An AI-based method, device, and medium for optimizing equipment operation control.

Abstract

This invention discloses an artificial intelligence-based method, device, and medium for optimizing equipment operation control, relating to the field of industrial automation control technology. The method includes: constructing a continuous operating state feature vector based on an enhanced multi-granularity time series; jointly encoding the feature vector using a temporal attention method to generate an equipment operating state description and a predicted equipment operating state value; receiving the operating results after receiving a real-time control command sequence; comparing and analyzing the operating results with the predicted equipment operating state value; and updating the equipment operating state description. This invention, by performing feature mapping and enhanced encoding on abnormal data, abnormal fluctuations, and statistical characteristics of multi-granularity time series, can accurately characterize the equipment operating state, retain abnormal information and time dependencies, provide high-quality input for constructing continuous operating state feature vectors, and improve the accuracy of state prediction and real-time control optimization capabilities.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and in particular to an artificial intelligence-based method, device, and medium for optimizing equipment operation control. Background Technology

[0002] With the improvement of industrial automation and intelligence, a large amount of multi-source heterogeneous data is generated during the operation of industrial equipment, including equipment status, operating parameters, control commands and work logs. The application of artificial intelligence in industrial control is constantly expanding, showing important value in equipment status analysis, anomaly detection and control strategy optimization. Modeling and predicting equipment operation through deep learning and time series analysis has become a research focus.

[0003] Existing technologies still have shortcomings in multi-source heterogeneous data processing and abnormal state analysis. They lack dynamic mapping and joint coding capabilities in multi-timescale data fusion and abnormal feature representation, resulting in coarse granularity of operation status identification, lagging traceability of abnormal root causes, and difficulty in achieving adaptive optimization control for complex operating conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based equipment operation control optimization method to solve the problem of insufficient accuracy in processing multi-source heterogeneous data and representing abnormal state characteristics of industrial equipment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an artificial intelligence-based method for optimizing equipment operation control, comprising: collecting multi-source heterogeneous time-series data and organizing it into an initial time series; asynchronously completing and removing low-confidence data for missing and outlier values ​​in the initial time series to generate a complete time series; dividing the complete time series into time windows and dynamically resampling the data in each time window to generate a multi-granularity time series; performing feature mapping and enhanced encoding on the abnormal data, abnormal fluctuations, and statistical characteristics of each time window in the multi-granularity time series based on the historical data of the complete time series and the statistical characteristics of the time windows to obtain an enhanced multi-granularity time series; constructing a continuous operating state feature vector based on the enhanced multi-granularity time series and jointly encoding the feature vector using a temporal attention method to generate an equipment operating state description and a predicted equipment operating state value; calculating the optimized value of each operating parameter based on the equipment operating state description, generating a real-time control command sequence, and executing the equipment operation; receiving the operation results after receiving the real-time control command sequence, comparing and analyzing the operation results with the predicted equipment operating state value, and updating the equipment operating state description.

[0007] As a preferred embodiment of the AI-based equipment operation control optimization method of the present invention, the specific steps for generating the complete time series are as follows: Multi-source heterogeneous time series data are organized and merged in chronological order to obtain an initial time series; Extract the missing and outlier values ​​identified in the initial time series, and use the characteristics of consecutive time points and historical complete time series to perform asynchronous completion on the missing and outlier values, generating continuous and complete data segments; Confidence assessments are performed on continuous and complete data segments, low-confidence data are removed, and the data are recombined to generate a complete time series.

[0008] As a preferred embodiment of the AI-based equipment operation control optimization method of the present invention, the specific steps for generating multi-granularity time series are as follows: The complete time series is divided into time windows according to the second, minute and hour levels to obtain a preliminary set of time windows; Missing values ​​are filled and outliers are identified in the data in the initial time window set. The continuous data sequence of each window is extracted and resampled to obtain the continuous data set within each time window. Based on the dataset, continuous data sequences within each time window are combined and integrated according to different time scales to generate multi-granularity time series.

[0009] As a preferred embodiment of the AI-based equipment operation control optimization method of the present invention, the specific steps for obtaining the enhanced multi-granularity time series are as follows: Extract outlier data and outlier fluctuations from each time window in the multi-granularity time series, and perform preliminary feature processing with statistical features to generate a preliminary feature set; Based on the preliminary feature set, correlation analysis is performed on abnormal data and abnormal fluctuations to identify feature dimensions that are associated with statistical features. Using the results of correlation analysis, abnormal data, abnormal fluctuations and statistical characteristics are grouped and reorganized, and each group of features is mapped to a continuous state space to generate an intermediate feature set for state description, while determining whether there are potential abnormal features. Information enhancement and integration are performed on the intermediate feature set, and the features of each time window are enhanced and encoded to generate an enhanced multi-granular time series.

[0010] As a preferred embodiment of the AI-based equipment operation control optimization method of the present invention, the specific steps for constructing a continuous operating state feature vector based on the enhanced multi-granularity time series are as follows: Based on the enhanced multi-granularity time series, feature extraction and information integration are performed on the abnormal data, abnormal fluctuations and statistical characteristics of each time window to obtain a continuous operating state feature set. For time series features in a continuous operating state feature set, determine the degree of deviation of each feature from its historical and predicted range, and identify abnormal and potential abnormal features; Based on the judgment results, the time series features are jointly encoded and combined, and the abnormal features are integrated with the normal features through dependency association to generate a continuous running state feature vector.

[0011] As a preferred embodiment of the AI-based equipment operation control optimization method of the present invention, the specific steps for generating the equipment operation status description and the equipment operation status prediction value are as follows: Extract data from each time period in the feature vector of continuous operation status, and perform time-dependent extraction and joint integration of abnormal data, abnormal fluctuations and statistical features to obtain the joint features of each time period; Perform time-series dependency analysis and correlation extraction on the joint features of each time period, determine whether the abnormal features of each time period deviate from the historical and prediction range, and generate a complete description of the equipment operating status. The judgment results are used to correct the predicted values ​​of each operating parameter within a future time window, generate predicted values ​​of equipment operating status, and identify operating parameters that deviate from the predicted values.

[0012] As a preferred embodiment of the AI-based equipment operation control optimization method of the present invention, the specific steps of generating a real-time control command sequence and executing equipment operation are as follows: Extract the current state value and trend of each operating parameter from the equipment operating status description and continuous operating status feature vector to obtain the corresponding optimization target range; Based on the optimization target range and state description, the current state value is transformed and reorganized to generate a real-time control command sequence. It is then determined whether each command meets the target adjustment direction and amplitude requirements, and adjustments are made to commands that do not meet the requirements. The system executes a real-time control command sequence to the device, determines the consistency between the device's execution feedback and the command, and collects the device's execution feedback information.

[0013] As a preferred embodiment of the AI-based device operation control optimization method of the present invention, the specific steps for updating the device operation status description are as follows: The system obtains the operating parameter values, sensor readings and control signals for each time period from the operating results after the real-time control command of the equipment, and extracts the data sequence of state update by combining the continuous operating state feature vector. The extracted data sequence is compared with the predicted values ​​of equipment operating status to determine whether each operating parameter and fluctuation characteristic deviates from the prediction range, and corresponding operating difference information is generated. By utilizing operational difference information, the parameter values, feature representation vectors, and feedback information in the continuous operation state feature vector are adjusted and updated. It is then determined whether the updated state meets the continuous operation conditions, and a corrected description of the equipment operation state is obtained, which serves as the input for the generation of the next round of real-time control commands.

