of this processing is to transform static difference information into dynamic change information, so that asynchronous features can be identified from the perspective of change trend. Among them, the cross-comparison result sequence is a data sequence structure formed by combining spatial distribution difference and temporal evolution difference. The continuous arrangement processing according to the order of data points refers to the orderly organization of data according to time or sampling order. Differential calculation refers to the numerical comparison of adjacent data points to obtain the change. The spatial difference change sequence is a change sequence formed by differential calculation of spatial distribution difference, which is used to describe the dynamic change process of spatial difference. The evolution difference change sequence is a change sequence formed by differential calculation of temporal evolution difference, which is used to describe the evolution process of time dimension difference.

[0039] For spatial difference change sequences and evolutionary difference change sequences, the change trend corresponding to each data point is marked with a direction. For the same data point, the spatial difference change direction and the evolutionary difference change direction are compared and calculated to determine the consistency of the signs. When the signs of the two directions are the same, they are marked as consistent. When the signs of the two directions are opposite, they are marked as inconsistent. The consistent and inconsistent states are then sequentially arranged according to the order of the data points to form a sequence of the degree of consistency of the change direction. When processing spatial and evolutionary variation sequences, the direction of change can be determined for the trend corresponding to each data point. For example, by comparing the magnitude of the change value between the current data point and the previous data point, the trend can be classified as increasing, decreasing, or remaining stable. Furthermore, increasing is marked as positive, decreasing as negative, and stable as zero, thus completing the direction identification process. Then, for the same data point, the direction of change in the spatial variation sequence is compared point by point with the direction of change in the evolutionary variation sequence. When both direction indicators are positive or both are negative, the direction is considered consistent; when one is positive and the other is negative, the direction is considered inconsistent. For example, if the spatial variation at a certain location shows an increasing trend while the evolutionary variation shows a decreasing trend, it can be identified as inconsistent. Subsequently, the determination results for each data point are arranged in their original order, marking those with consistent directions as consistent and those with inconsistent directions as inconsistent. The data points are marked as non-consistent states, ultimately forming a sequence of consistency in the direction of change. This process unifies the relationship between the two types of differences into a discrete state sequence, facilitating subsequent interval statistics and asynchronous offset identification. The change trend corresponding to each data point represents the directional characteristics of the difference changes between consecutive data points. Direction identification processing converts the change trend into a comparable direction category. The spatial difference change direction and the evolutionary difference change direction at the same data point location respectively represent the change orientation of the two types of differences at that location. Sign consistency comparison calculation is the process of matching and judging the two types of direction identifiers. If the direction signs are the same, the change trends are consistent, corresponding to a consistent state; if the direction signs are opposite, the change trends are opposite, corresponding to a non-consistent state. Serialization arranges the judgment results according to the order of the data points. The sequence of consistency in the direction of change is a sequence composed of consistent and non-consistent states, used to describe the degree of matching of the overall direction of change.

[0040] Based on the sequence of consistency of change direction, interval statistical processing is performed on the distribution of consistent and inconsistent states in continuous data points. Combined with the continuity of the change trend of spatial difference change sequence and evolutionary difference change sequence, joint discrimination calculation is performed. When there are continuous inconsistent state intervals in the sequence of consistency of change direction and the corresponding change trend does not have synchronous change characteristics, it is determined that the residual of multi-source industrial data after artificial intelligence reconstruction has asynchronous offset in the feature space.

[0041] When processing sequences of consistent change directions, we can first scan the consistent and inconsistent states continuously according to the order of data points. By traversing point by point, we can identify continuous intervals of the same state and record the start and end positions of each continuous inconsistent state, thus completing interval statistical processing. Subsequently, for each continuous inconsistent state interval, we combine the spatial difference change sequence and the evolutionary difference change sequence at the corresponding position to perform a continuity analysis of the change trends of the two types of change sequences within the interval. For example, by observing whether the spatial difference change within the interval shows a continuous unidirectional change and whether the evolutionary difference change shows different directions or fluctuations, we can identify whether the two types of change trends maintain a consistent rhythm. When the spatial difference change shows a stable increase while the evolutionary difference change shows a decrease or irregular fluctuations within a certain continuous interval, it can be determined that the change trends are asynchronous. Then, based on the interval statistical results and trend analysis results, we perform a joint discriminant calculation. In other words, under the condition that there are continuous non-consistent state intervals and the two types of change trends within the corresponding intervals do not have synchronous change characteristics, the existence of asynchronous offset is determined. This processing can avoid misjudgment caused by single-point anomalies and improve the stability of the judgment through interval-level continuity analysis. Among them, interval statistical processing refers to the process of segmenting and identifying the location of continuous identical states. The continuity of the change trend of spatial difference change sequence and evolutionary difference change sequence refers to the continuous characteristics of the change direction and change amplitude within a data interval. Joint discriminant calculation is the process of comprehensively judging the state distribution and trend continuity. The existence of continuous non-consistent state intervals in the change direction consistency sequence indicates that there is a continuous data point with inconsistent direction. The corresponding change trend does not have synchronous change characteristics, which means that the two types of changes do not show the same change direction or change rhythm within the interval. These conditions are used together to determine that there is asynchronous offset in the feature space of the residual.

[0042] In this embodiment, S203 specifically refers to: In cases where asynchronous offset is determined, spatial distribution difference and temporal evolution difference are extracted from the cross-comparison result sequence. The spatial distribution difference and temporal evolution difference are then paired and arranged according to the data point order. At the same time, the direction consistency result is combined to identify the corresponding direction consistency state for each pair of data. When asynchronous offset is detected, spatial distribution difference and temporal evolution difference can be extracted from the cross-comparison result sequence, and then arranged point by point according to the order of the data points in the original sequence, so that each data point position has both types of difference information. Then, the spatial distribution difference and temporal evolution difference at the same position are paired to form a pair of paired data. At the same time, combined with the previously obtained change direction consistency results, a direction label is added to each pair of paired data. For example, when the change direction consistency result corresponding to a certain data point is consistent, the paired data is marked as consistent in direction. When the change direction consistency result is inconsistent, it is marked as inconsistent in direction. In this way, a composite data structure containing difference and direction status can be formed at each data point position. For example, in a certain data segment, when the spatial distribution difference is large and the temporal evolution difference is small and the direction is inconsistent, the position can be clearly marked as a data point with significant offset and mismatched direction. The purpose of this processing is to integrate the difference information and direction consistency information before entering the weighted calculation, so as to provide a structured input for the construction of the subsequent residual offset characterization parameters.

[0043] Spatial distribution difference indicates the degree of difference between spatial positioning parameters and evolutionary correlation parameters at the numerical level, reflecting deviations in the spatial dimension. Temporal evolution difference indicates the degree of difference between the two types of parameters in the temporal relationship, reflecting differences in changes in the temporal dimension. Paired arrangement processing refers to combining the two types of difference quantities one-to-one according to the order of data points, so that each data point position forms a paired data structure. The consistency result of change direction indicates the matching status of spatial difference change and evolutionary difference change in direction. The corresponding direction consistency status is to attach the consistent or inconsistent judgment result to each pair of paired data in the form of an identifier, thereby forming a combined data containing numerical differences and directional relationships, which is used for subsequent offset characterization and fusion decision analysis.

[0044] For the spatial distribution difference and temporal evolution difference after pairing and arrangement, a weighted combination calculation is performed on each data point. The corresponding weights are determined according to the numerical ratio between the spatial distribution difference and the temporal evolution difference, and the weights are adjusted in combination with the consistency results of the change direction to form a combined difference sequence containing the weighted spatial distribution difference and the weighted temporal evolution difference. When processing the spatial distribution differences and temporal evolution differences after pairing and arrangement, a weighted combination calculation can be performed for the two types of differences corresponding to each data point. First, the weight allocation is determined based on the relative magnitude of the two types of differences. For example, when the spatial distribution difference in a data point is significantly greater than the temporal evolution difference, the spatial distribution difference can be given a higher weight, while when the two values ​​are close, they are assigned similar weights. Then, the weights are adjusted based on the consistency of the change direction. For example, when the change direction of the corresponding data point is consistent, the weights of the two types of differences are adjusted to be balanced, while when the change direction is inconsistent... When the difference has a large change, the weight ratio of the difference is increased to strengthen the expression of the non-consistency feature. Then the adjusted weights are applied to the spatial distribution difference and the temporal evolution difference respectively to obtain the weighted spatial distribution difference and the weighted temporal evolution difference. They are then combined and arranged according to the order of the data points to form a combined difference sequence. For example, if the spatial distribution difference is large and the direction is inconsistent in a certain data point, the difference will account for a higher proportion in the combined result after the weight adjustment, thus highlighting the offset feature at that position. This process can integrate the difference information from different sources according to the importance, making the subsequent residual offset characterization more refined.

