Industrial data processing method and system based on artificial intelligence

Abstract

The invention relates to the field of data processing, in particular to an industrial data processing method and system based on artificial intelligence. The method comprises the following steps: acquiring a historical industrial log; performing mean filtering processing on the historical industrial log, and performing cross-dimension heterogeneous feature fusion to generate heterogeneous fusion industrial feature data; obtaining time sequence trend change data based on the heterogeneous fusion industrial feature data; performing deep time sequence dependency mining on the time sequence trend change data to obtain a step point time sequence dependency relationship; performing mapping topology position reconstruction on the heterogeneous fusion industrial feature data based on a step point time sequence dependency relationship, and constructing an industrial feature mapping space model; performing multi-dimensional anomaly detection on the industrial feature mapping space model, and performing local anomaly distribution mining to obtain a high-frequency anomaly region; and performing multi-point abnormal evolution tracking on the high-frequency abnormal region so as to construct an abnormal evolution trajectory visual view. According to the invention, efficient and accurate industrial data processing is realized.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to an industrial data processing method and system based on artificial intelligence. Background Art

[0002] In modern industrial production, data collection and analysis capabilities are crucial. With the continuous improvement of industrial automation levels, various sensors and data acquisition equipment are widely used in the production process, generating massive amounts of data. However, traditional data processing methods often have difficulty coping with these large and complex data, and it is difficult to effectively extract key information. There is often a low data processing efficiency and an inability to extract accurate information. Therefore, in order to meet the needs of modern industrial data processing, a more intelligent industrial data processing method is needed. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes an industrial data processing method and system based on artificial intelligence to solve at least one of the above technical problems.

[0004] To achieve the above object, the present invention provides an industrial data processing method based on artificial intelligence, comprising the following steps: Step S1: Obtain historical industrial logs; perform mean filtering on the historical industrial logs, and perform cross-dimensional heterogeneous feature fusion to generate heterogeneous fused industrial feature data; Step S2: obtaining time series trend change data based on heterogeneous fusion industrial feature data; performing deep time series dependency mining on the time series trend change data to obtain step point time series dependency relationships; Step S3: reconstruct the mapping topology position of the heterogeneous fused industrial feature data based on the step point temporal dependency relationship, and construct an industrial feature mapping space model; Step S4: Perform multi-dimensional anomaly detection on the industrial feature mapping spatial model and perform local anomaly distribution mining to obtain high-frequency anomaly areas; Step S5: Track the evolution of multi-point anomalies in high-frequency anomaly areas and visualize their multi-dimensional distribution, thereby constructing a visualization view of the anomaly evolution trajectory; Step S6: Perform anomaly attribution quantitative analysis on the visualization view of the anomaly evolution trajectory to obtain attribution data of different anomaly evolution trajectories; make adaptive anomaly optimization decisions based on the attribution data of different anomaly evolution trajectories, and construct an anomaly trajectory optimization strategy.

[0005] The present invention acquires historical industrial logs to establish a comprehensive understanding of industrial processes, provides important data support for subsequent analysis, and uses mean filtering and cross-dimensional heterogeneous feature fusion to smooth data and reduce noise, thereby generating more reliable heterogeneous fused industrial feature data. Time series trend change data is obtained based on heterogeneous fused industrial feature data to reveal the changing trend of industrial data and provide information support in the time dimension for subsequent analysis. Deep time series dependency mining helps discover the complex relationships between data, obtains step point time series dependency relationships, and provides more in-depth data mining results. By mapping the topological position of heterogeneous fused industrial feature data and reconstructing it, an industrial feature mapping space model is constructed to visualize and integrate data relationships, build an industrial feature mapping space model to understand the relationship between data, improve the efficiency and accuracy of data analysis, and Mapping spatial models for multi-dimensional anomaly detection and local anomaly distribution mining, identifying high-frequency anomaly areas, quickly locating potential problems, and discovering high-frequency anomaly areas to help engineers and decision makers find problems early and take corresponding measures to avoid potential risks. Multi-point anomaly evolution tracking and multi-dimensional distribution visualization are performed on high-frequency anomaly areas, and an anomaly evolution trajectory visualization view is constructed to help understand the evolution process of abnormal data. Anomaly evolution trajectory visualization views are constructed to discover anomaly evolution laws and provide more intuitive references for anomaly handling and decision-making. Anomaly attribution quantitative analysis is performed on the anomaly evolution trajectory visualization view to obtain attribution data of different anomaly evolution trajectories, providing a basis for problem diagnosis and resolution. Adaptive anomaly optimization decisions are made based on attribution data of different anomaly evolution trajectories, and anomaly trajectory optimization strategies are constructed to improve problem solving efficiency and accuracy.

[0006] Preferably, step S1 comprises the following steps: Step S11: Obtain historical industrial logs; Step S12: Perform frequency-consistent multi-point sampling on historical industrial logs to extract multi-dimensional industrial data; Step S13: classify the multi-dimensional industrial data by data type, and extract parameterized industrial data and industrial image data; Step S14: performing mean filtering on the parameterized industrial data to obtain filtered optimized industrial data; Step S15: performing multi-dimensional convolution noise reduction processing on the industrial image data to obtain a convolution noise reduction image; Step S16: Perform cross-dimensional heterogeneous feature fusion on the filtered optimized industrial data and the convolution noise reduction image to generate a heterogeneous fused industrial feature vector.

[0007] The present invention obtains historical industrial logs to establish historical records of industrial processes, providing basic data for subsequent analysis and comparison. Frequency-consistent multi-point sampling is used to ensure data consistency, and multi-dimensional industrial data is extracted to fully understand the industrial production process. Multi-dimensional industrial data is classified and extracted, and the data is processed by type to improve the accuracy and efficiency of subsequent analysis. Mean filtering smoothes data, reduces noise, improves data quality and stability, and provides a more reliable data basis for subsequent analysis. Multi-dimensional convolution noise reduction processing processes noise and interference in industrial image data, improves image quality and clarity, and provides more accurate image data for subsequent analysis. Cross-dimensional heterogeneous feature fusion is performed on filtered optimized industrial data and convolution noise reduction images, and different types of data features are fused together to improve the richness of data dimensions and features. Heterogeneous fusion industrial feature vectors are generated to comprehensively analyze industrial data, discover deep correlations and laws between data, and provide support for optimization and decision-making of industrial production processes.

[0008] Preferably, the specific steps of step S16 are: Perform wavelet transform on the filtered optimized industrial data to extract the industrial frequency domain feature data; Perform abstract convolution feature extraction on the convolution denoised image to obtain the image abstract convolution feature; Perform dimensionality reduction compression and unification on the industrial frequency domain feature data and image abstract convolution features to obtain a frequency domain one-dimensional feature vector and an image abstract one-dimensional feature vector; Perform feature sequence aggregation on the frequency domain one-dimensional feature vector and the image abstract one-dimensional feature vector to obtain a one-dimensional feature vector sequence; Cross-dimensional heterogeneous feature fusion is performed on the one-dimensional feature vector sequence to generate a heterogeneous fused industrial feature vector.