[0014] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the artificial intelligence-based device operation control optimization method as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based device operation control optimization method as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by performing feature mapping and enhanced encoding on abnormal data, abnormal fluctuations and statistical features of multi-granularity time series, the operating status of equipment can be accurately characterized, abnormal information and time dependence can be preserved, providing high-quality input for the construction of continuous operating status feature vectors, and improving the accuracy of status prediction and real-time control optimization capabilities. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for an AI-based device operation control optimization method.

[0019] Figure 2 The flowcharts are for complete time series and multi-granularity time series.

[0020] Figure 3 This is a flowchart illustrating the enhanced multi-granularity time series and equipment operating status description.

[0021] Figure 4 A flowchart for updating the real-time control command sequence and equipment operating status description. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a device operation control optimization method based on artificial intelligence, including the following steps: S1: Collect multi-source heterogeneous time series data and organize them into an initial time series. Then, asynchronously complete the missing and outlier values ​​of the initial time series and remove low-confidence data to generate a complete time series.

[0026] S1.1: Organize and merge multi-source heterogeneous time series data in chronological order to obtain the initial time series.

[0027] Furthermore, the time series data from each source are sorted according to time labels, and it is determined whether there are duplicate or missing time points. For data with duplicate time points, the merging method is selected based on the statistical characteristics within the time window to generate a merged value for a single time point. For data from different sources that do not completely overlap at time points, the non-overlapping time points are filled in using historical data through linear interpolation to maintain the continuity of the sequence and verify time consistency, ensuring that the filled sequence meets the continuity requirements and is consistent with the original time order.

[0028] The merged repeating time points and the filled non-overlapping time points are arranged in chronological order to form a preliminary continuous time series. The series is then checked for continuity and temporal consistency to determine if there are still gaps that need to be filled or corrected. If the check fails, the filling and merging are repeated until the series meets the continuity requirements and the initial time series arranged in chronological order is obtained.

[0029] S1.2: Extract the missing and outlier values ​​identified in the initial time series, and use the characteristics of consecutive time points and historical complete time series to perform asynchronous completion on the missing and outlier values, generating continuous and complete data segments.

[0030] Furthermore, based on the chronological order, consecutive time points before and after missing values ​​and outliers are compared. At the same time, the mean, variance, and continuous data of the corresponding time periods of the historical complete time series are referenced to determine the time window type of each missing or outlier, including short-term missing, long-term missing, mild outlier, and strong outlier. For short-term missing windows, linear interpolation of the time points before and after is used. For long-term missing windows, multi-order interpolation is used and adjusted in combination with historical trend characteristics. At the same time, confidence is passed for verification. For mild outlier windows, a correction value is calculated based on the deviation before and after and used as the encoding input. For strong outlier windows, deviation correction is performed and potential outlier features are marked. The marking and correction results are used as verification and feedback conditions.

[0031] The completed or corrected data undergoes sequential continuity normalization, including verification with surrounding data, chronological alignment, and local statistical consistency adjustments. Continuity and statistical feature checks are then performed, comparing consecutive points with historical features to determine if the data segment meets integrity requirements. The check result serves as a feedback correction trigger; data segments that do not meet the requirements have their interpolation or correction parameters adjusted and are reprocessed. Data segments that meet the requirements are marked as valid and added to the output sequence. After processing all time windows, the results are integrated chronologically to generate a continuous and complete data segment.

[0032] Specifically, short-term missing data refers to short consecutive missing segments that can be filled in by linear interpolation relative to consecutive time points before and after them. Long-term missing data refers to long consecutive missing segments that can be filled in by multi-order interpolation relative to consecutive time points before and after them and the trend characteristics of the historical complete time series. Mild anomalies refer to observations that deviate from historical statistical characteristics, but the degree of deviation is not large. Deviation correction can be performed and the data can be used as encoding input. Strong anomalies refer to observations that deviate significantly from historical statistical characteristics. These should be marked as potential anomalies and feedback correction should be triggered.

[0033] S1.3: Analyze the confidence level of continuous and complete data segments, remove low-confidence data, and reassemble them to generate a complete time series.

[0034] Furthermore, each data point or data segment is compared with consecutive time points before and after it to analyze the statistical characteristics, trend patterns, and continuity matching degree with the corresponding time period in the historical complete time series. At the same time, the abnormal fluctuation amplitude is calculated to form a preliminary confidence index. The preliminary confidence index (which is determined by comparing the statistical characteristics and trend patterns of each data point or data segment with consecutive time points before and after it and the corresponding time period in the historical complete time series, and combining it with the abnormal fluctuation amplitude to determine the credibility of the data segment) is jointly judged with the continuity consistency to identify abnormal data segments with low confidence, that is, data segments that deviate from historical continuous data, have large fluctuation amplitudes, or are unstable in continuity in the time series. The remaining high-confidence data segments are combined and time-aligned according to the time order. By comparing the continuity and statistical characteristics of data segments in adjacent segments, the splicing gaps and fluctuations are regulated to generate a complete time series.

[0035] It should be noted that to improve the closed-loop processing of low-confidence data, when assessing the confidence of continuous and complete data segments and removing low-confidence data, the scope of confidence calculation indicators, the handling of low confidence and new gaps generated after removal should be clearly defined. For each removed data segment, it will be completed based on consecutive time points before and after and the characteristics of historical complete time series, forming a closed-loop processing. The completion method includes multi-order interpolation, combined with abnormal fluctuation correction and confidence feedback, to ensure that the generated complete time series maintains continuity and statistical consistency. By forming a closed loop between removal and completion, it is possible to avoid new missing data caused by the removal process, while ensuring the accuracy and completeness of subsequent multi-granularity time series construction and continuous running state feature vector generation.

[0036] S2: Divide the time series into time windows based on the complete time series, and dynamically resample the data in each time window to generate a multi-granularity time series.

[0037] S2.1: Divide the complete time series into time windows according to the second, minute and hour levels to obtain a preliminary set of time windows.

[0038] Furthermore, the complete time series is scanned in chronological order, and the data is divided into continuous second-level time intervals based on timestamps to form a set of second-level time windows. Based on the second-level time windows, the data for every consecutive number of seconds is statistically summarized to generate a set of minute-level time windows. This ensures that the data within each minute-level time window can reflect the overall change characteristics of the continuous time period. The data for several consecutive minutes is further summarized to obtain a set of hour-level time windows. Each hour-level time window contains continuous data from the corresponding minute-level time window, forming a preliminary set of time windows.

[0039] S2.2: Perform missing value completion and outlier identification on the data in the initial time window set, extract the continuous data sequence of each window and resample it to obtain a continuous data set within each time window.

[0040] Furthermore, the continuous data sequence within each time window is evaluated to determine whether there are missing values ​​or abnormal fluctuations in the data points, and the evaluation results are used as inputs for processing paths and parameter settings.