[0045] Weighted combination calculation refers to the process of fusing different types of difference data according to a certain proportion. The numerical ratio between spatial distribution difference and temporal evolution difference indicates the relative magnitude of the two types of differences in terms of numerical value, which is used to determine their respective degree of influence. The corresponding weight is a proportional coefficient allocated according to the numerical ratio, which is used to control the contribution of the two types of differences in the combination. The consistency of change direction result indicates whether the change direction of the two types of differences is consistent. Adjusting the weight refers to correcting the weight according to the consistency or inconsistency of the direction. The weighted spatial distribution difference is the spatial difference value after the weighting, and the weighted temporal evolution difference is the temporal difference value after the weighting. The combined difference sequence containing the weighted spatial distribution difference and the weighted temporal evolution difference is a data sequence formed by combining the two types of weighted results in the order of data points, which is used to reflect the comprehensive difference of each data point.

[0046] Based on the combined difference sequence, the weighted spatial distribution difference and weighted temporal evolution difference corresponding to each data point are fused and calculated to construct residual offset characterization parameters, and the residual offset characterization parameters are serialized and output according to the order of data points.

[0047] When further processing the combined difference sequence, a fusion calculation can be performed on the weighted spatial distribution difference and the weighted temporal evolution difference corresponding to each data point. The two types of weighted difference values ​​are jointly mapped at the same scale. For example, the two types of difference values ​​can be normalized and aligned before comprehensive calculation, so that differences in different dimensions can be fused under a unified standard. Then, a single characterization value is generated as a residual offset characterization parameter based on the relative change relationship between the two types of differences. For example, when the weighted spatial distribution difference and the weighted temporal evolution difference in a certain data point are both large, the fusion result at that position is also relatively large, thus reflecting that the offset degree of the data point is high. When both are small, it indicates that the offset degree is low. Subsequently, the fusion results corresponding to each data point are arranged according to the original data order to form a continuous residual offset characterization parameter sequence. In this way, multi-dimensional difference information can be compressed into a single-dimensional characterization result, which is convenient for subsequent consistency judgment and fusion processing.

[0048] The weighted spatial distribution difference and weighted temporal evolution difference for each data point represent the degree of difference after weight adjustment in the spatial and temporal dimensions, respectively. Fusion calculation refers to the process of jointly processing these two types of difference information on a unified scale to form a single representation. The residual offset characterization parameter is a value obtained through fusion calculation and is used to describe the comprehensive offset degree of each data point. Serialized output means that all residual offset characterization parameters are organized in an orderly manner according to the data point order to form a continuous data sequence. This sequence can reflect the offset distribution of the overall data at each position.

[0049] After constructing the residual multi-domain mapping sequence, a cross-comparison operation is performed on the spatial positioning parameters and evolutionary correlation parameters in the residual multi-domain mapping sequence to identify whether there are asynchronous offsets in the multi-source industrial data after artificial intelligence reconstruction. Specifically, the spatial positioning parameters and evolutionary correlation parameters are first synchronously paired according to the temporal order of the data points in the original sequence, so that each data point corresponds to both spatial distribution information and temporal evolution information. The spatial positioning parameters are represented by the normalized offset distance of the residual data points in a unified feature space, and the evolutionary correlation parameters are represented by the Pearson correlation coefficient of the residual change trend between adjacent time windows. For example, at the 15240th sampling point, the corresponding spatial positioning parameter is 0.814 and the evolutionary correlation parameter is 0.563. The numerical difference is calculated as 0.251 through point-by-point difference. At the same time, the direction of change of the spatial positioning parameter and the direction of change of the evolutionary correlation parameter between adjacent data points are judged to be consistent. When the spatial positioning parameter increases from 0.768 to 0.814, while the evolutionary correlation parameter decreases from 0.611 to 0.563, it is judged that the spatial direction changes in a positive direction and the evolutionary direction changes in a negative direction. Since the two directions of change are inconsistent, it is marked as a state of inconsistent direction. Then, the numerical difference results and the direction consistency results are combined and arranged according to the sampling order to form a cross-comparison result sequence, thereby realizing a unified expression of spatial distribution differences and temporal evolution differences.

[0050] After forming the cross-comparison result sequence, continuous difference calculations are performed on the spatial distribution difference and the temporal evolution difference to extract the dynamic change characteristics of the two types of differences during continuous sampling. Specifically, the spatial distribution difference is differentially calculated according to the sampling order, using the formula ΔSi=Si-S(i-1) to calculate the change in spatial difference between adjacent data points. Simultaneously, the temporal evolution difference is differentially calculated using the formula ΔTi=Ti-T(i-1) to form an evolutionary difference change sequence. For example, in three consecutive sampling points, the spatial distribution difference is 0.183, 0.276, and 0.412, corresponding to spatial difference changes of 0.093 and 0.136, indicating a continuous increasing trend in spatial difference. Meanwhile, the corresponding temporal evolution difference is 0.337, 0.318, and 0.291, corresponding to evolutionary difference changes of -0.019 and -0.027, indicating a decreasing trend in temporal evolution difference. Subsequently, directional labeling processing is performed on the spatial difference change sequence and the evolutionary difference change sequence. Positive values ​​are marked as positive changes, negative values ​​are marked as negative changes, and zero values ​​are marked as stable states. A sign consistency comparison calculation is then performed on the same data point location. When the two types of changes have the same direction, they are marked as consistent states, and when the two types of changes have opposite directions, they are marked as inconsistent states, thus forming a sequence of the degree of consistency of change direction.

[0051] After generating a sequence of consistent change directions, continuous interval statistical processing is performed on consistent and inconsistent states. This is combined with a joint judgment based on the continuity of the change trends in the spatial difference change sequence and the evolutionary difference change sequence to determine if asynchronous shifts exist. Specifically, the sequence of consistent change directions is scanned point-by-point to identify consecutive inconsistent state intervals, and the start and end positions of each interval are recorded. For example, if 79 consecutive inconsistent state data points appear between sampling points 15240 and 15318, and the spatial difference change sequence within the corresponding interval shows a continuously increasing trend while the evolutionary difference change sequence shows a fluctuating decreasing trend, then it is determined that there is a significant asynchronous change trend in this interval. Further joint judgment is performed by combining an interval length threshold and the continuity of the change trend. If the length of a consecutive inconsistent state interval exceeds a preset threshold of 20 sampling points, and the spatial difference change trend and the evolutionary difference change trend within the corresponding interval do not exhibit synchronous change characteristics, then asynchronous shifts are determined to exist in this region. This method avoids misjudgments caused by a single outlier, thereby improving the stability and reliability of the asynchronous shift identification process.

[0052] After determining the existence of asynchronous offset, a weighted combination calculation is performed on the spatial distribution difference and temporal evolution difference in the cross-comparison result sequence, and residual offset characterization parameters are constructed by combining the consistency results of change direction. Specifically, firstly, the spatial distribution difference and temporal evolution difference are paired and arranged point by point according to the data point order. For example, if the spatial distribution difference for a certain data point is 0.624 and the temporal evolution difference is 0.381, and the consistency result of change direction is inconsistent, then the initial weight of the spatial distribution difference is determined to be 0.621 and the initial weight of the temporal evolution difference is 0.379 based on the proportional relationship between the two types of difference values. Since the data point is in a non-consistent direction state, the weight of the spatial distribution difference with a large change amplitude is further increased to 0.708, while the weight of the temporal evolution difference is adjusted to 0.292. Subsequently, the weighted spatial distribution difference and the weighted temporal evolution difference are calculated separately, and a fusion calculation is performed through a normalized joint mapping method to generate the corresponding residual offset characterization parameters. For example, the residual offset parameter obtained for this data point is 0.836, indicating a significant data offset characteristic at this location. Conversely, when the residual offset parameter for some data points is below 0.200, it indicates a lower degree of offset at the corresponding location. Finally, the residual offset parameters corresponding to all data points are serialized and arranged in the original sampling order to form a continuous sequence of residual offset parameters.