[0009] The present invention effectively extracts the frequency domain features of industrial data through wavelet transform, helping to capture the periodicity and frequency domain information in the data. Extracting industrial frequency domain feature data understands the frequency domain structure of the data and reveals the hidden frequency domain features in the data. Abstract convolution feature extraction captures high-level abstract features from the image, and characterizes and analyzes the image. Obtaining the image abstract convolution feature extracts the important features of the image, providing a basis for subsequent feature fusion and analysis. Dimensionality reduction compression uniformly reduces the dimension of the data, removes redundant information, and improves data processing efficiency. Obtaining the frequency domain one-dimensional feature vector and the image abstract one-dimensional feature vector simplifies the data representation, providing a more efficient data structure for subsequent feature fusion. Feature sequence aggregation can fuse different feature vectors into a complete sequence, comprehensively considering the information of different features. Obtaining a one-dimensional feature vector sequence integrates features from different sources together, providing a more comprehensive data perspective. Cross-dimensional heterogeneous feature fusion organically integrates features from different dimensions and sources, and improves the expressiveness of features. Generating heterogeneous fusion industrial feature vectors comprehensively considers the information of different types of features, providing a more comprehensive feature representation for industrial data analysis and application.

[0010] Preferably, step S2 specifically comprises the following steps: Step S21: performing time series analysis on the heterogeneous fused industrial feature data to obtain time series feature data; Step S22: performing time series trend change analysis on the time series feature data to generate time series trend change data; Step S23: performing phase change fitting on the time series trend change data to construct a time series trend change curve; Step S24: performing morphological mutation identification on the time series trend change curve, and extracting a plurality of curve morphological step points; Step S25: Perform deep timing dependency mining on multiple curve morphology step points to obtain the step point timing dependency relationship.

[0011] The present invention organizes heterogeneous fused industrial feature data in chronological order through time series analysis, reveals the law of data change over time, obtains time series feature data, converts industrial feature data into time series form, and provides a basis for subsequent time series analysis. Time series trend change analysis reveals the trend characteristics of time series data over time, helps to understand the development and change law of data, generates time series trend change data to capture the dynamic change trend of data, and provides a basis for subsequent time series feature analysis. Stage change fitting fits time series data in sections, identifies the change trend and fitting curve of data in different stages, constructs time series trend change curve to analyze the stage change of data, and helps to understand the development process of data. Morphological mutation identification helps to identify the morphological change points in the curve, reveals important turning points and key features of data, extracts multiple curve morphological step points to capture the mutation characteristics of the curve, and provides key information for subsequent time series dependency mining. Deep time series dependency mining can analyze the time series dependency relationship between different step points in the curve, reveal the internal connection and regularity of data, obtain the step point time series dependency relationship to understand the dynamic mechanism of data change, and provide in-depth insights for the prediction and optimization of industrial data.

[0012] Preferably, step S3 specifically comprises the following steps: Step S31: performing inter-feature topological correlation analysis on heterogeneous fused industrial feature data based on the step point temporal dependency relationship, thereby obtaining feature topological correlation data; Step S32: performing feature space mapping on the heterogeneous fused industrial feature data to obtain feature mapping points; Step S33: Calculate the mapping space position of the feature mapping point to generate the mapping point spatial position coordinates; Step S34: reconstructing the mapping topology position of the spatial position coordinates of the mapping points according to the feature topology association data, and constructing an industrial feature mapping space model.

[0013] The present invention can discover the correlation and interaction between features through topological correlation analysis between features, reveal important correlation patterns in industrial feature data, obtain feature topological correlation data to understand the relationship between features, and provide key information for subsequent feature space mapping. Feature space mapping maps heterogeneous fused industrial feature data into a unified space, which is convenient for unified analysis and processing. The feature mapping points are obtained to convert the original data into a comparable spatial representation, which provides a basis for subsequent analysis. The mapping space position calculation determines the position of the feature mapping point in space, helps to understand the spatial distribution and relationship between features, generates mapping point spatial position coordinate visualization and spatial analysis, and provides a basis for subsequent mapping topological position reconstruction. The mapping topological position reconstruction can adjust the spatial position of the mapping point according to the topological correlation relationship between features, construct a feature mapping space model with correlation, and construct an industrial feature mapping space model to reveal the spatial relationship between features, providing a deeper understanding for the visualization and analysis of industrial data.

[0014] Preferably, the specific steps of step S4 are: Step S41: performing multi-dimensional anomaly detection on the industrial feature mapping space model to extract spatial multi-dimensional anomaly sample points; Step S42: Perform spatial distribution analysis on the spatial multi-dimensional abnormal sample points to obtain abnormal sample spatial distribution data; Step S43: performing cluster density calculation on the abnormal sample spatial distribution data to obtain an abnormal sample density value; Step S44: Based on the abnormal sample density value, the industrial feature mapping space model is mined for local abnormal distribution, thereby obtaining a high-frequency abnormal area.

[0015] The present invention identifies abnormal sample points in the mapping space model through multi-dimensional anomaly detection, and finds data points that show abnormalities in multiple dimensions. Extract abnormal patterns and features in the spatial multi-dimensional abnormal sample point identification data to provide a basis for subsequent abnormal analysis. Spatial distribution analysis helps understand the distribution law and trend of abnormal sample points in space, and reveals the spatial distribution characteristics of abnormal data. The spatial distribution data of abnormal samples is obtained to explore the spatial structure and distribution of abnormal data, providing a basis for subsequent clustering density calculation. Clustering density calculation can identify the clustering pattern and density distribution of abnormal sample points, and help find the clustering area and distribution law of abnormal data. The density distribution of abnormal data is determined by obtaining the abnormal sample density value, which provides key information for subsequent local abnormal distribution mining. Local abnormal distribution mining discovers high-frequency abnormal areas in the industrial feature mapping space model based on the density value of abnormal samples, and identifies local clustering areas of abnormal data. Obtaining high-frequency abnormal areas accurately locates abnormal hot spots in industrial data, providing important clues for abnormal detection and processing.

[0016] Preferably, the specific steps of step S5 are: Step S51: Tracking the multi-point anomaly evolution of the high-frequency anomaly area, thereby obtaining the anomaly evolution trajectories of multiple sample points in the area; Step S52: Perform dynamic spatial distribution evolution analysis on the abnormal evolution trajectories of multiple sample points in the region to generate multi-time point spatial distribution trajectories; Step S53: Perform multi-dimensional distribution visualization on the multi-time point spatial distribution trajectories, thereby constructing a visualization view of the abnormal evolution trajectory.