[0041] For missing values, continuous data is generated by linear interpolation based on the statistical characteristics of consecutive time points before and after the missing values ​​and the complete historical time series. For outliers, data points with significant deviations are marked by comparing the local mean and variance of the continuous data series within the time window with the abnormal fluctuation characteristics of the complete historical time series. At the same time, the outlier type and magnitude are recorded and passed to the joint integration.

[0042] During the resampling process of the completed continuous data sequence and the marked outliers, the data within each time window are arranged at a uniform time interval, and each outlier is associated with the continuous data sequence of the corresponding time period to form a preliminary continuous sequence. Continuity and integrity are judged for each time window, including whether missing values ​​have been continuously filled through completion and whether the outlier markings are consistent with the actual deviation magnitude and trend. For missing values ​​or outlier markings that fail the judgment, completion or outlier identification is re-executed, and the resampling parameters or the local time series arrangement order is adjusted according to the judgment results. The continuous sequence after completion and resampling is compared again to check the time series order, the accuracy of the outlier markings, and the local continuity. If the requirements are still not met, a feedback correction phase is triggered, and completion, outlier identification, and resampling are iteratively executed until all judgment conditions are met. After the verification and feedback correction are completed, a continuous data set within each time window is generated.

[0043] It should be noted that the equipment sampling cycle should be based on the actual sampling frequency of the equipment sensors to maintain the original temporal characteristics of the data. The control command cycle should be combined with the time interval of the commands issued by the control center to ensure that the resampled data can be directly used for control analysis. When there is missing or abnormal data in the historical data density, the time interval should be dynamically adjusted according to the data distribution of the complete historical time series to maintain continuity and statistical consistency.

[0044] During the iterative process of filling in missing values, identifying anomalies, and resampling data for each time window, the algorithm will perform cyclic adjustments when the judgment conditions are not fully met, until the convergence condition is reached or the maximum number of iterations is reached.

[0045] For example, for high-speed sensor data, each time window may contain hundreds of sampling points. The algorithm can be set to a maximum of 8 iterations. Each data sequence will undergo a maximum of 8 rounds of completion and anomaly correction. If there are still unprocessed missing or abnormal segments after the iteration, the continuous abnormal data will be marked and output as a partially complete time series with the label. At the same time, the abnormal information is retained for the generation and feedback correction of the continuous running state feature vector.

[0046] If there are still unprocessed missing or abnormal data segments during the iteration process, the persistent abnormal segments are marked and the partially complete time series with the markings is output. At the same time, the abnormal information is retained, and the continuous running state feature vector is generated and feedback is corrected.

[0047] S2.3: Based on the dataset, the continuous data sequences within each time window are combined and integrated according to different time scales to generate multi-granularity time series.

[0048] Furthermore, the continuous data sequence within each time window is divided into seconds, minutes, and hours, and the data segments at different time scales are judged. By comparing the time interval and fluctuation amplitude of adjacent data points, short time gaps, long time gaps, or abnormal fluctuations are determined, and the judgment results are used as processing paths and interpolation parameters.

[0049] For data segments with short time gaps, linear interpolation is used for consecutive time points before and after the data segment and then passed to the local continuity check. For data segments with long time gaps or large fluctuations, the mean, variance and fluctuation characteristics of the historical complete time series are combined to make inferences, and the confidence level is output for verification. Data segments with abnormal fluctuations that do not exceed the historical range are only fine-tuned and marked as enhanced candidate features.

[0050] Data segments at various time scales are aggregated to extract continuous data features, including mean, variance, maximum, minimum, abnormal fluctuation features, and continuity indicators. These features are then integrated based on the judgment results to ensure continuity in sequence, fluctuation amplitude, and statistical consistency across time scales. A continuity check is performed on the integrated multi-time scale series to determine whether the series meets the multi-granularity consistency requirements. If not, feedback is triggered to adjust parameters or complete the results until continuity is achieved. Second-level, minute-level, and hour-level data series are combined in time series order to generate a multi-granularity time series.

[0051] S3: Based on the historical data of the complete time series and the statistical characteristics of the time windows, feature mapping and enhanced coding are performed on the abnormal data, abnormal fluctuations and statistical characteristics of each time window in the multi-granularity time series to obtain the enhanced multi-granularity time series.

[0052] S3.1: Extract abnormal data and abnormal fluctuations from each time window in the multi-granularity time series, and perform preliminary feature processing with statistical features to generate a preliminary feature set.

[0053] Furthermore, the statistical characteristics of the continuous data sequence within each time window are compared with those of the preceding and following consecutive time points and the corresponding time periods of the historical complete time series, including local mean, variance, maximum value, minimum value and trend characteristics. The statistical indicators reflect the normal fluctuation range and trend pattern of the data, and determine whether the current observation point deviates from the historical pattern.

[0054] The system structurally combines various types of abnormal data, abnormal fluctuations, and statistical features according to time series order and dependency patterns. It then calls feature regularization to correct continuity and temporal consistency. The system performs continuity and temporal consistency checks on the preliminary feature set after combination. By comparing the time series relationship between abnormal features and normal features and historical statistical features, it determines whether there are abnormal deviations or potential abnormal patterns. The check results trigger feedback correction. If the check fails, the abnormal feature values, parameters, or combination order are adjusted, and the structured combination is re-executed to ensure that the features of each time window meet the continuity and consistency requirements. After the feedback correction is completed, the system outputs the preliminary feature set for each time window.

[0055] The formula for mapping anomaly features within a time window is: ; in, Indicates the first The first time window Preliminary mapping feature values ​​between anomalous data or anomalous fluctuations and statistical characteristics are used for feature grouping, recombination, and mapping to a continuous state space. Indicates the first The first time window An anomalous data point or an anomalous fluctuation observation is extracted from a continuous and complete data segment. This represents the corresponding time period in a complete historical time series. The mean of each feature is used to assess whether the current outlier deviates from the historical statistical level. This represents the corresponding time period in a complete historical time series. The standard deviation of each feature is used to standardize the degree of deviation. Indicates the first The first time window The anomalous data points or the local variation amplitude of anomalous fluctuations are used to reflect the characteristics of short-term anomalous fluctuations, and are obtained by comparing them with consecutive time points before and after.

[0056] It should be noted that the time window anomaly feature mapping formula reflects the comparison and preliminary feature mapping process of abnormal data and statistical features in continuous time series, and determines the output. It can be directly used as input for grouping, reorganizing, and mapping to a continuous state space.

[0057] In the time window anomaly feature mapping formula, to ensure consistency of various dimensions, the local variation amplitude is... According to the historical standard deviation Normalization is performed using the same scale so that the parts of the formula can be directly added together while maintaining the dimensionless property.

[0058] At the same time, when historical standard deviation When the denominator is zero, a small constant is used for numerical substitution during the calculation process to avoid zero denominator cases, ensure stable calculation under all time windows, and maintain the correctness of dimensions and numerical processing.

[0059] Historical fluctuation range refers to the allowable range of change for continuous data within a certain feature dimension or time window, determined based on the statistical characteristics of the corresponding time period in the complete historical time series, to determine whether the current observed data is abnormal.

