A device running modeling method, an electronic device, and a storage medium

Abstract

Embodiments of the present application disclose a device operation modeling method, an electronic device and a storage medium. The method comprises: obtaining and acquiring a plurality of state evolution sample pairs based on target item working condition parameters collected in real time by a target device in a historical operation period, the target item working condition parameters comprising state parameters and control parameters; constructing a preliminary dynamic model of the target device based on the plurality of state evolution sample pairs by using a Kuppmann-extended dynamic modal decomposition algorithm; performing reliability verification on the preliminary dynamic model based on reconstruction errors of the preliminary dynamic model; and when the preliminary dynamic model fails to pass the reliability verification, determining optimized state evolution sample pairs based on the reconstruction errors, and updating the preliminary dynamic model based on the optimized state evolution sample pairs to obtain an operation dynamic model of the target device. Embodiments of the present application can construct an operation dynamic model of a device with self-diagnosis and self-healing capabilities.

Description

Technical Field

[0001] This invention relates to the field of automatic equipment control technology, and in particular to a method for modeling equipment operation, an electronic device, and a storage medium. Background Technology

[0002] In the fields of industrial automation and intelligent control (such as model predictive control and adaptive control), the core challenge faced by engineers stems from the strong nonlinear characteristics of physical systems. In typical scenarios such as chemical reactions, semiconductor manufacturing, and tobacco drying, variables such as temperature, pressure, and flow rate are dynamically complex and difficult to describe accurately using traditional mathematical models.

[0003] To address this challenge, the industry has introduced data-driven modeling techniques such as the Koopman operator. The Koopman operator maps state variables to a high-dimensional feature space, transforming the evolution of a nonlinear system into a globally linear form. This allows for the establishment of linear matrix models in high-dimensional space, enabling the direct deployment of mature linear control strategies. However, existing control schemes based on the Koopman operator exhibit significant shortcomings in practical industrial deployments. Traditional methods employ offline training, where the model is permanently installed and operational once deployed. When physical anomalies occur in the production environment, such as material blockages or actuator jamming, the model lacks self-diagnosis and self-healing capabilities, leading to a significant decline or even instability in automated control performance. Summary of the Invention

[0004] This invention provides a device operation modeling method, an electronic device, and a storage medium, which can construct a dynamic device operation model with self-diagnosis and self-healing capabilities.

[0005] In a first aspect, embodiments of the present invention provide a device operation modeling method, comprising: Multiple state evolution sample pairs are obtained based on the target item operating condition parameters collected in real time during the historical running period of the target device. The target item operating condition parameters include state parameters and control parameters. Based on the multiple state evolution sample pairs, a preliminary dynamic model of the target device is constructed using the Koopman-Extended Dynamic Mode Decomposition algorithm; The reliability of the preliminary dynamic model is verified by acquiring and using the reconstruction error of the preliminary dynamic model. If the preliminary dynamic model fails the reliability verification, an optimized state evolution sample pair is determined based on the reconstruction error, and the preliminary dynamic model is updated based on the optimized state evolution sample pair to obtain the operating dynamic model of the target device.

[0006] Secondly, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the device operation modeling method as described in any of the embodiments of the present invention.

[0007] Thirdly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the device operation modeling method as described in any of the embodiments of the present invention.

[0008] This invention provides a device operation modeling method, electronic device, and storage medium. It acquires multiple state evolution sample pairs based on target item operating parameters collected in real-time during historical operation of the target device. A preliminary dynamic model of the target device is constructed using the Koopman-Extended Dynamic Mode Decomposition algorithm based on these sample pairs. The reliability of the preliminary dynamic model is further verified based on its reconstruction error. If the preliminary dynamic model fails the reliability verification, an optimized state evolution sample pair is determined based on the reconstruction error. The preliminary dynamic model is then updated based on the optimized state evolution sample pair to obtain the operating dynamic model of the target device. This method can construct a device operating dynamic model with self-diagnosis and self-healing capabilities, thereby significantly improving the control performance of automated control and enhancing the stability of the automated control system. Attached Figure Description