[0017] The present invention tracks the abnormal change trends and trajectories of multiple sample points in a high-frequency abnormal area through multi-point abnormal evolution tracking, revealing the evolution process of abnormal data. The abnormal evolution trajectories of multiple sample points in the region are obtained to understand the dynamic changes of abnormal data and provide time series information for subsequent analysis. Dynamic spatial distribution evolution analysis analyzes the spatial dynamic changes of the abnormal evolution trajectories of multiple sample points in the region, revealing the evolution trend and spatial distribution changes of abnormal data. Generate multi-time point spatial distribution trajectories to explore the spatiotemporal evolution laws of abnormal data, providing a data basis for subsequent multidimensional distribution visualization. Multidimensional distribution visualization can present multi-time point spatial distribution trajectories in an intuitive way and construct a visualization view of the abnormal evolution trajectory. Constructing a visualization view of the abnormal evolution trajectory comprehensively displays the dynamic evolution process of abnormal data, helping users to deeply understand the spatiotemporal characteristics of abnormal data.

[0018] Preferably, the specific steps of step S6 are: Step S61: classifying the abnormal trajectory evolution pattern of the abnormal evolution trajectory visualization view to obtain the abnormal evolution trajectory classification; Step S62: performing quantitative analysis of abnormal attribution on the abnormal evolution trajectory classification, thereby obtaining attribution data of different abnormal evolution trajectories; Step S63: performing anomaly risk assessment according to different anomaly evolution trajectory attribution data, thereby obtaining an anomaly evolution trajectory risk assessment result; Step S64: Make an adaptive abnormal optimization decision on the abnormal evolution trajectory risk assessment result and construct an abnormal trajectory optimization strategy.

[0019] The present invention classifies abnormal evolution trajectories according to their evolution patterns through abnormal trajectory evolution pattern classification, revealing different evolution types of abnormal data. The abnormal evolution trajectory classification is obtained to analyze and identify abnormal data in more detail, providing a classification basis for subsequent abnormal attribution quantitative analysis. Abnormal attribution quantitative analysis can quantitatively analyze the root causes and influencing factors of different abnormal evolution trajectories, revealing the generation mechanism of abnormal data. The attribution data of different abnormal evolution trajectories are obtained to understand the nature and characteristics of abnormal data, and provide a quantitative basis for abnormal risk assessment. Abnormal risk assessment evaluates and predicts abnormal situations based on the attribution data of different abnormal evolution trajectories, and quantifies the risk level of abnormal data. The risk assessment results of abnormal evolution trajectories are obtained to identify the risk level and potential impact of abnormal data, providing an important reference for subsequent decision-making. Adaptive abnormal optimization decision-making adjusts the decision-making strategy in real time according to the risk assessment results of abnormal evolution trajectories, and optimizes abnormal processing and countermeasures. Constructing abnormal trajectory optimization strategy takes effective countermeasures according to the actual situation and risk level of abnormal data, and improves the efficiency and accuracy of abnormal processing.

[0020] In this specification, an artificial intelligence-based industrial data processing system is provided, which is used to execute the artificial intelligence-based industrial data processing method as described above, including: The heterogeneous feature fusion module is used to obtain historical industrial logs; perform mean filtering on the historical industrial logs and perform cross-dimensional heterogeneous feature fusion to generate heterogeneous fused industrial feature data; The time series dependency module is used to obtain time series trend change data based on heterogeneous fusion industrial feature data; deep time series dependency mining is performed on the time series trend change data to obtain the step point time series dependency relationship; The feature space module is used to reconstruct the mapping topology position of heterogeneous fused industrial feature data based on the temporal dependency of step points and build an industrial feature mapping space model; The anomaly detection module is used to perform multi-dimensional anomaly detection on the industrial feature mapping spatial model and conduct local anomaly distribution mining to obtain high-frequency anomaly areas; The distribution visualization module is used to track the evolution of multi-point anomalies in high-frequency anomaly areas and perform multi-dimensional distribution visualization to build a visualization view of the anomaly evolution trajectory; The adaptive decision-making module is used to perform quantitative analysis of anomaly attribution on the visualization view of the anomaly evolution trajectory, so as to obtain the attribution data of different anomaly evolution trajectories; make adaptive anomaly optimization decisions based on the attribution data of different anomaly evolution trajectories, and build an anomaly trajectory optimization strategy.

[0021] The present invention obtains historical industrial logs and performs mean filtering to remove noise, improves data quality, integrates information of different dimensions through cross-dimensional heterogeneous feature fusion, and generates more comprehensive and representative industrial feature data. Based on heterogeneous fusion industrial feature data, time series trend change data is obtained to capture the time change law of data. Deep time series dependency mining reveals more complex time dependency relationships between data, providing deeper information for subsequent analysis. Mapping topological position reconstruction based on step point time dependency relationships can better understand the relationship and connection between data. An industrial feature mapping space model is constructed to effectively represent and analyze data in feature space. Multi-dimensional anomaly detection and local anomaly distribution mining accurately identify high-frequency anomaly areas, obtain high-frequency anomaly areas to help detect anomalies early, and improve the stability and security of industrial systems. Multi-point anomaly evolution tracking and multi-dimensional distribution visualization intuitively display the evolution process and characteristics of abnormal data. Construction of anomaly evolution trajectory visualization view helps users better understand the change law and trend of abnormal data. Anomaly attribution quantitative analysis and adaptive anomaly optimization decision-making deeply understand the root cause of abnormal data and formulate effective processing strategies. Construction of anomaly trajectory optimization strategy adjusts decisions in real time according to the actual situation of abnormal data, and improves problem solving efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a schematic diagram of the steps of an industrial data processing method based on artificial intelligence of the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0023] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0024] The present application example provides an industrial data processing method and system based on artificial intelligence. The execution subjects of the method and system include but are not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of audio and image management system, information management system, and cloud data management system.

[0025] See also Figures 1 to 4 The present invention provides an industrial data processing method based on artificial intelligence, and the industrial data processing method based on artificial intelligence comprises the following steps: Step S1: Obtain historical industrial logs; perform mean filtering on the historical industrial logs, and perform cross-dimensional heterogeneous feature fusion to generate heterogeneous fused industrial feature data; Step S2: obtaining time series trend change data based on heterogeneous fusion industrial feature data; performing deep time series dependency mining on the time series trend change data to obtain step point time series dependency relationships; Step S3: reconstruct the mapping topology position of the heterogeneous fused industrial feature data based on the step point temporal dependency relationship, and construct an industrial feature mapping space model; Step S4: Perform multi-dimensional anomaly detection on the industrial feature mapping spatial model and perform local anomaly distribution mining to obtain high-frequency anomaly areas; Step S5: Track the evolution of multi-point anomalies in high-frequency anomaly areas and visualize their multi-dimensional distribution, thereby constructing a visualization view of the anomaly evolution trajectory; Step S6: Perform anomaly attribution quantitative analysis on the visualization view of the anomaly evolution trajectory to obtain attribution data of different anomaly evolution trajectories; make adaptive anomaly optimization decisions based on the attribution data of different anomaly evolution trajectories, and construct an anomaly trajectory optimization strategy.