[0060] The historical fluctuation range consists of three parts. First, by analyzing the mean and variance of data within a historical time period, the mean deviation range is determined to reflect the overall fluctuation trend of the data. Second, by comparing the change range of continuous historical data points, the local continuous change range is obtained to identify short-term abnormal fluctuations. Third, by referring to the maximum and minimum values ​​of data within a historical time period, an extreme value fluctuation reference is formed to determine abnormal peaks or troughs.

[0061] During the determination process, the feature value of each current time window is compared with the historical fluctuation range. The feature value is compared with the mean deviation range. If it falls within the range and the continuous change amplitude does not exceed the historical continuous change range, it is determined to be a normal feature or a fine-tunable enhanced candidate feature. If it exceeds the mean deviation range or the local fluctuation exceeds the historical extreme value, it is determined to be an abnormal feature, triggering interpolation, fine-tuning and resampling.

[0062] The historical fluctuation range consists of three parts: mean deviation, local continuous changes, and extreme value fluctuations. In anomaly detection, mean deviation is calculated by determining the deviation between the current time window's characteristic value and the mean of the corresponding historical time period, thus obtaining the mean deviation interval.

[0063] For local continuous changes, calculate the maximum and minimum values ​​of the continuous change amplitude at adjacent time points to obtain the local fluctuation range.

[0064] Extreme value fluctuations are determined by referencing the maximum and minimum values ​​for the corresponding historical time periods to obtain the extreme value fluctuation range.

[0065] The three intervals are merged element by element (taking the minimum lower bound and maximum upper bound of each point) to form the final historical fluctuation range. This range is then directly compared with the current observed value to obtain the comparison range (mean deviation lower bound, local continuous lower bound, minimum extreme value) and (mean deviation upper bound, local continuous upper bound, maximum extreme value).

[0066] S3.2: Based on the preliminary feature set, conduct correlation analysis on abnormal data and abnormal fluctuations to identify feature dimensions that are related to statistical features.

[0067] Furthermore, anomalous data and anomalous fluctuation features are extracted from each time window in the multi-granularity time series, and statistical features of the corresponding time windows are extracted. The anomalous data and anomalous fluctuation features are compared with statistical features of the same time window or similar time scale in the historical complete time series. Through similarity measurement, the correlation features between each anomalous feature and statistical feature are generated.

[0068] The correlation features are determined, and feature dimensions whose similarity measure exceeds the historical fluctuation range are marked as significantly correlated feature dimensions. Low-correlation feature dimensions are eliminated. After determination and screening, abnormal data and abnormal fluctuation feature dimensions that are significantly correlated with statistical features are identified. The correlation feature dimensions are used as inputs for grouping, reorganizing and mapping abnormal data, abnormal fluctuations and statistical features to continuous state space.

[0069] It should be noted that when performing correlation analysis on abnormal data and statistical features, the similarity measure and the historical fluctuation range have different dimensions. Therefore, when determining the dimension of significant correlation features, the similarity measure is normalized to be consistent with the dimension of the historical fluctuation range. After calculating the similarity or distance between each abnormal feature and the statistical feature, the similarity index is standardized according to the fluctuation range of the corresponding time period of the complete historical time series. It is then directly compared with the historical fluctuation range. Only when the standardized similarity measure exceeds the corresponding historical fluctuation range is the feature dimension marked as a significant correlation feature dimension, ensuring that the dimensions are consistent in the determination process.

[0070] S3.3: Using the results of correlation analysis, abnormal data, abnormal fluctuations and statistical characteristics are grouped and reorganized, and each group of features is mapped to a continuous state space to generate an intermediate feature set for state description, while determining whether there are potential abnormal features.

[0071] Furthermore, based on the association analysis, the dimensions of significant association features are determined, the association and dependency levels of each feature are judged, and the judgment results are used as the processing path and combination parameters for grouping, recombination and mapping.

[0072] Features are divided into different groups. For features with high correlation within a group, linear combination and integration are performed, and the combination results and parameters are passed to the mapping. For potentially anomalous feature groups, deviation information is recorded during integration and passed to the verification. For low-correlation feature groups, only structured arrangement is performed to maintain the time series order and passed to the mapping. After the recombined features of each group are mapped to the continuous state space, they are compared with the historical continuous running state feature vectors. By calculating the degree of deviation between the fluctuation patterns of features within the group and historical features, the deviation is used as the triggering condition for potential anomalous features and the mapping adjustment parameter.

[0073] If the deviation of a feature within a continuous time period exceeds the historical fluctuation range or shows an abnormal trend, it is marked as a potential abnormal feature and passed to feedback correction. If it does not exceed the historical fluctuation range, it remains in normal status and is retained in the intermediate feature set. For the marked potential abnormal feature group, the combination and mapping parameters are adjusted according to the verification results, and the combination and mapping are re-executed until the deviation of the features within the group is within a controllable range. After the feedback correction is completed, the mapping features of each group are integrated to form an intermediate feature set for state description, and potential abnormal features are marked at the same time.

[0074] It should be noted that when performing feedback correction on the intermediate feature set, the judgment of whether the features within the group deviate from the controllable range does not rely on fixed values, but is based on the dynamic calculation of historical complete time series, local continuous changes and statistical characteristics. The controllable range also takes into account the historical state of equipment operation, actual control parameters and process upper and lower limits as the basis for determining whether potential abnormal features have completed feedback correction. When the deviation of the features within the group is within the dynamic controllable range, it is marked as corrected. When it exceeds the controllable range, it is still retained as a potential abnormal feature and continues to be used for subsequent feedback correction.

[0075] S3.4: Enhance and integrate the information of the intermediate feature set, strengthen the encoding of the features of each time window, and generate an enhanced multi-granular time series.

[0076] Furthermore, feature mapping is performed on the abnormal data, abnormal fluctuations, and statistical characteristics contained in the state description of each time window, and they are arranged according to time order and correlation dimension to form a preliminary continuous sequence. By comparing with the historical complete time series and continuous running state feature vector, the degree of deviation of each feature in terms of mean, variance, maximum value, minimum value, and fluctuation pattern is determined.

[0077] Features that are significantly consistent with historical features are integrated, and the integration results and parameters are passed to the enhanced coding. Features that deviate significantly but can be corrected are marked as enhanced candidate features, and deviation information and adjustment coefficients are passed to the coding. Features that do not deviate significantly are only sorted and time series normalized. Enhanced coding is performed on the integrated or sorted features. The dependency relationship between abnormal features and normal features is mapped to the continuous state space through joint coding, and the combination order, coefficients and deviation information of each feature are passed to the local continuity check.

[0078] The system determines whether the feature sequences within the time window meet the requirements of continuity and consistency. Features and coefficients that fail the verification are marked as needing adjustment and passed to feedback correction. Features that pass the verification are marked as valid and enter the output. For features that need adjustment, the parameters and combination order are updated according to the verification results. The encoding and continuity verification are repeated until the sequence meets the requirements of continuity and statistical consistency. After the feedback correction is completed, the features of each time window are integrated in chronological order to form an enhanced multi-granular time series.