[0009] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the device operation modeling method provided in an embodiment of the present invention; Figure 2 This is another flowchart illustrating the device operation modeling method provided in this embodiment of the invention; Figure 3 This is another flowchart illustrating the device operation modeling method provided in this embodiment of the invention; Figure 4 This is another flowchart illustrating the device operation modeling method provided in this embodiment of the invention; Figure 5 This is another flowchart illustrating the device operation modeling method provided in this embodiment of the invention; Figure 6This is another flowchart illustrating the device operation modeling method provided in this embodiment of the invention; Figure 7 This is a schematic diagram illustrating the prediction of the moisture content of the exit tobacco shreds using the dynamic model of the tobacco drying equipment provided by the equipment operation modeling method in this embodiment of the invention. Figure 8 This is another flowchart illustrating the performance of the tobacco drying equipment's dynamic model in predicting the moisture content of the exit tobacco shreds, provided by the equipment operation modeling method provided in this embodiment of the invention. Figure 9 This is a schematic diagram of the device operation modeling apparatus provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0012] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0013] Figure 1 This is a flowchart illustrating a device operation modeling method provided in an embodiment of the present invention. This embodiment is applicable to the operation modeling of chemical reaction equipment, semiconductor manufacturing equipment, and tobacco drying equipment. The method can be executed by the device operation modeling device provided in this embodiment, which can be implemented using software and / or hardware. In a specific embodiment, the device can be integrated into an electronic device, such as a computer or server. The following embodiments will illustrate this using the integration of the device into an electronic device as an example. (Reference) Figure 1The method may specifically include the following steps: Step 101: Obtain and acquire multiple state evolution sample pairs based on the target item operating condition parameters collected in real time during the historical operation period of the target device. The target item operating condition parameters include state parameters and control parameters. This step facilitates the construction of a preliminary dynamic model of the target device based on multiple state evolution sample pairs.

[0014] Specifically, the aforementioned target equipment can be material processing equipment.

[0015] In a specific instance, the aforementioned target equipment is a tobacco drying device.

[0016] Specifically, the aforementioned historical runtime segment can be any historical runtime segment with a preset duration.

[0017] Specifically, the aforementioned target item operating condition parameters may include all or part of the operating condition parameters of the target equipment.

[0018] Specifically, the aforementioned state parameters may include one or more state parameters, and the aforementioned control parameters may include one or more control parameters.

[0019] Specifically, the aforementioned multiple state parameters may include: moisture content of the exported tobacco, temperature of the exported tobacco, hot air temperature, ventilation pressure of the drying duct, and / or flow rate of the drying hot air.

[0020] Specifically, the above control parameters may include: water valve opening, tobacco leaf re-moistening water supply, heating steam valve opening, drying damper opening, and / or fan operating frequency.

[0021] Specifically, the aforementioned state evolution sample pairs can be understood as being based on the operating conditions of the equipment at different times. The historical operating condition sample of the previous moment is paired with the real-time operating condition sample that follows the evolution of the next moment to form a paired dataset that can reflect the continuous change of the equipment's operating conditions over time. This dataset is used to characterize the correspondence between the dynamic migration and step-by-step evolution of the system's operating conditions from the previous state to the subsequent state.

[0022] Specifically, the process of obtaining multiple state evolution sample pairs based on the target item operating condition parameters collected in real time during the historical operation period of the target device may include: obtaining the original target item operating condition parameters, preprocessing the original target item operating condition parameters, and obtaining multiple state evolution sample pairs based on the preprocessed target item operating condition parameters.

[0023] Specifically, the preprocessing of the original target item operating parameters may include performing preprocessing operations such as data alignment and synchronization, data cleaning, data filtering and noise reduction, missing value imputation, outlier removal and / or feature extraction on the target item operating parameters.

[0024] Step 102 involves constructing a preliminary dynamic model of the target device using the Koopman-Extended Dynamic Mode Decomposition algorithm based on multiple state evolution sample pairs. This step facilitates the verification of the preliminary dynamic model and the acquisition of the operational dynamic model of the target device based on the verification results.

[0025] Optionally, prior to step 102, a Koopman lifting dictionary is constructed, comprising constant terms, linear terms, quadratic terms, and cross terms.

[0026] What is noteworthy is that during the operation of tobacco drying equipment, there are often linear effects and single-variable nonlinear effects, such as the nonlinear effect of temperature rise on moisture change and variable coupling effects, such as the coupling effect between hot air temperature and outlet moisture content. Therefore, the embodiments of the present invention construct a Koopman lifting dictionary including constant terms, linear terms, quadratic terms and cross terms, which can better describe the different types of influence between operating parameters, thereby improving the modeling accuracy.

[0027] Specifically, the aforementioned Koopman lifting dictionary can also include some of the four terms: constant terms, linear terms, quadratic terms, and cross terms.