[0026] The present invention acquires historical industrial logs to establish a comprehensive understanding of industrial processes, provides important data support for subsequent analysis, and uses mean filtering and cross-dimensional heterogeneous feature fusion to smooth data and reduce noise, thereby generating more reliable heterogeneous fused industrial feature data. Time series trend change data is obtained based on heterogeneous fused industrial feature data to reveal the changing trend of industrial data and provide information support in the time dimension for subsequent analysis. Deep time series dependency mining helps discover the complex relationships between data, obtains step point time series dependency relationships, and provides more in-depth data mining results. By mapping the topological position of heterogeneous fused industrial feature data and reconstructing it, an industrial feature mapping space model is constructed to visualize and integrate data relationships, build an industrial feature mapping space model to understand the relationship between data, improve the efficiency and accuracy of data analysis, and Mapping spatial models for multi-dimensional anomaly detection and local anomaly distribution mining, identifying high-frequency anomaly areas, quickly locating potential problems, and discovering high-frequency anomaly areas to help engineers and decision makers find problems early and take corresponding measures to avoid potential risks. Multi-point anomaly evolution tracking and multi-dimensional distribution visualization are performed on high-frequency anomaly areas, and an anomaly evolution trajectory visualization view is constructed to help understand the evolution process of abnormal data. Anomaly evolution trajectory visualization views are constructed to discover anomaly evolution laws and provide more intuitive references for anomaly handling and decision-making. Anomaly attribution quantitative analysis is performed on the anomaly evolution trajectory visualization view to obtain attribution data of different anomaly evolution trajectories, providing a basis for problem diagnosis and resolution. Adaptive anomaly optimization decisions are made based on attribution data of different anomaly evolution trajectories, and anomaly trajectory optimization strategies are constructed to improve problem solving efficiency and accuracy.

[0027] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of the steps of an industrial data processing method based on artificial intelligence of the present invention. In this example, the steps of the industrial data processing method based on artificial intelligence include: Step S1: Obtain historical industrial logs; perform mean filtering on the historical industrial logs, and perform cross-dimensional heterogeneous feature fusion to generate heterogeneous fused industrial feature data; In this embodiment, historical monitoring log data of various equipment, processes, environment, etc. accumulated in the factory or production line for a long time are collected. These log data include various types of heterogeneous information such as sensor data, operation logs, fault reports, etc. The obtained original industrial log data are preprocessed. First, the mean filtering algorithm is used to eliminate noise interference. For different types of industrial log data, such as equipment parameters, environmental conditions, operating behaviors, etc., targeted feature extraction methods are designed. For example, the frequency domain features of equipment parameters are extracted by wavelet transform, and the visual features of images are extracted by convolutional neural network. These different types of features are compressed by dimensionality reduction and uniformly expressed as one-dimensional feature vectors. Finally, advanced feature fusion technology, such as attention mechanism or self-attention mechanism, is used to fuse these one-dimensional feature vectors across dimensions to generate highly abstract heterogeneous fused industrial feature data.

[0028] Step S2: obtaining time series trend change data based on heterogeneous fusion industrial feature data; performing deep time series dependency mining on the time series trend change data to obtain step point time series dependency relationships; In this embodiment, the heterogeneous fusion industrial feature vector sequence obtained in the previous step is divided into multiple time windows by using sliding window technology. In each time window, the statistical indicators of the feature vector, such as mean, variance, peak value, etc., are calculated as the time series feature data of the window. The time series feature data of each time window are connected in series to form a complete time series feature data sequence. The time series feature data is trend extracted by using time series analysis techniques, such as moving average, exponential smoothing, etc. The overall change trend of the time series feature data in the time dimension is extracted and quantified as trend change data. The time series trend change data is input into a piecewise function fitting model, such as piecewise linear fitting or piecewise polynomial fitting. Through the fitting process, the time series trend change data is divided into several change stages, each stage is represented by a continuous function. These continuous functions are spliced ​​together to construct a complete time series trend change curve. The time series trend change curve is analyzed for morphological mutation using anomaly detection algorithms, such as CUSUM control charts or probability transition models. The key time point at which the curve morphology changes significantly, i.e., the morphological step point, is identified. These morphological step points reflect the major turning points in the state of the industrial system over time. Using time series analysis and causal relationship mining technology, the time series dependencies between different morphological step points are analyzed. Through this deep time series dependency analysis, the potential driving factors and internal mechanisms of the changes in the state of the industrial system are discovered.

[0029] Step S3: reconstruct the mapping topology position of the heterogeneous fused industrial feature data based on the step point temporal dependency relationship, and construct an industrial feature mapping space model; In this embodiment, based on the temporal dependency between step points, each feature data sequence is mapped and reconstructed in topological position. Mapping topological position reconstruction refers to mapping each feature data sequence into a multidimensional space, where each dimension corresponds to a feature, and the position of each feature data sequence in space is determined by the temporal dependency between its step points. Multidimensional scaling (MDS) or other topological mapping methods are used to achieve mapping topological position reconstruction. All feature data sequences are mapped into the same multidimensional space to construct an industrial feature mapping space model. The model is used to visualize the relationship between different feature data sequences and analyze the mutual influence between different feature data sequences. A step point refers to a point in a data sequence where a significant change occurs, such as a sudden increase in temperature or a sudden decrease in pressure. Step points usually represent a change in the state of the system, so they are of great significance in industrial feature data analysis. The temporal dependency refers to the order and time interval of the appearance of step points between different feature data sequences, such as when the temperature increases, the pressure also increases, and the time interval is very short. The temporal dependency reflects the mutual influence between different feature data sequences. Mapping topological position reconstruction means mapping each feature data sequence into a multidimensional space, where each dimension corresponds to a feature, and the position of each feature data sequence in the space is determined by the temporal dependency between its step points. This visualizes the relationship between different feature data sequences and analyzes their mutual influence.

[0030] Step S4: Perform multi-dimensional anomaly detection on the industrial feature mapping spatial model and perform local anomaly distribution mining to obtain high-frequency anomaly areas; In this embodiment, a suitable anomaly detection algorithm is selected, such as a distance-based anomaly detection algorithm (such as KNN), a density-based method (such as DBSCAN), or a clustering-based anomaly detection algorithm (such as K-means), and all data points in the industrial feature mapping space model are used as input to perform multi-dimensional anomaly detection, identify abnormal data points, and perform local anomaly distribution mining on the identified abnormal data points, that is, analyze the distribution law of abnormal data points in the mapping space, and use cluster analysis or density estimation methods to mine local abnormal distribution, such as using the DBSCAN algorithm or Gaussian Mixture Model (GMM), local anomaly distribution mining helps identify areas where abnormal data points appear in large numbers, namely high-frequency anomaly areas. Based on the results of local anomaly distribution mining, high-frequency anomaly areas are identified. High-frequency anomaly areas refer to areas where abnormal data points appear in large numbers in the mapping space. These areas represent potential problems or failures in the system operation process. Multidimensional anomaly detection refers to identifying data points that are significantly different from other data points in the multidimensional data space. In the industrial feature mapping space model, each data point represents a feature data sequence, and abnormal data points represent abnormal situations in the system operation process. Local anomaly distribution mining refers to analyzing the distribution pattern of abnormal data points in the mapping space and identifying areas where abnormal data points appear in large numbers.