[0053] Based on the data from the aforementioned lithium battery electrode coating production line, a normal operating condition dataset and an abnormal offset dataset were constructed. The abnormal dataset was created by artificially introducing a sampling time drift of 2 to 5 seconds and a 5% data misalignment. The traditional anomaly detection method based on mean fluctuation and the method described in this embodiment were then compared and tested. Under the same experimental environment, the traditional mean fluctuation detection method achieved an accuracy of 82.4% and a false detection rate of 11.3% for identifying asynchronous offsets, while the method described in this embodiment achieved an accuracy of 96.1% and a false detection rate reduced to 3.8%. Furthermore, in the presence of continuous offset intervals, the average positioning error of this embodiment for the offset interval was 3.2 sampling points, while the average positioning error of the traditional method was 14.7 sampling points. This demonstrates that this embodiment can effectively identify asynchronous offset features in multi-source industrial data and improve the accuracy and stability of residual offset characterization.

[0054] S3. Jointly calculate the residual migration characterization parameter and the evolution correlation parameter to construct the migration decoupling parameter and the consistency correlation parameter, and determine the degree of residual migration based on the joint calculation results; In this embodiment, S3 specifically includes the following steps: S301. The residual migration characterization parameters and evolution correlation parameters are paired according to the data point order, and the difference calculation of adjacent data points is performed for each paired data point. The change of the residual migration characterization parameters in continuous data points is extracted as the residual migration change information. At the same time, the change of the evolution correlation parameters in continuous data points is extracted as the evolution correlation change information. A joint calculation sequence is formed based on the two types of changes. When processing residual migration parameters and evolutionary correlation parameters, a one-to-one correspondence can be established according to the order of data points in the original sequence, so that each position contains both residual migration parameters and evolutionary correlation parameters. Then, difference calculations are performed on adjacent data points. For example, by comparing the changes in the values ​​of the current data point and the previous data point, the changes in residual migration parameters and evolutionary correlation parameters between consecutive data points are calculated, thereby obtaining dynamic change information of the two types of parameters. The change in residual migration parameters reflects the trend of migration degree in the time dimension, and the change in evolutionary correlation parameters reflects the trend of correlation in the time dimension. For example, when the residual migration parameters continuously increase while the evolutionary correlation parameters change relatively slowly in a certain data segment, the difference between migration change and correlation change can be identified. Then, the two types of changes are combined and arranged according to the order of data points to form a joint calculation sequence, so that each data point simultaneously has migration change information and correlation change information, thus providing a foundation for subsequent separation mapping and migration decoupling calculations.

[0055] The residual offset characterization parameter and the evolutionary correlation parameter are used to represent the numerical characteristics of the residual in terms of the overall offset degree and the change correlation relationship, respectively. Correspondence pairing processing refers to matching the two types of parameters point by point according to the data point order, so that each position forms a paired data structure. Adjacent data point difference calculation refers to comparing the values ​​between consecutive data points to obtain the change amount. The change amount of the residual offset characterization parameter in consecutive data points is the offset change value obtained by difference calculation, which is used to describe the dynamic change of the offset degree. The residual offset change information is a data set composed of these changes. The change amount of the evolutionary correlation parameter in consecutive data points is the correlation change value obtained by the same difference calculation, which is used to describe the change of the correlation relationship. The evolutionary correlation change information is a data set composed of these changes. The joint calculation sequence is a sequence structure formed by combining the residual offset change information and the evolutionary correlation change information according to the data point order, which is used to express the two types of change characteristics simultaneously.

[0056] S302. For the joint calculation sequence, perform separation mapping calculation on the residual offset change information and the evolution association change information. By comparing the difference and synchronization relationship of the two types of change information in adjacent data points, construct the offset decoupling parameter to characterize the independent change characteristics of the residual offset and the consistency association parameter to characterize the consistent relationship between the two types of change, and arrange them accordingly according to the order of data points. S303. Based on the offset decoupling parameter and the consistency association parameter, perform a combined metric calculation on each data point. By performing hierarchical mapping processing on the numerical ratio relationship between the offset decoupling parameter and the consistency association parameter, and combining the degree of consistency of their changes in continuous data points, perform a weighted superposition calculation. The combined result is used as the residual offset degree of the corresponding data point, and is serialized and output according to the order of the data points.

[0057] When processing offset decoupling parameters and consistency association parameters, a combined metric calculation can be performed on the two types of parameters corresponding to each data point. First, the values ​​of the two types of parameters are compared and analyzed, and they are classified according to their relative magnitude. For example, the numerical ratio can be divided into three states: offset-dominant, association-dominant, and balanced. Then, a corresponding mapping level is set for different states, and this level is used as the basis for subsequent weight allocation. At the same time, the weights are adjusted based on the consistency of the changes of the two types of parameters in continuous data points. For example, when the changing trends of the two types of parameters are consistent in multiple consecutive data points, the influence of the offset decoupling parameter is reduced, while when the changing trends are inconsistent, the weight of the offset decoupling parameter is increased. Then, a weighted superposition calculation is performed on the adjusted two types of parameters, and the superposition result is used as the residual offset degree of the current data point. For example, if the offset decoupling parameter is large in a certain data point and its changing trend is inconsistent with the consistency association parameter, the superposition result will be significantly increased, thus reflecting a high degree of offset. Finally, the calculation results of all data points are arranged in the original order to form a continuous residual offset degree sequence for subsequent consistency determination and data fusion processing.

[0058] Combined metric calculation refers to the calculation process of uniformly quantifying multiple types of parameters, used to integrate information from different sources into a single evaluation result. The numerical ratio between offset decoupling parameters and consistent correlation parameters represents the relative proportion of the two types of parameters in terms of value, which is an important basis for determining the influence weight. Hierarchical mapping processing is the process of converting the numerical ratio into discrete levels to facilitate subsequent weight allocation. The consistency of the changes of the two in continuous data points is used to describe the synchronization of the changes of the two types of parameters in the time dimension. Weighted superposition calculation refers to the combined operation of the two types of parameters according to the weight to form a comprehensive result. The residual offset degree of the corresponding data point is a value obtained through combined calculation, used to represent the comprehensive offset intensity at that position. Serialization output arranges the residual offset degrees of all data points in order to form a continuous sequence, used to reflect the offset distribution of the overall data.

[0059] In this embodiment, S302 specifically refers to: For the residual offset change information and evolution association change information in the joint calculation sequence, corresponding pairing processing is performed according to the data point order, and the difference comparison calculation is performed on the residual offset change information of adjacent data points to form an offset difference sequence, and the difference comparison calculation is performed on the evolution association change information of adjacent data points to form an association difference sequence. When further processing the jointly calculated sequence, we can first arrange the residual offset change information and evolutionary association change information point by point according to the order of the data points in the original sequence, so that each position simultaneously has both types of change data. Then, we perform difference comparison calculation for adjacent data points, that is, we compare the residual offset change information of the current data point with that of the previous data point to obtain the change difference between adjacent positions, thus forming an offset difference sequence. At the same time, we perform the same form of difference comparison calculation on the evolutionary association change information to extract the change difference between adjacent data points and form an association difference sequence. For example, when the residual offset change information fluctuates greatly between adjacent points in a certain continuous data segment, while the evolutionary association change information changes less, it will show a large difference value in the offset difference sequence, while showing a small difference value in the association difference sequence. This processing method can further transform the change information of a single point into the change difference information between adjacent points, thereby more accurately depicting the differences in the rhythm and magnitude of change, and providing basic data support for subsequent synchronization relationship determination and decoupling analysis. Among them, the residual offset change information and evolutionary association change information in the joint calculation sequence refer to the combined data that simultaneously contains offset change and association change at the same data point location. The difference comparison calculation is the process of comparing the numerical changes between adjacent data points to obtain the difference amount. The offset difference sequence is a set of change differences formed by the residual offset change information after adjacent difference calculation, which is used to describe the fluctuation of offset change in the time dimension. The association difference sequence is a set of change differences formed by the evolutionary association change information after adjacent difference calculation, which is used to describe the fluctuation of association change in the time dimension.