[0079] S4: Construct a continuous operating state feature vector based on the enhanced multi-granularity time series, and use the temporal attention method to jointly encode the feature vector to generate equipment operating state description and equipment operating state prediction value.

[0080] S4.1: Based on the enhanced multi-granularity time series, feature extraction and information integration are performed on the abnormal data, abnormal fluctuations and statistical characteristics of each time window to obtain a continuous operating state feature set.

[0081] Furthermore, based on the enhanced multi-granularity time series, the abnormal data, abnormal fluctuations, and statistical characteristics of each time window are analyzed. By extracting the outliers and fluctuation ranges of the continuous data series in each time window and combining them with the statistical characteristics within the time window, preliminary feature processing is performed to obtain the feature set of each time window.

[0082] By integrating the interrelationships between abnormal data and abnormal fluctuations in time series and their association with statistical features, and by mapping them to a continuous state space for feature fusion and information integration, a set of continuous operating state features is formed.

[0083] Specifically, time-dependent analysis refers to comparing the abnormal data, abnormal fluctuations, and statistical characteristics of each time window in the continuous operating state feature vector with the characteristics of the preceding and following windows in chronological order. It analyzes the changing trend, fluctuation amplitude, and persistence of the characteristics over time. When the direction or amplitude of a certain feature changes significantly deviates from the historical trend, or when the deviation continues for more than a certain period of time, it is judged as an abnormal feature. If the deviation amplitude is small but shows an abnormal trend change, it is judged as a potential abnormal feature.

[0084] S4.2: For time series features in the continuous operating state feature set, determine the degree of deviation of each feature from the historical and prediction range, and identify abnormal and potential abnormal features.

[0085] Furthermore, based on the continuous operating state feature vector constructed from the enhanced multi-granularity time series, a comparative analysis of the feature values ​​of abnormal data, abnormal fluctuations, and statistical characteristics in each time window is performed. The historical data segment and prediction time period corresponding to each time series feature in the continuous operating state feature vector are extracted. By calculating the degree of difference and fluctuation consistency of feature values ​​in the historical time period and the prediction time period, the magnitude of deviation of each time series feature from the historical and prediction ranges is determined. Features with significant deviations are identified as abnormal features, and features with gradually expanding deviation trends but not yet exceeding the historical and prediction ranges are identified as potential abnormal features, thus forming the abnormal and potential abnormal identification of each time series feature in the continuous operating state feature set.

[0086] S4.3: Based on the judgment results, the time series features are jointly encoded and combined, and the abnormal features are integrated with the normal features through dependency association to generate a continuous running state feature vector.

[0087] Furthermore, feature mapping and enhanced encoding are performed on the abnormal data, abnormal fluctuations and statistical features of each time window to obtain the representation vector of each time window. By comparing the statistical features with those of consecutive time windows before and after and the complete historical time series, the degree of significant correlation between abnormal features and normal features in terms of time dependence is determined. The determination result is used as the processing path and parameter input for joint encoding.

[0088] For anomalous features with significant time dependencies, joint combination processing is performed to integrate them with corresponding normal features into joint representation features, and the parameters are passed to the encoding integration. For anomalous features with large deviations but which can be corrected, they are marked as enhanced candidate features, and the deviation index and adjustment parameters are passed to the joint encoding. For features without significant deviations, they are simply sorted in order while maintaining the time dependency order, and the sorting result is used as the input of the joint encoding.

[0089] In the local continuity and time dependency consistency verification stage, by comparing feature order, fluctuation pattern and statistical consistency, it is determined whether the joint representation vector meets the continuous operation state requirements. Features and parameters that fail the verification are marked as needing adjustment and passed to feedback correction. Features that pass the verification are marked as valid and enter the continuous operation state feature vector output. Based on the verification results, the parameters and combination order are updated, and the joint encoding and continuity verification are repeated until all time window features meet the continuity and time dependency consistency requirements. After the feedback correction is completed, the joint representation vector of each time window is integrated in time order to generate the continuous operation state feature vector.

[0090] It should be noted that the temporal attention method is used to jointly encode the abnormal data, abnormal fluctuations and statistical features of each time window in the continuous operating state feature vector. By analyzing the dependencies and importance of features in the time series, it generates equipment operating state descriptions and predicted values.

[0091] Time-dependent information is extracted from the continuous feature vector of each time window. This involves comparing the features of each time window with those of the preceding and following windows to determine the importance of each feature in the overall sequence. Based on the determination results, abnormal and normal features are mapped and combined so that abnormal features that deviate from the historical fluctuation range have a higher representation in the joint representation vector, while preserving the continuity and trend of normal features. The joint features of each time window are sequentially encoded and enhanced and integrated. By comparing with the complete historical time series and the enhanced multi-granular time series, potential abnormal features and abnormal fluctuations are corrected to ensure that the continuous operating state feature vector can accurately reflect the operating state of the equipment.

[0092] For input and feature construction, the continuous feature vector for each time window consists of outlier data points, outlier fluctuation amplitudes, and statistical features (mean, variance, and extreme values). The number of feature dimensions is related to the number of sampling points and statistical indicators within the window. The data in each time window are arranged in chronological order to form an input sequence for attention encoding.

[0093] The attention model structure and temporal attention method jointly encode the features of each time window through a self-attention mechanism. The model can capture the dependencies between different time windows and integrate the importance of abnormal and normal features in the time series. The result of joint encoding generates a high-dimensional representation vector for each time window.

[0094] Time-dependent analysis and feature integration compare the differences and trends of features before and after each time window, determine the importance of each feature in the overall sequence, give higher weight to anomalous features, make anomalous features that deviate from the historical fluctuation range more significant in the joint representation, while maintaining the continuity and change trend of normal features. The joint features of each time window are integrated and strengthened in chronological order to form a continuous running state feature vector.

[0095] Training methods and loss functions: Model training uses historical device operating data, including normal and abnormal states.

[0096] The loss function consists of two parts: one part measures the state prediction error and the other part measures the accuracy of anomaly feature recognition. They are combined to optimize prediction performance and anomaly detection capability simultaneously.

[0097] Training termination conditions include the loss function converging or reaching the maximum number of training epochs, ensuring stable model training.

[0098] The rule-based encoding method (no training required) dynamically calculates attention weights based on statistical deviations, anomaly frequencies, or local fluctuation amplitudes of features when no training is performed. The more significant the deviation of anomaly features, the higher their weight, thereby strengthening the role of potential anomaly features in the joint representation.

[0099] Output and mapping: The model outputs a continuous running state feature vector, and the vector for each time window reflects the joint feature representation of that window.

[0100] The continuous operation status feature vector is mapped to describe the device's operating status, providing usable information for monitoring. At the same time, the feature vector is mapped to predictive values ​​of the device's operating status, providing predictive information for various operating parameters in future time windows.

[0101] Joint encoding ensures that the continuity of abnormal and normal features is preserved in the output, providing reliable input descriptions and predictions for the generation of real-time control commands.

[0102] S4.4: Extract data from each time period in the continuous operation state feature vector, extract and integrate time-dependent abnormal data, abnormal fluctuations and statistical features to obtain the joint features of each time period.