[0031] Step S5: Track the evolution of multi-point anomalies in high-frequency anomaly areas and visualize their multi-dimensional distribution, thereby constructing a visualization view of the anomaly evolution trajectory; In this embodiment, in the high-frequency anomaly area, multiple abnormal data points are selected as tracking targets, and the evolution trajectories of these abnormal data points in the time dimension are tracked, that is, their positions in the mapping space are observed to change over time. Time series analysis methods, such as sliding windows or autoregressive models, are used to track the evolution trajectories of the abnormal data points, and the positions of the tracked abnormal data points in the mapping space are visualized to show their evolution trajectories. Scatter plots, line graphs, or three-dimensional graphics are used for visualization to show the distribution and change trends of the abnormal data points in different dimensions. In order to better display multidimensional data, dimensionality reduction techniques, such as principal component analysis (PCA) or t-SNE, are used to reduce high-dimensional data to two-dimensional or three-dimensional space for visualization. The multidimensional distribution visualization results are integrated into a visualization view to construct a visualization view of the abnormal evolution trajectory, which displays the evolution trajectories of multiple abnormal data points in the high-frequency anomaly area and their mutual relationships. Interactive visualization tools, such as Tableau or D3.js, are used to construct a visualization view of the abnormal evolution trajectory to facilitate users to conduct interactive exploration and analysis.

[0032] Step S6: Perform anomaly attribution quantitative analysis on the visualization view of the anomaly evolution trajectory to obtain attribution data of different anomaly evolution trajectories; make adaptive anomaly optimization decisions based on the attribution data of different anomaly evolution trajectories, and construct an anomaly trajectory optimization strategy.

[0033] In this embodiment, the visualization view of the abnormal evolution trajectory is analyzed to identify the characteristics of different abnormal evolution trajectories, such as the shape, direction, speed, etc. of the trajectory. Based on these characteristics, each abnormal evolution trajectory is quantitatively analyzed to determine its cause and influencing factors. A statistical analysis method, such as regression analysis or principal component analysis, is used to perform an abnormal attribution quantitative analysis. Based on the results of the abnormal attribution quantitative analysis, attribution data is generated for each abnormal evolution trajectory, such as the cause of the abnormality, the weight of the influencing factors, etc. These attribution data help us understand the reasons behind different abnormal evolution trajectories and their impact on the system. According to the attribution data of different abnormal evolution trajectories, adaptive abnormal optimization decisions are made, that is, different optimization measures are taken according to different abnormal situations. For example, if a certain abnormal evolution trajectory is caused by a sensor failure, then measures such as replacing the sensor or calibrating are taken to solve the problem. The adaptive abnormal optimization decision is integrated into a strategy to construct an abnormal trajectory optimization strategy. The strategy automatically selects appropriate optimization measures according to different abnormal evolution trajectories, thereby improving the reliability and stability of the system.

[0034] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Step S11: Obtain historical industrial logs; Step S12: Perform frequency-consistent multi-point sampling on historical industrial logs to extract multi-dimensional industrial data; Step S13: classify the multi-dimensional industrial data by data type, and extract parameterized industrial data and industrial image data; Step S14: performing mean filtering on the parameterized industrial data to obtain filtered optimized industrial data; Step S15: performing multi-dimensional convolution noise reduction processing on the industrial image data to obtain a convolution noise reduction image; Step S16: Perform cross-dimensional heterogeneous feature fusion on the filtered optimized industrial data and the convolution noise reduction image to generate a heterogeneous fused industrial feature vector.

[0035] In this embodiment, the source of the historical industrial log is determined, such as PLC data, sensor data, database records, etc. The format of the historical industrial log is identified, such as text files, CSV files, database records, etc. According to the data source and format, the historical industrial log data is obtained using appropriate tools or methods. According to the actual needs and data characteristics, the appropriate sampling frequency is determined, such as sampling once per second, sampling once per minute, etc. The historical industrial log is sampled at multiple points, such as collecting data from multiple sensors and the operating status of multiple devices at the same time. The collected data is preprocessed, such as data cleaning, data conversion, etc., to ensure the integrity and consistency of the data. According to the data characteristics and application scenarios, the data type is identified, such as numerical data, text data, image data, etc. Numerical data, such as temperature, pressure, flow, etc., are extracted, and data format conversion is performed, such as converting string types to numerical types. Image data, such as images captured by cameras, images collected by sensors, etc., are extracted, and image format conversion is performed, such as converting original images to standard image formats. Select a suitable mean filter, such as a sliding average filter, a weighted average filter, etc., according to actual needs and data characteristics. According to the data noise level and filtering effect, determine the size of the filter window, such as 3-point mean filtering, 5-point mean filtering, etc. Use the selected filter and window size to perform mean filtering on the parameterized industrial data to obtain filtered optimized industrial data. Select a suitable convolution kernel, such as Gaussian kernel, Laplace kernel, etc., according to the image noise type and noise reduction effect. According to the image noise level and noise reduction effect, determine the size of the convolution kernel, such as 3x3 convolution kernel, 5x5 convolution kernel, etc. Use the selected convolution kernel and size to perform multi-dimensional convolution noise reduction on the industrial image data to obtain a convolution noise reduction image. Extract features from the filtered optimized industrial data and the convolution noise reduction image, such as statistical features, texture features, shape features, etc. Select a suitable cross-dimensional heterogeneous feature fusion method, such as principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network, etc., according to actual needs and data characteristics. Use the selected feature fusion method to fuse the features of the filtered optimized industrial data and the convolution noise reduction image to generate a heterogeneous fused industrial feature vector.

[0036] In this embodiment, the specific steps of step S16 are: Perform wavelet transform on the filtered optimized industrial data to extract the industrial frequency domain feature data; Perform abstract convolution feature extraction on the convolution denoised image to obtain the image abstract convolution feature; Perform dimensionality reduction compression and unification on the industrial frequency domain feature data and image abstract convolution features to obtain a frequency domain one-dimensional feature vector and an image abstract one-dimensional feature vector; Perform feature sequence aggregation on the frequency domain one-dimensional feature vector and the image abstract one-dimensional feature vector to obtain a one-dimensional feature vector sequence; Cross-dimensional heterogeneous feature fusion is performed on the one-dimensional feature vector sequence to generate a heterogeneous fused industrial feature vector.