[0060] Based on the offset difference sequence and the associated difference sequence, the synchronization relationship is determined for the residual offset change information and the evolutionary associated change information corresponding to each data point. By performing point-by-point comparison calculation on the change direction and change magnitude of the offset difference sequence and the associated difference sequence at the same data point position, a change difference relationship identification sequence is formed, and interval division is performed according to the data point order. When processing offset difference sequences and associated difference sequences, for each data point, the corresponding offset difference value and associated difference value can be compared and analyzed point by point. First, the direction of change is determined based on the positive and negative changes of adjacent differences. For example, by comparing the magnitude of the current difference value with the previous difference value, it can be determined whether the change is increasing or decreasing. At the same time, the magnitude of the change is assessed by combining the changes in the magnitude of the difference value. Then, the direction and magnitude of change of the offset difference sequence and the associated difference sequence at the same data point are matched and judged. When the change directions are consistent and the change trends of the magnitudes are similar, they are judged as synchronous changes. When the directions are opposite or the change trends of the magnitudes are inconsistent, they are judged as asynchronous changes. The judgment result of each data point is marked. For example, synchronous changes are marked as consistent, and asynchronous changes are marked as different. Arranged in the order of data points to form a sequence of change difference relationship marks. Then, the sequence of marks is divided into intervals. By continuously scanning the distribution of identical identifiers, consecutive consistent identifiers are divided into one interval, and consecutive differing identifiers are divided into another interval. For example, when a segment of data shows consecutive differing identifiers, a continuous difference interval can be formed. This process can expand the single-point judgment result into an interval-level structure, avoiding the impact of occasional fluctuations on the judgment result. Among them, the synchronous relationship judgment refers to the process of judging whether two types of difference changes have consistent change characteristics. The direction and magnitude of change of the offset difference sequence and the associated difference sequence at the same data point position represent the change trend and change intensity of the two types of differences at that position, respectively. The point-by-point comparison calculation is the calculation process of matching the direction and magnitude of each corresponding position. The change difference relationship identifier sequence is a sequence structure composed of the synchronous or asynchronous judgment results of each data point. The interval division process is the process of segmenting consecutive identical identifiers to describe the continuous distribution of change relationships in the time dimension.

[0061] For the change difference relationship identifier sequence, a separation mapping calculation is performed on the residual offset change information and the evolution association change information. The residual offset change information in the data point interval corresponding to the non-consistent state in the change difference relationship identifier sequence is extracted to form an offset decoupling parameter. At the same time, the residual offset change information in the data point interval corresponding to the consistent state in the change difference relationship identifier sequence is combined with the evolution association change information to form a consistent association parameter, and then arranged in the corresponding order according to the data point order.

[0062] When processing the sequence of identifiers indicating differences in changes, we can first divide the data points into multiple continuous intervals based on the distribution of consistent and inconsistent states in the identifier sequence, and then locate the data point intervals corresponding to inconsistent and consistent states. Next, for the data point intervals of inconsistent states, we extract the residual offset change information at the corresponding positions from the joint calculation sequence and arrange them continuously according to the data point order. We then collect the residual offset change information within this interval as independent change data, thus forming the offset decoupling parameter. For example, when all identifiers within a continuous interval are inconsistent states, the residual offset within that interval is... Change information is directly extracted and its change characteristics are retained. At the same time, for data point intervals in a consistent state, the residual offset change information and evolutionary correlation change information at the corresponding positions are combined and calculated point by point. For example, by synchronously superimposing or proportionally fusing the two types of change information, the two types of changes can be reflected in the same data structure to form a consistent correlation parameter, which is arranged according to the order of data points. Finally, two types of parameter structures containing independent offset information and correlation change information are obtained. This processing can separate the mixed change information and express the offset feature and correlation feature separately, thus providing a clear data basis for subsequent offset degree determination.

[0063] In the variation difference relationship identifier sequence, the data point interval corresponding to the inconsistent state refers to the interval where multiple consecutive data points show inconsistency in the direction or magnitude of change. These intervals reflect the difference between offset change and correlation change. The offset decoupling parameter is a data set composed of residual offset change information extracted from the inconsistent state interval, used to represent independent offset features unaffected by correlation change. In the variation difference relationship identifier sequence, the data point interval corresponding to the consistent state refers to the interval where multiple consecutive data points maintain consistency in the direction and magnitude of change. These intervals reflect the synchronous relationship between the two types of changes. Combined calculation refers to the process of jointly processing residual offset change information and evolution correlation change information at the same data point location. The consistent correlation parameter is a data set formed by combined calculation, used to represent the synergistic relationship between offset change and correlation change. These two types of parameters are used to describe different types of change characteristics and provide hierarchical structure support for subsequent data analysis.

[0064] After constructing the residual migration characterization parameters, joint calculations are performed using evolutionary correlation parameters to determine the migration variation characteristics of the residuals during continuous sampling. Specifically, firstly, the residual migration characterization parameters and evolutionary correlation parameters are paired point-by-point according to the chronological order of the data points in the original sequence, ensuring that each data point possesses both migration characterization information and correlation variation information. Subsequently, difference calculations are performed on adjacent data points, where the change in the residual migration characterization parameters is expressed as ΔPi = Pi. P(i 1) Perform calculations, using ΔEi=Ei for the changes in evolutionary correlation parameters. E(i) 1) Perform calculations. For example, at the 18620th sampling point, the residual offset characterization parameter changes from 0.684 to 0.793, corresponding to a residual offset change of 0.109, while the evolutionary correlation parameter changes from 0.522 to 0.537, corresponding to an evolutionary correlation change of 0.015. This indicates that the degree of offset change at this position is significantly higher than the degree of correlation change. Subsequently, the two types of changes are combined and arranged according to the order of data points to form a joint calculation sequence, so that each data point contains both offset change information and correlation change information, thereby providing a basic data structure for subsequent offset decoupling analysis.

[0065] After forming the joint calculation sequence, further difference comparison calculations are performed on the residual offset change information and evolutionary association change information in the joint calculation sequence to extract the difference in the change rhythm between adjacent data points. Specifically, difference calculations are performed on the residual offset change information between consecutive data points to form an offset difference sequence, and difference calculations are performed on the evolutionary association change information to form an association difference sequence. For example, in four consecutive sampling points, the residual offset changes are 0.071, 0.109, 0.146, and 0.183, respectively, and the corresponding offset difference values ​​are 0.038, 0.037, and 0.037, respectively, indicating a continuous increasing trend in offset change; while the corresponding evolutionary association changes are 0.019, 0.015, 0.012, and 0.011, respectively, and the corresponding association difference values ​​are respectively... 0.004 0.003 and A value of 0.001 indicates that the overall correlation change is relatively gradual. Subsequently, a synchronization relationship determination is performed on the offset difference value and the correlation difference value at the same data point location. When the two values ​​change in the same direction and the difference in magnitude is less than a preset threshold of 0.02, they are marked as consistent. When the two values ​​change in different directions or the difference in magnitude exceeds a preset threshold of 0.05, they are marked as different, and a sequence of change difference relationship identifiers is formed according to the order of the data points. For example, if the difference status identifiers appear consecutively between sampling points 18620 and 18695, it indicates that there is a significant asynchrony between the offset change and the correlation change within this interval.

[0066] After constructing the sequence of identification of changes in difference relationships, interval division processing is performed on the consistent state interval and the difference state interval, and offset decoupling parameters and consistent association parameters are constructed respectively. Specifically, for continuous difference state intervals, the residual offset change information at the corresponding position is extracted from the joint calculation sequence and arranged continuously according to the data point order to form the offset decoupling parameter. For example, in the interval from sampling point 18620 to 18695, the corresponding residual offset change amount is generally maintained between 0.120 and 0.210, while the evolutionary association change amount is always lower than 0.030. Therefore, the residual offset change information in this interval is extracted to form the offset decoupling parameter, which is used to characterize the independent offset feature that is not affected by the association change. At the same time, for the consistent state interval, the residual offset change information at the corresponding position and the evolutionary association change information are synchronously combined and calculated. In this embodiment, the joint result of the two types of change values ​​is calculated by normalized proportional fusion method, thereby forming the consistent association parameter. For example, within a certain continuous and consistent state interval, the residual offset change and the evolutionary correlation change change synchronously. The combined consistent correlation parameter remains stable between 0.420 and 0.510, indicating that the offset change and correlation change maintain a high degree of synchronicity within this interval. This method achieves the separate expression of independent offset features and correlation co-occurrence features, thereby avoiding the offset misjudgment problem caused by the coupling of different change information.

[0067] After obtaining the offset decoupling parameter and the consistency correlation parameter, a combined metric calculation is performed on the two types of parameters to determine the degree of residual offset corresponding to each data point. Specifically, firstly, a hierarchical mapping process is performed based on the numerical ratio between the offset decoupling parameter and the consistency correlation parameter. When the offset decoupling parameter accounts for more than 70%, it is determined to be an offset-dominant state; when the consistency correlation parameter accounts for more than 70%, it is determined to be a correlation-dominant state; and the rest are determined to be in an equilibrium state. Subsequently, a weighted superposition calculation is performed based on the degree of consistency of the changes of the two types of parameters in continuous data points. For example, at a certain data point, the offset decoupling parameter is 0.812 and the consistency association parameter is 0.264, corresponding to an offset proportion of 75.4%. Therefore, this is determined to be an offset-dominated state, and the weight of the offset decoupling parameter is increased to 0.73. After weighted summation, the corresponding residual offset degree is 0.846. At another data point, the offset decoupling parameter and the consistency association parameter are 0.431 and 0.472 respectively, with their proportions being relatively close. Therefore, this is determined to be a balanced state, and the final residual offset degree is 0.458. Finally, the residual offset degrees corresponding to all data points are serialized and arranged according to the original sampling order to form a continuous residual offset degree sequence for subsequent consistency determination and data fusion alignment processing.