[0103] Furthermore, the abnormal data, abnormal fluctuations, and statistical features within each time period are sequentially arranged according to time order. The correlation between different time points is identified through time dependency analysis. The sequentially arranged abnormal data, abnormal fluctuations, and statistical features are jointly integrated. Using the statistical features of the corresponding time windows in the historical data and the enhanced multi-granularity time series, the features of each time period are correlated, mapped, and encoded to form the joint features of each time period.

[0104] S4.5: Perform time-series dependency analysis and correlation extraction on the joint features of each time period, determine whether the abnormal features of each time period deviate from the historical and prediction range, and generate a complete description of the equipment operating status.

[0105] Furthermore, the abnormal data, abnormal fluctuations, and statistical features of each time period in the continuous operation status feature vector are compared with the historical complete time series and the enhanced multi-granular time series to determine whether the features are within the historical or predicted range.

[0106] Features are categorized into three types: those with insignificant deviations are marked as normal features and directly passed to joint integration; those with significant but correctable deviations are marked as potential anomalous features, and the deviation magnitude and correction coefficient are passed to association extraction and integration; and those with deviations exceeding the range are marked as anomalous features. Deviation indicators are used as verification and fusion trigger conditions to verify the continuity, dependency, and degree of deviation of features. Features and parameters that fail the verification are passed to feedback correction; those that pass the verification are marked as valid and enter the state description integration. The feature representation vector and combination parameters are updated according to the verification results. The temporal dependency analysis and integration are repeated until the feature deviation, dependency, and continuity of all time periods meet the requirements, generating a complete description of the equipment's operating status.

[0107] It should be noted that the correction coefficient is used to adjust abnormal or deviated features to ensure the continuity and controllability of the feature vector and predicted value of the continuous operation state. The calculation is based on three quantities: the deviation of the current time window feature from the mean and standard deviation of the historical complete time series, the local change amplitude of consecutive time points before and after, and the deviation degree of the feature itself.

[0108] The specific operation involves standardizing historical deviations and local continuous change amplitudes to a uniform scale and weighting them with the degree of feature deviation to obtain preliminary correction coefficients. These coefficients are dynamically generated over time windows and updated in real-time feedback to correct current features while retaining anomaly indicators, thus achieving closed-loop adjustment. No fixed values ​​are required; the coefficients can be adjusted according to changes in equipment type, operating parameters, or time window characteristics, reflecting the operational sensitivity and anomaly impact of different equipment or parameters. By referencing historical data and local features, the correction coefficients can be used to amplify or weaken anomaly features during the transmission process, affecting the description of equipment operating status and the generation of predicted values.

[0109] S4.6: Use the judgment results to correct the predicted value of each operating parameter in the future time window, generate the predicted value of the equipment operating status, and identify the operating parameters that deviate.

[0110] Furthermore, the joint features of each operating parameter in the continuous operating state feature vector corresponding to the time period are extracted, and combined with the abnormal features and potential abnormal features in the equipment operating state description, it is determined whether the original predicted value deviates from the historical or predicted range.

[0111] Operating parameters are categorized into three types: parameters with no or slight deviations retain their original predicted values ​​and are integrated into the predicted value set; parameters with significant but correctable deviations undergo time-dependent extraction and joint integration based on abnormal data, abnormal fluctuations, and statistical characteristics, calculating correction amounts and integrating them; parameters deviating beyond the correctable range are marked as abnormal predicted parameters, and their deviation magnitude and correction status are transmitted to feedback correction, while deviation flags are enabled for output. During joint integration, adjustment strategies are applied to ensure that the corrected predicted values ​​account for both the impact of abnormal characteristics and the continuity of state within the time window. Continuity and consistency checks are performed on the corrected predicted values. Parameters that fail the check trigger feedback correction and update their correction amounts; those that pass the check are marked as valid and retain their deviation flags. After iterative adjustment, the corrected predicted values ​​and deviation flags are integrated to generate complete equipment operating status predicted values.

[0112] The formula for correcting the predicted value of the eigenvector of the continuous operating state is: ; in, Indicates the first A set of operating parameters in the future time window The internal correction prediction values ​​are used to generate the next round of real-time control commands. Indicates the first Each running parameter in the current time window The predicted value is generated from the continuous operating state feature vector and the predicted value of the equipment operating state. Indicates the first Each running parameter in the time window The joint feature values ​​are obtained by feature mapping and reinforcement coding from the abnormal data, abnormal fluctuations and statistical features in the enhanced multi-granularity time series and continuous operating state feature vector.

[0113] It should be noted that the formula for correcting the predicted value of the continuous operating state feature vector reflects the process of correcting the predicted value of each operating parameter in the continuous operating state feature vector. It directly uses the joint features to adjust the current predicted value, thereby realizing the linkage between judgment, abnormal feature extraction and predicted value correction.

[0114] For the correction formula of the predicted value of the eigenvector of the continuous running state, the joint eigenvalues It can be derived from abnormal data, abnormal fluctuations, and statistical features in enhanced multi-granularity time series and continuous operation state feature vectors. Before use, the values ​​will undergo feature mapping and normalization to ensure that the dimensions are consistent with the original predicted values. Matching.

[0115] Through this normalization process, the correction formula for the predicted value of the continuous running state feature vector is logically implemented. The correction effect is applied to the joint eigenvalues ​​without directly offsetting the current predicted value, while maintaining the consistency of the predicted value's dimensions.

[0116] S5: Calculate the optimized value of each operating parameter based on the equipment operating status description, generate a real-time control command sequence, and execute the equipment operation.

[0117] S5.1: Extract the current state value and trend of each operating parameter from the equipment operating status description and continuous operating status feature vector to obtain the corresponding optimization target range.

[0118] Furthermore, when extracting the current state value and trend of each operating parameter from the equipment operating status description and continuous operating status feature vector, feature extraction and information integration are performed on the data of each time window in the continuous operating status feature vector to obtain the change trajectory of each operating parameter in a continuous time period, including mean, variance, maximum value, minimum value and abnormal fluctuation information.

[0119] The trend of each operating parameter is analyzed based on the change trajectory, including the direction of increase or decrease, the fluctuation amplitude, and abnormal fluctuations. The analysis results are compared with the joint features in the equipment operating status description to determine the degree of deviation of the current status value from the historical pattern and multi-granular features. Based on the degree of deviation and the change trend, the optimization target range of each operating parameter is determined, including the target value range, the allowed change direction, and the fluctuation amplitude. The optimization target range is then used to generate a real-time control command sequence.

[0120] Ideally, the linkage between the continuous operating state feature vector, the equipment state description, and the generation of control commands is realized, so that the real-time control commands for each operating parameter are dynamically adjusted to meet the optimization objectives.

[0121] Specifically, the target range is optimized, including the target value range, the allowed direction of change and the fluctuation range. The deviation and trend of the continuous operating state feature vector are determined based on the historical operating data and the operating patterns observed in multi-granularity time series. The target range of the current operating condition is generated to guide the generation of real-time control commands and achieve control optimization consistent with the equipment operating status and historical data.