[0037] In this embodiment, various monitoring data collected during the industrial production process are filtered and optimized to eliminate noise interference. Then, the filtered data sequence is subjected to wavelet transform to obtain the characteristic components of the data in different frequency bands. The key frequency bands containing key process information are selected to extract the industrial frequency domain feature data. The image data collected on the industrial site is subjected to noise reduction processing using a convolutional neural network to remove environmental interference. The output of the hidden layer of the convolutional network is extracted as the abstract convolution feature of the image, which can reflect the high-level semantic information of the image. These abstract features can effectively describe the visual features of equipment and processes in industrial scenes. For high-dimensional industrial frequency domain feature data and image abstract features, principal component analysis (PCA) and other methods are used for dimensionality reduction and compression. The compressed feature data is unified into a one-dimensional feature vector, including a frequency domain one-dimensional feature vector and an image abstract one-dimensional feature vector. This makes the feature expression forms of different types of industrial data tend to be unified, preparing for subsequent feature fusion. The compressed frequency domain one-dimensional feature vector and the image abstract one-dimensional feature vector are organized into a feature vector sequence in time series order. These feature vector sequences reflect the evolution of multi-source heterogeneous features of industrial systems over time. By performing time window aggregation on the feature vector sequence, higher-level time series features can be further extracted. The one-dimensional feature vector sequence obtained in the previous step is fused across dimensions using the attention mechanism or self-attention mechanism. This fusion process can mine the implicit associations between different types of data sources to form richer and more representative heterogeneous feature vectors. This heterogeneous feature fusion method makes full use of the multi-dimensional semantic information contained in industrial data, providing strong support for subsequent intelligent analysis and decision-making.

[0038] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Step S21: performing time series analysis on the heterogeneous fused industrial feature data to obtain time series feature data; Step S22: performing time series trend change analysis on the time series feature data to generate time series trend change data; Step S23: performing phase change fitting on the time series trend change data to construct a time series trend change curve; Step S24: performing morphological mutation identification on the time series trend change curve, and extracting a plurality of curve morphological step points; Step S25: Perform deep timing dependency mining on multiple curve morphology step points to obtain the step point timing dependency relationship.

[0039] In this embodiment, the sliding window technology is used to divide the heterogeneous fusion industrial feature vector sequence obtained in the previous step into multiple time windows. In each time window, the statistical indicators of the feature vector, such as mean, variance, peak value, etc., are calculated as the time series feature data of the window. The time series feature data of each time window are connected in series to form a complete time series feature data sequence. Key features with time dependence are extracted from high-dimensional heterogeneous features. Trend extraction is performed on the time series feature data using time series analysis techniques, such as moving average, exponential smoothing, etc. The overall change trend of the time series feature data in the time dimension is extracted and quantified as trend change data. This step effectively eliminates noise interference and highlights the potential laws of the industrial system state in time. The time series trend change data is input into a piecewise function fitting model, such as piecewise linear fitting or piecewise polynomial fitting. Through the fitting process, the time series trend change data is divided into several change stages, each stage is represented by a continuous function. These continuous functions are spliced ​​together to construct a complete time series trend change curve. Use anomaly detection algorithms, such as CUSUM control charts or probability transition models, to perform morphological mutation analysis on the time series trend change curve. Identify the key time points where the curve morphology changes significantly, namely the morphological step points. These morphological step points reflect the significant turning points in the state of the industrial system over time. Use time series analysis and causal relationship mining technology to analyze the time series dependencies between different morphological step points. Through this deep time series dependency analysis, the potential driving factors and internal mechanisms of changes in the state of the industrial system are discovered. Better understand the time correlation between subsystems or process links in the industrial process, and provide support for fault diagnosis and process optimization.

[0040] In this embodiment, refer to Figure 4 , is a flowchart of detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Step S31: performing inter-feature topological correlation analysis on heterogeneous fused industrial feature data based on the step point temporal dependency relationship, thereby obtaining feature topological correlation data; Step S32: performing feature space mapping on the heterogeneous fused industrial feature data to obtain feature mapping points; Step S33: Calculate the mapping space position of the feature mapping point to generate the mapping point spatial position coordinates; Step S34: reconstructing the mapping topology position of the spatial position coordinates of the mapping points according to the feature topology association data, and constructing an industrial feature mapping space model.

[0041] In this embodiment, the step point temporal dependency relationship obtained in the previous step is used to analyze the mutual dependence degree and time sequence relationship between different industrial features. A topological association diagram between features is constructed, and the edge weight is used to represent the association strength between features, and the edge vector is used to represent the temporal dependency direction between features. These feature topological association information is organized into structured feature topological association data to provide a basis for subsequent feature space construction. Dimensionality reduction techniques, such as principal component analysis (PCA) or t-SNE, are used to map high-dimensional heterogeneous fusion industrial feature data to low-dimensional feature space. Each industrial feature data sample corresponds to a coordinate point in the low-dimensional feature space, that is, a feature mapping point. The specific coordinate value of each feature mapping point in the low-dimensional feature space is calculated to form a set of spatial position coordinates. These coordinate values ​​reflect the relative position relationship of each industrial feature in the feature space. By analyzing these position coordinates, the association distribution between features is preliminarily observed. Using the feature topological association data obtained in the previous step, the spatial position coordinates of the feature mapping points are further topologically reconstructed. Through optimization algorithms, such as force-directed layout algorithms, the positions of feature mapping points in the feature space can better reflect the topological relationship between them. The reconstructed feature mapping point coordinates are organized into a complete industrial feature mapping space model, providing a basis for subsequent intelligent analysis.

[0042] In this embodiment, step S4 includes the following steps: Step S41: performing multi-dimensional anomaly detection on the industrial feature mapping space model to extract spatial multi-dimensional anomaly sample points; Step S42: Perform spatial distribution analysis on the spatial multi-dimensional abnormal sample points to obtain abnormal sample spatial distribution data; Step S43: performing cluster density calculation on the abnormal sample spatial distribution data to obtain an abnormal sample density value; Step S44: Based on the abnormal sample density value, the industrial feature mapping space model is mined for local abnormal distribution, thereby obtaining a high-frequency abnormal area.

[0043] In this embodiment, an anomaly detection algorithm, such as outlier detection based on Gaussian mixture model (GMM), is used to comprehensively scan the data points in the industrial feature mapping space model. Identify those points that deviate significantly from the normal data distribution, that is, abnormal sample points in the multidimensional space. Analyze the specific distribution of abnormal sample points in the industrial feature mapping space, including location coordinates, area range, etc. The spatial distribution information of these abnormal samples is organized into structured data to form abnormal sample spatial distribution data. A density clustering algorithm, such as DBSCAN, is used to perform cluster analysis on the abnormal sample spatial distribution data. The sample density of each cluster cluster is calculated, that is, the average density of sample points in the cluster. Combined with the abnormal sample density value obtained in the previous step, the local abnormal distribution analysis of the industrial feature mapping space is performed. Identify those areas with high abnormal sample density, that is, high-frequency abnormal areas in the feature mapping space. These high-frequency abnormal areas reflect potential fault hazards or abnormal working conditions in the industrial system, providing important clues for subsequent root cause analysis.