[0068] like Figure 4The diagram shown illustrates the temporal distribution of the residual offset characterization parameters according to the present invention. The residual offset characterization parameters remain at a low level overall within the normal data range, while consistently exceeding a preset threshold within the abnormal offset range, forming a continuous high-amplitude fluctuation range. This allows for the characterization of asynchronous offset features in multi-source industrial data.

[0069] Based on the data from the lithium battery electrode coating production line described above, a normal operating condition dataset and an abnormal offset dataset were constructed. The abnormal data was generated by artificially introducing a 3-6 second sampling time drift and an 8% local misalignment. The traditional fixed-threshold anomaly detection method and the method described in this embodiment were then compared and tested. During the experiment, 48 hours of continuous production data were selected as the test sample, containing approximately 174,000 sets of valid sampled data. Experimental results show that the traditional fixed-threshold method has an accuracy rate of 79.3% for identifying continuous offset intervals, but a false negative rate of 14.8% for local gradual offsets. However, using the method in this embodiment, the accuracy rate for identifying continuous offset intervals increased to 95.7%, the false negative rate for local gradual offsets decreased to 3.6%, and the average positioning error for offset intervals decreased from 16.2 sampling points in the traditional method to 4.1 sampling points. Furthermore, when performing retrospective analysis on actual abnormal shutdown events of the production line, this embodiment can identify abnormal offset trends between multi-source data approximately 11 minutes in advance. This demonstrates that this embodiment can not only effectively characterize the degree of residual offset, but also improve the ability to identify abnormal changes in industrial data in advance.

[0070] S4. Couple the residual offset degree with the consistency correlation parameter to construct the consistency evaluation parameter. Based on the consistency evaluation parameter, complete the consistency judgment in the industrial data quality consistency test, and perform the alignment processing in data fusion in combination with the residual offset degree. In this embodiment, S4 specifically refers to: The residual offset degree and the consistency correlation parameter are paired according to the data point order, and a coupling calculation is performed on each pair of data. After the residual offset degree and the consistency correlation parameter are processed by a unified scale, a numerical ratio division is performed, and the ratio division result is mapped to the corresponding level weight. At the same time, the weight is adjusted according to the stability of the change of the consistency correlation parameter in continuous data points, and the adjusted weight is weighted and calculated with the residual offset degree to construct the consistency evaluation parameter and arrange it according to the data point order. When processing the residual offset and the consistency correlation parameter, we can first match the two types of data point by point according to the data point order, so that each data point has both offset and correlation information. Then, we perform coupling calculations on each pair of matched data. During the calculation, we first unify the scale of the two types of values, for example, by standardizing or compressing the numerical range, so that the two types of data are within the same numerical range. Then, we divide the numerical values ​​according to their relative size relationship and map the division results to different levels of weight. For example, when the residual offset is significantly greater than the consistency correlation parameter, a higher weight is assigned. High offset weights are assigned, and balanced weights are given when the two are close. At the same time, the stability of the consistency correlation parameter is judged by the changes in the continuous data points. For example, if the changes in the continuous data points are small, they are considered stable, and the corresponding weights are maintained or enhanced. If the changes are large, the weights are reduced. After the weight adjustment is completed, the adjusted weights are applied to the residual offset degree for weighted calculation to form the consistency evaluation parameter for each data point. For example, when a certain data point has a high offset degree and the correlation change is unstable, its weight adjustment will make the evaluation result biased towards the non-consistent state. This can achieve a comprehensive quantitative expression of offset and correlation information.

[0071] The residual offset degree and the consistency association parameter represent numerical descriptions of the data in terms of offset degree and association consistency, respectively. They are paired to form a one-to-one matching relationship. Coupled computation refers to the process of jointly processing two types of data under a unified framework. Unified scaling is the process of transforming data of different dimensions or ranges into the same numerical interval for comparison. Numerical proportion division is the process of classifying data according to the relative size relationship between the two types of data. The proportion division result reflects the classification result of the relative proportion of the two types of data. Corresponding level weights are weight values ​​assigned according to the proportion division result, used to control the degree of influence of different data in the calculation. The stability of the consistency association parameter in continuous data points represents its fluctuation in the time dimension, used to adjust the weight allocation. Weighted computation is the process of applying weights to the data to form a comprehensive result. The consistency evaluation parameter is a comprehensive value obtained through coupled computation, used to describe the consistency status of each data point.

[0072] Based on the consistency assessment parameters, consistency determination calculations are performed on each data point. The consistency assessment parameters are divided into intervals, and the interval to which the consistency assessment parameters belong is determined for each data point. The corresponding interval is marked as a consistent state or a non-consistent state. At the same time, the continuity of the consistency state distribution in continuous data points is verified, and the consistency determination in the industrial data quality consistency test is completed. When processing consistency assessment parameters, consistency judgment calculations can be performed on the assessment values ​​of each data point. First, the overall numerical distribution of the consistency assessment parameters is divided into intervals, for example, the assessment values ​​are divided into multiple continuous intervals according to their magnitude, with each interval corresponding to a different consistency level. Then, for each data point, the interval range into which its consistency assessment parameter falls is determined, and the data point is marked as consistent or inconsistent according to the interval category. For example, when the assessment value falls into a high consistency interval, it is marked as consistent, and when it falls into a low consistency interval, it is marked as inconsistent. After completing the single-point judgment, the continuity of the state distribution in continuous data points is checked. By scanning the state changes of adjacent data points, short-term abrupt state changes are smoothed. For example, when a data point is judged as inconsistent but the data points before and after it are consistent, consistency correction can be performed on that point to avoid the impact of isolated fluctuations on the overall judgment. This can ensure the stability and continuity of the consistency judgment results in the time dimension.

[0073] Consistency determination calculation refers to the calculation process of classifying and judging the consistency of data points based on consistency evaluation parameters. Interval grading is the process of dividing a continuous numerical range into multiple level intervals to define different consistency level standards. The interval to which the consistency evaluation parameter belongs represents the interval range in which the evaluation value of a certain data point is located, and is the basis for state determination. Consistent state or non-consistent state is the classification label of data points based on the interval division results. Continuity verification is the process of verifying the consistency of the determination results of adjacent data points. By analyzing the continuity of the state distribution, abnormal changes are corrected to ensure the stability and reliability of the determination results.

[0074] By combining the consistency judgment result and the residual offset degree, alignment processing calculation is performed on each data point. The position of the data point in the feature space is offset correction according to the residual offset degree, and the correction magnitude is controlled in stages according to the consistency judgment result. The corrected data is arranged according to the order of the data points to achieve alignment processing in data fusion.

[0075] When performing alignment processing on each data point, the consistency judgment result and the residual offset degree can be used together to adjust the position of each data point in the feature space. Specifically, the offset direction and offset amount of each data point can be determined first based on the residual offset degree. For example, when the residual offset degree of a data point is large, its coordinates in the feature space are moved in the opposite direction to the reference position, while when the offset degree is small, only a fine adjustment is made. Then, the correction magnitude is controlled in stages based on the consistency judgment result. For example, for data points judged to be in a consistent state, only a small range of position adjustment is allowed, while for data points judged to be in a non-consistent state, a larger range of offset correction is allowed, thereby achieving differentiated processing. For example, in a certain data segment, if some data points have a high offset degree and are judged to be in a non-consistent state, their coordinates are adjusted by a large amount to move them closer to the overall distribution center, while data points that are already in a consistent state maintain their original positions or only make slight adjustments. Finally, all the corrected data points are arranged in the original order to form an aligned data sequence, thereby achieving unified alignment of multi-source data in the feature space.

[0076] The consistency judgment result and residual offset degree represent the consistency classification status and offset degree quantification result of the data points, respectively, and both serve as the control basis for alignment processing. Alignment processing calculation refers to the calculation process of adjusting the position of data points according to the offset degree and consistency status. Feature space is a multi-dimensional spatial structure used to represent the positional relationship of data points. The position of a data point in the feature space represents its value state on the multi-dimensional features. Offset correction refers to the operation of adjusting the position of data points according to the offset degree. Correction magnitude represents the degree of movement of data points during the adjustment process. Hierarchical control refers to the hierarchical management of the correction magnitude according to the consistency judgment result, so as to realize the differentiated adjustment strategy for different data points.