[0122] Using the optimization target range as the basis for generating real-time control command sequences means that the optimization target range is determined by the current state value, change trend, and historical and predictive information in the continuous operating state feature vector of each operating parameter. It includes the allowable numerical range, change direction, and fluctuation amplitude. This range serves as the basis for generating real-time control commands, guiding the numerical conversion, mapping, and reorganization of each parameter. This ensures that the generated control commands not only meet the optimization target but also maintain the continuity and constraint relationship between parameters, thereby achieving dynamic and controllable equipment operation adjustments.

[0123] S5.2: Based on the optimization target range and state description, the current state value is transformed and reorganized to generate a real-time control command sequence, and it is determined whether each command meets the target adjustment direction and amplitude requirements. If the command does not meet the requirements, adjustments are made.

[0124] Furthermore, the current state value and trend of each operating parameter are extracted and compared with the historical data in the continuous operating state feature vector and the statistical features of the enhanced multi-granularity time series to determine the deviation between the current state value and the optimization target range.

[0125] The operating parameters are divided into three categories: parameters with insignificant deviations are directly mapped to generate instructions and passed to the verification; parameters with significant but correctable deviations are aligned to the target range through numerical conversion and mapping, and the mapping coefficients and arrangement order are passed to the reorganization; parameters with deviations exceeding the controllable range are marked as control instructions that need to be corrected, and the deviation magnitude, mapping parameters, and correction trigger information are passed to the feedback correction.

[0126] The parameter mapping values ​​are arranged and combined according to the characteristics of continuous operation state to generate a preliminary real-time control command sequence. Each command is compared and analyzed. The command is judged to meet the requirements based on the target adjustment direction and magnitude. Commands that fail the verification trigger feedback correction. Commands that pass the verification are marked as valid and included in the output. Based on the continuous operation state feature vector and the enhanced multi-granularity time series, the parameters, adjustment magnitude and arrangement order are recalculated. The correction is iteratively executed until all commands meet the optimization target requirements. The complete real-time control command sequence is then generated by integrating them in chronological order.

[0127] For example, increasing one parameter may cause another parameter to exceed the safety limit. Potential conflicts are analyzed and handled by referring to historical operating data, process limitations and safety boundaries to determine the priority and allowable adjustment range of each parameter. Limitation and adjustment are performed when combining parameters to ensure that the combined real-time control commands not only meet the sequence of continuous operating state characteristics, but also take into account the constraint relationship between multiple parameters, so as to achieve coordinated control and safety optimization.

[0128] S5.3: Execute the real-time control command sequence to the device, determine the consistency between the device execution feedback and the command, and collect the device execution feedback information.

[0129] Furthermore, instructions are sent to the device execution nodes through the device interface, and the operating parameters are adjusted according to the instructions. At the same time, the actual operating status of each node is collected, including the operating parameter values ​​and control signal feedback. The actual values ​​are compared with the instruction values ​​one by one. For parameters that deviate from the allowable range, they are marked as consistent with the execution and included in the feedback information integration. For parameters that exceed the allowable deviation, they are marked as deviating from the execution, and the deviation magnitude, direction and duration are recorded to trigger feedback correction.

[0130] Integrate the consistency results of all operating parameters with the feedback information of each node to generate complete equipment execution feedback information. By comparing execution consistency and deviation characteristics, determine whether the feedback information meets the requirements of continuous operation. Parameters that fail the verification trigger feedback correction and readjustment or correction, while parameters that pass the verification are marked as valid.

[0131] Specifically, for parameters that deviate from the allowable range, the allowable deviation is determined by a combination of historical fluctuation characteristics in the continuous operating state feature vector, the type of each operating parameter and its corresponding process requirements, and the optimization target range.

[0132] Different parameters can be set with different allowable deviation ranges based on sensitivity and process constraints. When performing feedback judgment, it can identify significant deviations without misjudging normal fluctuations. The allowable deviation standard serves as the basis for judging the consistency of equipment execution, ensuring that the execution results of real-time control commands are consistent with the characteristics of continuous operation.

[0133] S6: Receive the operation results after receiving the real-time control command sequence of the equipment, compare and analyze the operation results with the predicted value of the equipment operation status, and update the equipment operation status description.

[0134] S6.1: Obtain the operating parameter values, sensor readings and control signals for each time period from the operating results after the real-time control command of the equipment, and extract the data sequence of state update by combining the continuous operating state feature vector.

[0135] Furthermore, the collected operating parameter values, sensor readings, and control signals for each time period are organized in chronological order and aligned with the continuous operating state features of the corresponding time period in the continuous operating state feature vector. The actual operating values ​​for each time period are compared with historical features and joint coding results. For parameters and signals that deviate from the historical fluctuation range or have significant abnormal fluctuations, the deviation magnitude, direction, and duration are recorded. For parameters and signals with normal fluctuations, the original feature values ​​are maintained.

[0136] The identified abnormal and normal features are combined according to time sequence and feature dimensions to generate a status update data sequence. The continuity, time dependency and feature completeness of the sequence are checked. If missing or abnormal markers are found that do not meet the requirements for continuous operation, feedback correction is triggered to correct or complete the abnormal features until a complete, continuous status update data sequence that can be used to determine operational differences and update the description of equipment operation status is generated.

[0137] S6.2: Compare the extracted data sequence with the predicted values ​​of equipment operating status, determine whether each operating parameter and fluctuation characteristic deviates from the prediction range, and generate corresponding operating difference information.

[0138] Furthermore, the operating parameter values ​​for each time period are matched one by one with the corresponding predicted values. The deviation magnitude, direction, and duration are calculated. Based on the judgment results, the operating parameters are divided into three categories: those with insignificant deviations are marked as normal and directly transmitted to the output merging; those with significant but correctable deviations are transmitted to joint analysis and correction; and those with deviations exceeding the controllable range are marked as abnormal difference parameters. The deviation information and correction trigger identifier are transmitted to feedback correction.

[0139] The deviation magnitude, direction, and duration are used to form structured operational difference information. The marked state is then transmitted as a chain control signal to the continuous operational state feature vector adjustment. By comparing the fluctuation characteristics of continuous time periods with the predicted value trend, the completeness and rationality of the difference information are determined. If the verification fails, feedback correction is triggered and the deviation parameters are updated. If the verification passes, it is marked as valid and included in the state vector. The analysis, structure generation, and feedback correction are repeated until the deviation of each operational parameter and fluctuation characteristic accurately reflects the predicted value deviation, and the complete operational difference information is generated.

[0140] S6.3: Utilize the operational difference information to adjust and update the parameter values, feature representation vectors, and feedback information in the continuous operation state feature vector, determine whether the updated state meets the continuous operation conditions, obtain the corrected equipment operation state description, and use it as the input for the generation of the next round of real-time control commands.

[0141] Furthermore, the operational discrepancy information is compared one by one with the parameter values, feature representation vectors, and feedback information of each time period in the continuous operational status feature vector to determine whether the feature in each time period deviates from the prediction range. Based on the determination results, the features are divided into three categories: features with insignificant deviations retain their original values ​​and are directly passed to the integration; features with significant but correctable deviations calculate correction amounts based on the operational discrepancy information, update parameter values, and pass the correction amounts and correction flags to the encoding integration; features with deviations exceeding the controllable range are marked as abnormal features, and the deviation magnitude, direction, and trigger correction flag are passed to the feedback correction.