[0044] In this embodiment, the specific steps of step S5 are: Step S51: Tracking the multi-point anomaly evolution of the high-frequency anomaly area, thereby obtaining the anomaly evolution trajectories of multiple sample points in the area; Step S52: Perform dynamic spatial distribution evolution analysis on the abnormal evolution trajectories of multiple sample points in the region to generate multi-time point spatial distribution trajectories; Step S53: Perform multi-dimensional distribution visualization on the multi-time point spatial distribution trajectories, thereby constructing a visualization view of the abnormal evolution trajectory.

[0045] In this embodiment, the industrial feature mapping space model constructed previously is used to identify high-frequency abnormal areas in the feature space. For multiple sample points in these abnormal areas, their abnormal evolution process in the time series is tracked, and the abnormal evolution trajectory of each sample point is recorded. These abnormal evolution trajectories reflect the dynamic changes of the abnormal state of the industrial system in time and space. Based on the abnormal evolution trajectories of multiple sample points, they are aggregated and analyzed according to the time series, and the spatial distribution characteristics of the sample points in the abnormal area at each time point are calculated, such as the average position, variance and other indicators. The spatial distribution characteristics on these time series are connected to construct the spatial distribution trajectories of multiple time points. This step reflects the overall evolution dynamics of the abnormal area in time. The multi-time point spatial distribution trajectories obtained in the previous step are intuitively displayed using visualization techniques such as heat maps and scatter plots. In the same view, multi-dimensional information such as the boundary contour of the abnormal area, the abnormal evolution trajectory of the sample points, and the spatial distribution trajectory are integrated.

[0046] In this embodiment, the specific steps of step S6 are: Step S61: classifying the abnormal trajectory evolution pattern of the abnormal evolution trajectory visualization view to obtain the abnormal evolution trajectory classification; Step S62: performing quantitative analysis of abnormal attribution on the abnormal evolution trajectory classification, thereby obtaining attribution data of different abnormal evolution trajectories; Step S63: performing anomaly risk assessment according to different anomaly evolution trajectory attribution data, thereby obtaining an anomaly evolution trajectory risk assessment result; Step S64: Make an adaptive abnormal optimization decision on the abnormal evolution trajectory risk assessment result and construct an abnormal trajectory optimization strategy.

[0047] In this embodiment, a machine learning algorithm, such as cluster analysis or pattern recognition, is used to analyze the visualization view of the abnormal evolution trajectory constructed in the previous step. Different types of abnormal evolution trajectory patterns are identified, such as rapid deterioration type, slow deterioration type, periodic fluctuation type, etc. The observed abnormal evolution trajectory is classified into different pattern categories to form the abnormal evolution trajectory classification result. Combined with process knowledge and expert experience, the potential root causes of different types of abnormal evolution trajectories are analyzed. For each abnormal evolution pattern, the degree of association between it and each root cause factor is quantified to form abnormal attribution data. This root cause attribution analysis deeply understands the mechanism of abnormal state of industrial system. An abnormal risk assessment model is designed, and the abnormal evolution trajectory attribution data obtained above is used as an input variable. According to the degree of influence of different root causes on abnormal development, the risk level of various abnormal evolution trajectories is quantitatively calculated. The risk assessment results are classified or sorted to form a complete abnormal evolution trajectory risk assessment result. This step helps industrial enterprises better identify and control major abnormal risks. Combined with historical experience data and expert knowledge, an abnormal optimization decision strategy based on risk assessment results is designed. According to different types of high-risk abnormal evolution trajectories, corresponding process adjustments, equipment maintenance or other optimization measures are formulated. A closed-loop optimization feedback mechanism is established to dynamically adjust the optimization strategy according to the actual evolution effect of the abnormal state. Through this adaptive abnormal optimization decision, major risks in industrial system operations can be effectively reduced.

[0048] In this embodiment, an artificial intelligence-based industrial data processing system is provided, which is used to execute the artificial intelligence-based industrial data processing method as described above, including: The heterogeneous feature fusion module is used to obtain historical industrial logs; perform mean filtering on the historical industrial logs and perform cross-dimensional heterogeneous feature fusion to generate heterogeneous fused industrial feature data; The time series dependency module is used to obtain time series trend change data based on heterogeneous fusion industrial feature data; deep time series dependency mining is performed on the time series trend change data to obtain the step point time series dependency relationship; The feature space module is used to reconstruct the mapping topology position of heterogeneous fused industrial feature data based on the temporal dependency of step points and build an industrial feature mapping space model; The anomaly detection module is used to perform multi-dimensional anomaly detection on the industrial feature mapping spatial model and conduct local anomaly distribution mining to obtain high-frequency anomaly areas; The distribution visualization module is used to track the evolution of multi-point anomalies in high-frequency anomaly areas and perform multi-dimensional distribution visualization to build a visualization view of the anomaly evolution trajectory; The adaptive decision-making module is used to perform quantitative analysis of anomaly attribution on the visualization view of the anomaly evolution trajectory, so as to obtain the attribution data of different anomaly evolution trajectories; make adaptive anomaly optimization decisions based on the attribution data of different anomaly evolution trajectories, and build an anomaly trajectory optimization strategy.

[0049] The present invention obtains historical industrial logs and performs mean filtering to remove noise, improves data quality, integrates information of different dimensions through cross-dimensional heterogeneous feature fusion, and generates more comprehensive and representative industrial feature data. Based on heterogeneous fusion industrial feature data, time series trend change data is obtained to capture the time change law of data. Deep time series dependency mining reveals more complex time dependency relationships between data, providing deeper information for subsequent analysis. Mapping topological position reconstruction based on step point time dependency relationships can better understand the relationship and connection between data. An industrial feature mapping space model is constructed to effectively represent and analyze data in feature space. Multi-dimensional anomaly detection and local anomaly distribution mining accurately identify high-frequency anomaly areas, obtain high-frequency anomaly areas to help detect anomalies early, and improve the stability and security of industrial systems. Multi-point anomaly evolution tracking and multi-dimensional distribution visualization intuitively display the evolution process and characteristics of abnormal data. Construction of anomaly evolution trajectory visualization view helps users better understand the change law and trend of abnormal data. Anomaly attribution quantitative analysis and adaptive anomaly optimization decision-making deeply understand the root cause of abnormal data and formulate effective processing strategies. Construction of anomaly trajectory optimization strategy adjusts decisions in real time according to the actual situation of abnormal data, and improves problem solving efficiency and accuracy.