[0077] After constructing the residual offset degree and consistency correlation parameters, the two types of parameters are further coupled and calculated to form consistency evaluation parameters for industrial data quality consistency testing. Specifically, firstly, the residual offset degree and consistency correlation parameters are paired point by point according to the time order of data points in the original sequence, and the two types of data are uniformly scaled using a min-max normalization method to uniformly map data of different dimensions to the [0,1] interval; then, a level division is performed according to the numerical ratio of the two types of data. When the ratio of the residual offset degree to the consistency correlation parameter exceeds 1.5, it is determined to be an offset-dominant state; when the ratio is less than 0.7, it is determined to be a correlation-dominant state; and the rest are determined to be an equilibrium state, and are assigned basic weights of 0.75, 0.35, and 0.55 respectively. For example, at the 21430th sampling point, the residual offset is 0.812 and the consistency correlation parameter is 0.428. After normalization, the ratio of the two reaches 1.89, thus indicating a offset-dominated state, and a base weight of 0.75 is assigned. Simultaneously, the weight is dynamically adjusted based on the stability of the consistency correlation parameter across consecutive data points. When the fluctuation of the consistency correlation parameter within five consecutive sampling points is less than 0.03, the corresponding weight is increased; when the fluctuation exceeds 0.08, the corresponding weight is decreased. In this embodiment, the fluctuation of the consistency correlation parameter between sampling points 21430 and 21435 reaches 0.11, so the corresponding weight for this interval is adjusted from 0.75 to 0.82. Subsequently, the adjusted weight is used to perform a weighted calculation on the residual offset, ultimately obtaining a consistency evaluation parameter of 0.846 for the corresponding data point, thereby achieving a comprehensive quantitative expression of the offset and correlation stability.

[0078] After the consistency assessment parameters are generated, consistency determination calculations are performed on each data point to complete the consistency analysis in the industrial data quality consistency test. Specifically, based on the overall distribution of the consistency assessment parameters, the assessment results are divided into a high consistency interval, a middle transition interval, and a low consistency interval. A consistency assessment parameter greater than 0.75 is considered consistent, less than 0.45 is considered inconsistent, and between 0.45 and 0.75 is considered pending confirmation. Then, for each data point, the interval to which its consistency assessment parameter belongs is determined, and a corresponding consistency status identifier is assigned. For example, at the 21430th sampling point, the consistency assessment parameter reaches 0.846, thus it is considered consistent; however, at the 21705th sampling point, the consistency assessment parameter is only 0.382, thus it is considered inconsistent. After completing the single-point judgment, a continuity check is further performed on the state distribution of continuous data points. When an isolated data point is judged to be in an inconsistent state, but its eight consecutive sampling points before and after it are in an consistent state, the isolated point is corrected to an consistent state to reduce the impact of short-term random fluctuations on the overall consistency judgment result. When the length of the continuous inconsistency state interval exceeds 15 sampling points, its inconsistency state identifier is retained, thereby ensuring the stability and reliability of the industrial data quality consistency test results in the time dimension. In addition, during the actual test, when verifying continuous 48-hour production data, this embodiment achieved an accuracy rate of 96.4% in identifying abnormal consistency states of multi-source industrial data, which is a significant improvement compared to the 82.7% accuracy rate of traditional detection methods based on fixed thresholds.

[0079] After completing the consistency determination, alignment processing in data fusion is performed on each data point based on the consistency determination result and the degree of residual offset. Specifically, the offset direction and amount of each data point in the feature space are determined according to the degree of residual offset. When the residual offset of a data point exceeds 0.70, its feature space coordinates are corrected in the opposite direction towards the overall distribution center, while when the offset is less than 0.30, only fine-tuning is performed. At the same time, the correction magnitude is controlled in stages based on the consistency determination result. For data points in a consistent state, the correction magnitude is limited to within 15% of the original offset, while for data points in a non-consistent state, a position correction magnitude of up to 60% is allowed. For example, in a continuous abnormal interval, the data point corresponding to the vibration monitoring data has an offset of 0.884 and is determined to be in a non-consistent state. Therefore, a feature space position correction of 0.53 units is performed to move it closer to the overall data distribution center; while for data points in a normal state, only a fine-tuning of 0.08 units is performed. During the experiment, verification was conducted based on 48 hours of continuous lithium battery production line data, including approximately 174,000 sets of valid sampled data. The experimental results showed that when using the traditional fixed window alignment method, the Pearson consistency coefficient after multi-source industrial data fusion was 0.81, and the average time offset error after data alignment was 2.8s. However, after using the method in this embodiment, the Pearson consistency coefficient increased to 0.95, the average time offset error decreased to 0.6s, and the accuracy of identifying continuous abnormal intervals increased from 83.7% to 96.4%. This indicates that this embodiment can effectively realize consistency judgment in industrial data quality consistency testing and alignment processing in the process of multi-source industrial data fusion.

[0080] S5. Based on the data fusion results, feedback correction parameters are generated, and iterative updates are performed on the residual offset characterization parameters, consistency correlation parameters, and residual offset degree.

[0081] In this embodiment, S5 specifically refers to: Based on the data fusion results, the position difference calculation is performed on the corresponding data points in the feature space of the data before and after fusion. The difference results of each data point are decomposed and processed to extract the difference amount representing the offset change and the difference amount representing the correlation change. The offset change difference amount and the correlation change difference amount are combined and mapped to form feedback correction parameters that correspond one-to-one with the data points and are arranged in the order of the data points. When processing the data fusion results, we can first perform point-by-point matching between the data before and after fusion in the feature space, so that each data point has a comparable positional representation before and after fusion. Then, we perform difference calculation on the coordinate positions of the same data point before and after fusion. For example, by comparing the positional changes in each feature dimension before and after fusion, we can obtain the overall positional offset of the data point. Subsequently, we decompose the difference result, dividing the positional change into offset change difference and correlation change difference. The offset change difference is used to characterize the degree of deviation of the data point's own position, while the correlation change difference is used to characterize the change in the relationship structure between the data point and the surrounding data points. For example, when a data point moves towards the overall center after fusion but its relationship with its neighboring data points remains stable, the offset change difference is large and the correlation change difference is small. Then, we perform combined mapping processing on the two types of difference, for example, by jointly encoding according to the relative size of the two types of difference, so that each data point forms a comprehensive correction value. Finally, we arrange the data points in order to form feedback correction parameters, thereby providing a basis for subsequent parameter updates.

[0082] The data fusion result represents the unified data representation formed after alignment processing; the corresponding data points in the feature space between the data before and after fusion refer to the mapping relationship between the data before and after fusion at the same position or index; the position difference calculation is the process of quantifying the position changes of data points in the feature space before and after fusion; the difference result decomposition processing is the process of breaking down the overall difference into different types of change information; the offset change difference quantity represents the degree of change in the position of the data point itself; the association change difference quantity represents the degree of change in the relationship structure between the data point and other data; the combined mapping calculation is the process of fusing the two types of difference quantities according to a certain mapping rule; the feedback correction parameter is a set of data formed by the combined mapping result to guide the subsequent parameter adjustment. This parameter corresponds one-to-one with the data points and is arranged in order to describe the impact of changes before and after fusion on the model parameters.

[0083] Based on the feedback correction parameters, iterative update calculations are performed on the residual offset characterization parameters, the consistency correlation parameters, and the residual offset degree. By mapping the feedback correction parameters to the residual offset characterization parameters and the consistency correlation parameters respectively, numerical adjustment processing is performed on the two types of parameters. Based on the updated residual offset characterization parameters and the consistency correlation parameters, a joint correction calculation is performed on the residual offset degree, resulting in updated residual offset characterization parameters, consistency correlation parameters, and residual offset degree, which are arranged in the order of data points.

[0084] When applying feedback correction parameters, they can be used as adjustment criteria to apply to the residual offset characterization parameter and the consistency correlation parameter respectively. Specifically, a one-to-one correspondence between the feedback correction parameter and the two types of parameters can be established according to the data point order. Then, for each data point, the residual offset characterization parameter is incrementally adjusted according to the magnitude and direction of the feedback correction parameter. For example, when the feedback correction parameter reflects the convergence of the fused data position towards the center, the value of the residual offset characterization parameter is reduced accordingly, while when it reflects the expansion of the offset, its value is increased. At the same time, a similar adjustment is performed on the consistency correlation parameter to reflect the changes in the correlation in the fusion result. After completing the update of the two types of parameters, the joint calculation is re-executed based on the updated residual offset characterization parameter and the consistency correlation parameter to correct the original residual offset degree. For example, when the updated offset characterization parameter decreases and the consistency correlation parameter increases, the residual offset degree is reduced accordingly, so that the overall offset description is more consistent with the fused data state. Finally, the updated results are arranged according to the data point order to form a new parameter sequence for subsequent iterative calculations or continuous optimization.