[0142] The features of each time period are re-encoded. Abnormal fluctuations and statistical features are jointly encoded according to the judgment results. The adjusted parameter values, feature representation vectors and feedback information are integrated into a new continuous operation state feature vector. At the same time, the encoding order, adjustment parameters and deviation flags are passed to the continuity check. By comparing the feature order, fluctuation mode and statistical features, it is determined whether the adjusted state meets the continuous operation conditions. Features and parameters that fail the check are triggered for feedback correction. Features that pass the check are marked as valid and included in the state description output. The joint encoding and continuity check are repeated for features that need to be adjusted until all time period features meet the continuity and prediction consistency requirements. The features of each time period are integrated to generate a corrected equipment operation state description, which is used as the input for the next round of real-time control commands.

[0143] This embodiment also provides a computer device applicable to the device operation control optimization method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the device operation control optimization method based on artificial intelligence as proposed in the above embodiment.

[0144] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0145] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the device operation control optimization method based on artificial intelligence as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0146] In summary, this invention, by performing feature mapping and enhanced encoding on abnormal data, abnormal fluctuations, and statistical features of multi-granularity time series, can accurately characterize the operating status of equipment, retain abnormal information and time dependencies, provide high-quality input for the construction of continuous operating status feature vectors, and improve the accuracy of status prediction and real-time control optimization capabilities.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing equipment operation control based on artificial intelligence, characterized in that, include: Collect heterogeneous time series data from multiple sources and organize them into an initial time series. Then, perform asynchronous completion and removal of missing and outlier values ​​and low-confidence data in the initial time series to generate a complete time series. The data is divided into time windows based on the complete time series, and the data in each time window is dynamically resampled to generate a multi-granularity time series. Based on the historical data of the complete time series and the statistical characteristics of the time windows, feature mapping and enhanced coding are performed on the abnormal data, abnormal fluctuations and statistical characteristics of each time window in the multi-granularity time series to obtain the enhanced multi-granularity time series. A continuous operating state feature vector is constructed based on the enhanced multi-granularity time series, and the feature vector is jointly encoded using the temporal attention method to generate equipment operating state description and equipment operating state prediction value. Based on the description of the equipment's operating status, the optimized value of each operating parameter is calculated, a real-time control command sequence is generated, and the equipment operation is executed. After receiving the sequence of real-time control commands for equipment operation, the system compares and analyzes the results with the predicted values ​​of equipment operation status, and updates the equipment operation status description.

2. The AI-based equipment operation control optimization method as described in claim 1, characterized in that, The specific steps for generating the complete time series are as follows: Multi-source heterogeneous time series data are organized and merged in chronological order to obtain an initial time series; Extract the missing and outlier values ​​identified in the initial time series, and use the characteristics of consecutive time points and historical complete time series to perform asynchronous completion on the missing and outlier values, generating continuous and complete data segments; Confidence assessments are performed on continuous and complete data segments, low-confidence data are removed, and the data are recombined to generate a complete time series.

3. The AI-based equipment operation control optimization method as described in claim 1, characterized in that, The specific steps for generating multi-granularity time series are as follows: The complete time series is divided into time windows according to the second, minute and hour levels to obtain a preliminary set of time windows; Missing values ​​are filled and outliers are identified in the data in the initial time window set. The continuous data sequence of each window is extracted and resampled to obtain the continuous data set within each time window. Based on the dataset, continuous data sequences within each time window are combined and integrated according to different time scales to generate multi-granularity time series.

4. The AI-based equipment operation control optimization method as described in claim 3, characterized in that, The specific steps to obtain the enhanced multi-granularity time series are as follows: Extract outlier data and outlier fluctuations from each time window in the multi-granularity time series, and perform preliminary feature processing with statistical features to generate a preliminary feature set; Based on the preliminary feature set, correlation analysis is performed on abnormal data and abnormal fluctuations to identify feature dimensions that are associated with statistical features. Using the results of correlation analysis, abnormal data, abnormal fluctuations and statistical characteristics are grouped and reorganized, and each group of features is mapped to a continuous state space to generate an intermediate feature set for state description, while determining whether there are potential abnormal features. Information enhancement and integration are performed on the intermediate feature set, and the features of each time window are enhanced and encoded to generate an enhanced multi-granular time series.

5. The equipment operation control optimization method based on artificial intelligence as described in claim 1, characterized in that, The specific steps for constructing a continuous operating state feature vector based on the enhanced multi-granularity time series are as follows: Based on the enhanced multi-granularity time series, feature extraction and information integration are performed on the abnormal data, abnormal fluctuations and statistical characteristics of each time window to obtain a continuous operating state feature set; For time series features in a continuous operating state feature set, determine the degree of deviation of each feature from its historical and predicted range, and identify abnormal and potential abnormal features; Based on the judgment results, the time series features are jointly encoded and combined, and the abnormal features are integrated with the normal features through dependency association to generate a continuous running state feature vector.

6. The AI-based equipment operation control optimization method as described in claim 5, characterized in that, The specific steps for generating the device operating status description and the device operating status prediction value are as follows: Extract data from each time period in the feature vector of continuous operation status, and perform time-dependent extraction and joint integration of abnormal data, abnormal fluctuations and statistical features to obtain the joint features of each time period; Perform time-series dependency analysis and correlation extraction on the joint features of each time period, determine whether the abnormal features of each time period deviate from the historical and prediction range, and generate a complete description of the equipment operating status. The judgment results are used to correct the predicted values ​​of each operating parameter within a future time window, generate predicted values ​​of equipment operating status, and identify operating parameters that deviate from the predicted values.

7. The AI-based equipment operation control optimization method as described in claim 1, characterized in that, The specific steps for generating the real-time control command sequence and executing the equipment are as follows: Extract the current state value and trend of each operating parameter from the equipment operating status description and continuous operating status feature vector to obtain the corresponding optimization target range; Based on the optimization target range and state description, the current state value is transformed and reorganized to generate a real-time control command sequence. It is then determined whether each command meets the target adjustment direction and amplitude requirements, and adjustments are made to commands that do not meet the requirements. The system executes a real-time control command sequence to the device, determines the consistency between the device's execution feedback and the command, and collects the device's execution feedback information.

8. The AI-based equipment operation control optimization method as described in claim 7, characterized in that, The specific steps for updating the device operating status description are as follows: The system obtains the operating parameter values, sensor readings and control signals for each time period from the operating results after the real-time control command of the equipment, and extracts the data sequence of state update by combining the continuous operating state feature vector. The extracted data sequence is compared with the predicted values ​​of equipment operating status to determine whether each operating parameter and fluctuation characteristic deviates from the prediction range, and corresponding operating difference information is generated. By utilizing operational difference information, the parameter values, feature representation vectors, and feedback information in the continuous operation state feature vector are adjusted and updated. It is then determined whether the updated state meets the continuous operation conditions, and a corrected description of the equipment operation state is obtained, which serves as the input for the generation of the next round of real-time control commands.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the artificial intelligence-based device operation control optimization method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based device operation control optimization method according to any one of claims 1 to 8.

Images