[0050] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0051] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An industrial data processing method based on artificial intelligence, characterized in that: The following steps are involved: Step S1: Obtain historical industrial logs; perform mean filtering on the historical industrial logs, and perform cross-dimensional heterogeneous feature fusion to generate heterogeneous fused industrial feature data; Step S2: obtaining time series trend change data based on heterogeneous fusion industrial feature data; Conduct deep time series dependency mining on time series trend change data to obtain step point time series dependency relationships; Step S3: reconstructing the mapping topology position of the heterogeneous fused industrial feature data based on the step point temporal dependency relationship, and constructing an industrial feature mapping space model; Step S4: Perform multi-dimensional anomaly detection on the industrial feature mapping spatial model and perform local anomaly distribution mining to obtain high-frequency anomaly areas; Step S5: Track the evolution of multi-point anomalies in high-frequency anomaly areas and visualize their multi-dimensional distribution, thereby constructing a visualization view of the anomaly evolution trajectory; Step S6: Perform anomaly attribution quantitative analysis on the visualization view of the anomaly evolution trajectory to obtain attribution data of different anomaly evolution trajectories; make adaptive anomaly optimization decisions based on the attribution data of different anomaly evolution trajectories, and construct an anomaly trajectory optimization strategy.

2. The industrial data processing method based on artificial intelligence according to claim 1 is characterized in that: The specific steps of step S1 are: Step S11: Obtain historical industrial logs; Step S12: Perform frequency-consistent multi-point sampling on historical industrial logs to extract multi-dimensional industrial data; Step S13: classify the multi-dimensional industrial data by data type, and extract parameterized industrial data and industrial image data; Step S14: performing mean filtering on the parameterized industrial data to obtain filtered optimized industrial data; Step S15: performing multi-dimensional convolution noise reduction processing on the industrial image data to obtain a convolution noise reduction image; Step S16: Perform cross-dimensional heterogeneous feature fusion on the filtered optimized industrial data and the convolution noise reduction image to generate a heterogeneous fused industrial feature vector.

3. The industrial data processing method based on artificial intelligence according to claim 2 is characterized in that: The specific steps of step S16 are: Perform wavelet transform on the filtered optimized industrial data to extract the industrial frequency domain feature data; Perform abstract convolution feature extraction on the convolution denoised image to obtain the image abstract convolution feature; Perform dimensionality reduction compression and unification on the industrial frequency domain feature data and image abstract convolution features to obtain a frequency domain one-dimensional feature vector and an image abstract one-dimensional feature vector; Perform feature sequence aggregation on the frequency domain one-dimensional feature vector and the image abstract one-dimensional feature vector to obtain a one-dimensional feature vector sequence; Cross-dimensional heterogeneous feature fusion is performed on the one-dimensional feature vector sequence to generate a heterogeneous fused industrial feature vector.

4. The industrial data processing method based on artificial intelligence according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: performing time series analysis on the heterogeneous fused industrial feature data to obtain time series feature data; Step S22: performing time series trend change analysis on the time series feature data to generate time series trend change data; Step S23: performing phase change fitting on the time series trend change data to construct a time series trend change curve; Step S24: performing morphological mutation identification on the time series trend change curve, and extracting a plurality of curve morphological step points; Step S25: Perform deep timing dependency mining on multiple curve morphology step points to obtain the step point timing dependency relationship.

5. The method according to claim 1, characterized in that The specific steps of step S3 are: Step S31: performing inter-feature topological correlation analysis on heterogeneous fused industrial feature data based on the step point temporal dependency relationship, thereby obtaining feature topological correlation data; Step S32: performing feature space mapping on the heterogeneous fused industrial feature data to obtain feature mapping points; Step S33: Calculate the mapping space position of the feature mapping point to generate the mapping point spatial position coordinates; Step S34: reconstructing the mapping topology position of the spatial position coordinates of the mapping points according to the feature topology association data, and constructing an industrial feature mapping space model.

6. The method according to claim 1, characterized in that The specific steps of step S4 are: Step S41: performing multi-dimensional anomaly detection on the industrial feature mapping space model to extract spatial multi-dimensional anomaly sample points; Step S42: Perform spatial distribution analysis on the spatial multi-dimensional abnormal sample points to obtain abnormal sample spatial distribution data; Step S43: performing cluster density calculation on the abnormal sample spatial distribution data to obtain an abnormal sample density value; Step S44: Based on the abnormal sample density value, the industrial feature mapping space model is mined for local abnormal distribution, thereby obtaining a high-frequency abnormal area.

7. The industrial data processing method based on artificial intelligence according to claim 1 is characterized in that: The specific steps of step S5 are: Step S51: Tracking the multi-point anomaly evolution of the high-frequency anomaly area, thereby obtaining the anomaly evolution trajectories of multiple sample points in the area; Step S52: Perform dynamic spatial distribution evolution analysis on the abnormal evolution trajectories of multiple sample points in the region to generate multi-time point spatial distribution trajectories; Step S53: Perform multi-dimensional distribution visualization on the multi-time point spatial distribution trajectories, thereby constructing a visualization view of the abnormal evolution trajectory.

8. The industrial data processing method based on artificial intelligence according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: classifying the abnormal trajectory evolution pattern of the abnormal evolution trajectory visualization view to obtain the abnormal evolution trajectory classification; Step S62: performing quantitative analysis of abnormal attribution on the abnormal evolution trajectory classification, thereby obtaining attribution data of different abnormal evolution trajectories; Step S63: performing anomaly risk assessment according to different anomaly evolution trajectory attribution data, thereby obtaining an anomaly evolution trajectory risk assessment result; Step S64: Make an adaptive abnormal optimization decision on the abnormal evolution trajectory risk assessment result and construct an abnormal trajectory optimization strategy.

9. An industrial data processing system based on artificial intelligence, characterized in that: Used to execute the artificial intelligence-based industrial data processing method as claimed in claim 1, comprising: The heterogeneous feature fusion module is used to obtain historical industrial logs; perform mean filtering on the historical industrial logs and perform cross-dimensional heterogeneous feature fusion to generate heterogeneous fused industrial feature data; The time series dependency module is used to obtain time series trend change data based on heterogeneous fusion industrial feature data; deep time series dependency mining is performed on the time series trend change data to obtain the step point time series dependency relationship; The feature space module is used to reconstruct the mapping topology position of heterogeneous fused industrial feature data based on the temporal dependency of step points and build an industrial feature mapping space model; The anomaly detection module is used to perform multi-dimensional anomaly detection on the industrial feature mapping spatial model and conduct local anomaly distribution mining to obtain high-frequency anomaly areas; The distribution visualization module is used to track the evolution of multi-point anomalies in high-frequency anomaly areas and perform multi-dimensional distribution visualization to build a visualization view of the anomaly evolution trajectory; The adaptive decision-making module is used to perform quantitative analysis of anomaly attribution on the visualization view of the anomaly evolution trajectory, so as to obtain the attribution data of different anomaly evolution trajectories; make adaptive anomaly optimization decisions based on the attribution data of different anomaly evolution trajectories, and build an anomaly trajectory optimization strategy.

Images