[0085] Feedback correction parameters represent the set of values ​​extracted from the differences before and after fusion to adjust the internal parameters of the model; residual offset parameters represent a quantitative description of the data offset characteristics; consistent correlation parameters represent the degree of consistency of the correlation between data, and both are adjusted synchronously through feedback correction parameters; iterative update calculation refers to the calculation process of gradually approaching the target state through multiple corrections based on existing parameters; numerical adjustment processing is the process of increasing or decreasing the original values ​​according to the feedback correction parameters; joint correction calculation is the process of recalculating the degree of offset using the updated two types of parameters; residual offset degree represents the quantitative result of the comprehensive offset of data points; arranging according to the order of data points means maintaining the same sequential structure of the updated parameters as the original data to ensure the continuity and consistency of subsequent processing.

[0086] After completing the consistency determination and data fusion alignment processing, feedback correction parameters are further formed based on the data fusion results, and iterative updates are performed on the residual offset characterization parameters, consistency correlation parameters, and residual offset degree. Specifically, the data before and after fusion are first matched point-by-point in a unified feature space. In this embodiment, the unified feature space adopts a four-dimensional feature vector space composed of temperature, pressure, vibration amplitude, and time-series change rate, and Euclidean distance is used as the position difference measure of data points before and after fusion. The overall position offset value is obtained by calculating the coordinate changes in each feature dimension before and after fusion. For example, at the 23640th sampling point, the feature coordinates before fusion are (0.82, 0.47, 0.31, 0.28), and the feature coordinates after fusion are (0.63, 0.52, 0.36, 0.25), resulting in a positional difference of 0.203. This difference is then decomposed, defining the change in the data point's own position as the offset difference and the change in the distance relationship between the data point and its five neighboring data points as the correlation difference. In this embodiment, the correlation difference is obtained by performing difference calculations on the neighborhood distance matrices before and after fusion. For example, the offset difference for the aforementioned data point... The displacement difference is 0.171, and the correlation difference is 0.032, indicating that after fusion, the data point mainly shows that its own position converges towards the overall distribution center, while the neighborhood correlation structure changes little. Subsequently, a combined mapping calculation is performed on the two types of differences. In this embodiment, a weighted proportional mapping method is used to generate feedback correction parameters, where the weight of the offset difference is set to 0.7 and the weight of the correlation difference is set to 0.3. Finally, the feedback correction parameter of the corresponding data point is 0.129, and a continuous feedback correction parameter sequence is formed according to the order of the data points, thereby providing clear data input and calculation basis for subsequent parameter iteration updates.

[0087] After forming the feedback correction parameters, they are mapped to the residual offset characterization parameters and the consistency association parameters, respectively. The residual offset degree is then recalculated based on the updated parameters to complete the iterative update process. Specifically, a one-to-one correspondence is first established between the feedback correction parameters, residual offset characterization parameters, and consistency association parameters according to the data point order. When the feedback correction parameters indicate that the fused data converges towards the overall distribution center, the corresponding residual offset characterization parameter value is decreased, and the consistency association parameter value is increased. Conversely, when the feedback correction parameters indicate that the offset expands after fusion, the residual offset characterization parameter is increased, and the consistency association parameter is decreased. In this embodiment, a linear increment / decrement update method is used to perform numerical adjustment, with update formulas Pi' = Pi'. λFi and Ci' = Ci + μFi, where Pi represents the residual offset characterization parameter before the update, Ci represents the consistency correlation parameter before the update, and Fi represents the feedback correction parameter, with λ and μ set to 0.65 and 0.35 respectively. For example, at the 23640th sampling point, the residual offset characterization parameter before the update is 0.784, the consistency correlation parameter is 0.426, and the corresponding feedback correction parameter is 0.129. After the update, the residual offset characterization parameter is adjusted to 0.700, and the consistency correlation parameter is adjusted to 0.471. Subsequently, based on the updated two types of parameters, a joint correction calculation is re-performed, and the degree of residual offset after the update is reduced from 0.812 to 0.594, indicating that the data offset state after fusion is significantly improved. During the experiment, iterative verification was performed based on 72 hours of continuous lithium battery production line data, comprising approximately 268,000 sets of valid sampled data. After three rounds of feedback correction iterations, the Pearson consistency coefficient of the multi-source industrial data fusion result increased from the initial 0.81 to 0.96, the average time offset error decreased from 2.6s to 0.4s, the proportion of abnormal offset data points decreased from 12.3% to 3.1%, and the mean square error after data fusion decreased from 0.137 to 0.042. This demonstrates that this embodiment can achieve dynamic optimization and updating of residual offset characterization parameters, consistency correlation parameters, and residual offset degree through feedback correction parameters. Figure 3 The diagram illustrates the changes in key indicators of the multi-source industrial data fusion results of this invention. After feedback correction iterations, the Pearson consistency coefficient among the multi-source industrial data gradually increases, the average time offset error continues to decrease, and the proportion of abnormal offset data points significantly decreases, indicating that this invention can effectively improve the consistency and alignment accuracy in the industrial data fusion process.

[0088] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0089] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0090] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0091] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0092] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0093] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0094] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data fusion method for artificial intelligence-based industrial data quality consistency testing, characterized in that, Specifically, the following steps are included: S1. Perform artificial intelligence reconstruction processing on the multi-source industrial data and obtain residual sequences. Construct a residual multi-domain mapping sequence containing spatial positioning parameters and evolutionary correlation parameters based on the residual sequences. The multi-source industrial data refers to the multi-source industrial data in the coating production line of lithium battery electrode sheets for new energy vehicles. The collected industrial data includes coating temperature sensor data, roller pressure sensor data, and equipment vibration monitoring data. The residual sequence is a sequence formed by calculating the difference between the multi-source industrial data and the corresponding reconstructed data sequence point by point. S1 specifically refers to: The multi-source industrial data is reconstructed using artificial intelligence. The multi-source industrial data is input into the corresponding artificial intelligence reconstruction model for mapping calculation, generating a reconstructed data sequence corresponding to the multi-source industrial data. The difference between the multi-source industrial data and the reconstructed data sequence is calculated for each data point to obtain the residual sequence. For the residual sequence, position mapping calculation is performed in a unified feature space to extract the distribution coordinates of each data point in the residual sequence. Then, cluster center calculation and relative position offset calculation are performed on the distribution coordinates to construct spatial positioning parameters based on the distribution center position and offset vector. After completing the construction of spatial positioning parameters, differential calculation and trend extraction are performed on adjacent data points of the residual sequence in continuous time series to form a time series change sequence. Correlation calculation is then performed on the time series change sequence to construct evolutionary correlation parameters. At the same time, the spatial positioning parameters and evolutionary correlation parameters are combined and mapped at the sequence level according to the correspondence of data points to form a residual multi-domain mapping sequence. S2. Perform cross-comparison operations on the spatial positioning parameters and evolutionary correlation parameters in the residual multi-domain mapping sequence to determine whether there is asynchronous shift in the residual in the feature space after the multi-source industrial data is reconstructed by artificial intelligence. If there is asynchronous shift, perform weighted combination calculation on the spatial distribution difference and temporal evolution difference in the cross-comparison result sequence, and construct residual shift characterization parameters by combining the consistency results of change direction. Serialize and output the residual shift characterization parameters according to the order of data points. S3. Joint calculations are performed using residual offset characterization parameters and evolutionary correlation parameters to construct offset decoupling parameters and consistency correlation parameters. The degree of residual offset is determined based on the joint calculation results. The offset decoupling parameters characterize the independent change characteristics of residual offset relative to evolutionary correlation changes, while the consistency correlation parameters characterize the consistent relationship between residual offset changes and evolutionary correlation changes. Based on the offset decoupling parameters and consistency correlation parameters, a combined metric calculation is performed on each data point. A hierarchical mapping process is applied to the numerical ratio between the offset decoupling parameters and the consistency correlation parameters, and a weighted superposition calculation is performed combining the degree of consistency of their changes in continuous data points. The combined result is used as the degree of residual offset for the corresponding data point, and the data is serialized and output according to the data point order. S4. Couple the residual offset degree with the consistency correlation parameter to construct the consistency evaluation parameter. Based on the consistency evaluation parameter, complete the consistency judgment in the industrial data quality consistency test, and perform the alignment processing in data fusion in combination with the residual offset degree. S4 specifically refers to: The residual offset degree and the consistency correlation parameter are paired according to the data point order, and a coupling calculation is performed on each pair of data. After the residual offset degree and the consistency correlation parameter are processed by a unified scale, a numerical ratio division is performed, and the ratio division result is mapped to the corresponding level weight. At the same time, the weight is adjusted according to the stability of the change of the consistency correlation parameter in continuous data points, and the adjusted weight is weighted and calculated with the residual offset degree to construct the consistency evaluation parameter and arrange it according to the data point order. Based on the consistency assessment parameters, consistency determination calculations are performed on each data point. The consistency assessment parameters are divided into intervals, and the interval to which the consistency assessment parameters belong is determined for each data point. The corresponding interval is marked as a consistent state or a non-consistent state. At the same time, the continuity of the consistency state distribution in continuous data points is verified, and the consistency determination in the industrial data quality consistency test is completed. Combining the consistency judgment result and the residual offset degree, alignment processing calculation is performed on each data point. The position of the data point in the feature space is offset by the residual offset degree, and the correction magnitude is controlled in stages according to the consistency judgment result. The corrected data is arranged according to the order of the data points to achieve alignment processing in data fusion. S5. Based on the data fusion results, feedback correction parameters are generated, and iterative updates are performed on the residual offset characterization parameters, consistency correlation parameters, and residual offset degree. S5 specifically refers to: Based on the data fusion results, the position difference calculation is performed on the corresponding data points in the feature space of the data before and after fusion. The difference results of each data point are decomposed and processed to extract the difference amount representing the offset change and the difference amount representing the correlation change. The offset change difference amount and the correlation change difference amount are combined and mapped to form feedback correction parameters that correspond one-to-one with the data points and are arranged in the order of the data points. Based on the feedback correction parameters, iterative update calculations are performed on the residual offset characterization parameters, the consistency correlation parameters, and the residual offset degree. By mapping the feedback correction parameters to the residual offset characterization parameters and the consistency correlation parameters respectively, numerical adjustment processing is performed on the two types of parameters. Based on the updated residual offset characterization parameters and the consistency correlation parameters, a joint correction calculation is performed on the residual offset degree, resulting in updated residual offset characterization parameters, consistency correlation parameters, and residual offset degree, which are arranged in the order of data points.

2. The data fusion method for artificial intelligence-based industrial data quality consistency testing according to claim 1, characterized in that, S2 specifically includes the following steps: S201. The spatial positioning parameters and evolutionary correlation parameters in the residual multi-domain mapping sequence are synchronously paired according to the correspondence of data points, and numerical difference calculation is performed for each paired data to obtain the numerical difference between the spatial positioning parameters and the evolutionary correlation parameters. At the same time, the direction of change of the spatial positioning parameters corresponding to adjacent data points and the direction of change of the evolutionary correlation parameters are determined to be consistent. Based on the numerical difference and direction consistency results, a cross-comparison result sequence containing spatial distribution difference and temporal evolution difference is constructed. S202. Based on the cross-comparison result sequence, perform joint discrimination calculation on the spatial distribution difference and the temporal evolution difference. By judging the consistency of the change trend and change direction of the difference in continuous data points, determine whether there is asynchronous shift in the feature space of the residual after the reconstruction of multi-source industrial data by artificial intelligence. S203. In the case of asynchronous offset, the spatial distribution difference and temporal evolution difference in the cross-comparison result sequence are weighted and combined, and the residual offset characterization parameter is constructed by combining the consistency result of the change direction. The residual offset characterization parameter is serialized and output according to the data point order. 3.The data fusion method of artificial intelligence-based industrial data quality consistency testing according to claim 2, characterized in that, S202 specifically refers to: Based on the cross-comparison result sequence, the spatial distribution difference and the temporal evolution difference are arranged continuously according to the data point order. The spatial distribution difference corresponding to adjacent data points is differentially calculated to form a spatial difference change sequence, and the temporal evolution difference corresponding to adjacent data points is differentially calculated to form an evolution difference change sequence. For spatial difference change sequences and evolutionary difference change sequences, the change trend corresponding to each data point is marked with a direction. For the same data point, the spatial difference change direction and the evolutionary difference change direction are compared and calculated to determine the consistency of the signs. When the signs of the two directions are the same, they are marked as consistent. When the signs of the two directions are opposite, they are marked as inconsistent. The consistent and inconsistent states are then sequentially arranged according to the order of the data points to form a sequence of the degree of consistency of the change direction. Based on the sequence of consistency of change direction, interval statistical processing is performed on the distribution of consistent and inconsistent states in continuous data points. Combined with the continuity of the change trend of spatial difference change sequence and evolutionary difference change sequence, joint discrimination calculation is performed. When there are continuous inconsistent state intervals in the sequence of consistency of change direction and the corresponding change trend does not have synchronous change characteristics, it is determined that the residual of multi-source industrial data after artificial intelligence reconstruction has asynchronous offset in the feature space.

4. The data fusion method for artificial intelligence-based industrial data quality consistency testing according to claim 2, characterized in that, S203 specifically refers to: In cases where asynchronous offset is determined, spatial distribution difference and temporal evolution difference are extracted from the cross-comparison result sequence. The spatial distribution difference and temporal evolution difference are then paired and arranged according to the data point order. At the same time, the direction consistency result is combined to identify the corresponding direction consistency state for each pair of data. For the spatial distribution difference and temporal evolution difference after pairing and arrangement, a weighted combination calculation is performed on each data point. The corresponding weights are determined according to the numerical ratio between the spatial distribution difference and the temporal evolution difference, and the weights are adjusted in combination with the consistency results of the change direction to form a combined difference sequence containing the weighted spatial distribution difference and the weighted temporal evolution difference. Based on the combined difference sequence, the weighted spatial distribution difference and weighted temporal evolution difference corresponding to each data point are fused and calculated to construct residual offset characterization parameters, and the residual offset characterization parameters are serialized and output according to the order of data points.

5. The data fusion method for industrial data quality consistency testing based on artificial intelligence according to claim 1, characterized in that, S3 specifically includes the following steps: S301. The residual migration characterization parameters and evolution correlation parameters are paired according to the data point order, and the difference calculation of adjacent data points is performed for each paired data point. The change of the residual migration characterization parameters in continuous data points is extracted as the residual migration change information. At the same time, the change of the evolution correlation parameters in continuous data points is extracted as the evolution correlation change information. A joint calculation sequence is formed based on the two types of changes. S302. For the joint calculation sequence, perform separation mapping calculation on the residual offset change information and the evolution association change information. By comparing the difference and synchronization relationship of the two types of change information in adjacent data points, construct the offset decoupling parameter to characterize the independent change characteristics of the residual offset and the consistency association parameter to characterize the consistent relationship between the two types of change, and arrange them accordingly according to the order of data points. S303. Based on the offset decoupling parameter and the consistency association parameter, perform a combined metric calculation on each data point. By performing hierarchical mapping processing on the numerical ratio relationship between the offset decoupling parameter and the consistency association parameter, and combining the degree of consistency of their changes in continuous data points, perform a weighted superposition calculation. The combined result is used as the residual offset degree of the corresponding data point, and is serialized and output according to the order of the data points.

6. The data fusion method for industrial data quality consistency testing based on artificial intelligence according to claim 5, characterized in that, S302 specifically refers to: For the residual offset change information and evolution association change information in the joint calculation sequence, corresponding pairing processing is performed according to the data point order, and the difference comparison calculation is performed on the residual offset change information of adjacent data points to form an offset difference sequence, and the difference comparison calculation is performed on the evolution association change information of adjacent data points to form an association difference sequence. Based on the offset difference sequence and the associated difference sequence, the synchronization relationship is determined for the residual offset change information and the evolutionary associated change information corresponding to each data point. By performing point-by-point comparison calculation on the change direction and change magnitude of the offset difference sequence and the associated difference sequence at the same data point position, a change difference relationship identification sequence is formed, and interval division is performed according to the data point order. For the change difference relationship identifier sequence, a separation mapping calculation is performed on the residual offset change information and the evolution association change information. The residual offset change information in the data point interval corresponding to the non-consistent state in the change difference relationship identifier sequence is extracted to form an offset decoupling parameter. At the same time, the residual offset change information in the data point interval corresponding to the consistent state in the change difference relationship identifier sequence is combined with the evolution association change information to form a consistent association parameter, and then arranged in the corresponding order according to the data point order.

Images