[0028] Specifically, the aforementioned Koopman lifting dictionary can also be constructed based on scalar functions such as trigonometric functions, exponential functions, logarithmic functions, and / or radial basis functions.

[0029] Specifically, the process of constructing a preliminary dynamic model of the target device using the Koopman-Extended Dynamic Mode Decomposition algorithm based on multiple state evolution sample pairs may include: obtaining training sample pairs based on state evolution sample pairs and the Koopman boosting dictionary; constructing a boosted state matrix, a control input matrix, and a next-step boosted state matrix based on the training sample pairs, and constructing the Koopman linear prediction equation based on the boosted state matrix, the control input matrix, and the next-step boosted state matrix; solving the coefficient matrix of the Koopman linear prediction equation, and constructing the running dynamic model using the Koopman boosting dictionary and the coefficient matrix.

[0030] Step 103: Obtain and verify the reliability of the preliminary dynamic model based on the reconstruction error. This step facilitates the acquisition of the target device's operational dynamic model based on the reliability verification results of the preliminary dynamic model.

[0031] Specifically, the process of obtaining the reconstruction error of the preliminary dynamic model includes: obtaining the predicted value sequence of state parameters through the preliminary dynamic model; performing time-by-time difference calculation on the predicted value sequence of state parameters and the corresponding true value sequence to obtain the single-point reconstruction deviation; and statistically integrating the single-point reconstruction deviations at each time point through mean square error, root mean square error, or mean absolute error to obtain the reconstruction error of the preliminary dynamic model.

[0032] Specifically, the process of verifying the reliability of the preliminary dynamic model based on the reconstruction error of the preliminary dynamic model may include: when the reconstruction error falls within the error value range corresponding to the unreliable judgment result, the preliminary dynamic model is judged as failing the reliability verification; when the reconstruction error falls within the error value range corresponding to the reliable judgment result, the preliminary dynamic model is judged as passing the reliability verification.

[0033] Step 104: If the preliminary dynamic model fails reliability verification, an optimized state evolution sample pair is determined based on the reconstruction error, and the preliminary dynamic model is updated based on the optimized state evolution sample pair to obtain the target equipment's operational dynamic model. This step, based on steps 101 and 103, obtains an equipment operational dynamic model with self-diagnosis and self-healing capabilities, thereby significantly improving the control performance of automated control and enhancing the stability of the automated control system.

[0034] Specifically, the process of determining the optimized state evolution sample pair based on the reconstruction error may include: when the reconstruction error meets the first requirement, obtaining the optimized state evolution sample pair based on the normal target item parameters in the real value sequence corresponding to the aforementioned predicted value sequence.

[0035] Specifically, the process of updating the preliminary dynamic model based on the optimized state evolution sample pairs to obtain the target device's operational dynamic model may include: constructing the operational dynamic model based on the optimized state evolution sample pairs using the Koopman-Extended Dynamic Mode Decomposition algorithm, and replacing the preliminary dynamic model with the operational dynamic model.

[0036] Specifically, steps 103 and 104 can be performed during the offline construction of the target device's runtime dynamic model, or they can be performed after the runtime dynamic model of the target device has been deployed and the runtime dynamic model is used to predict the state parameters of the target device online.

[0037] The following further describes the device operation modeling method provided by the embodiments of the present invention, such as... Figure 2 As shown, that is Figure 1 Step 101 may include the following steps: Step 1011: Obtain the complete numerical time series of the target item's operating condition parameters. The complete numerical time series is a sequence of the target item's operating condition parameters that are continuously and completely collected in real time during historical operation periods and are ordered in time.

[0038] Specifically, the aforementioned target item operating condition parameter can be one of multiple target item operating condition parameters.

[0039] Specifically, the time axis of each operating condition parameter can be aligned first, and the problems of sampling period differences, timestamp offsets, missing data and duplicate sampling points between each operating condition parameter can be uniformly handled to obtain the complete numerical time series corresponding to each operating condition parameter.

[0040] Step 1012: Divide the complete numerical time series into multiple time series segments based on the operating condition labels corresponding to the complete numerical time series. The multiple time series segments include one or more continuous stable operating condition segments and one or more unstable operating condition segments.

[0041] Specifically, the aforementioned operating condition labels can be understood as one or more labels that can be used to accurately determine whether a running time point is the running time corresponding to the same continuous stable operating condition segment. Specifically, they may include: processing task labels, material process labels, and / or equipment operating condition labels.

[0042] Specifically, the aforementioned processing task tags can be understood as tags used to record information related to processing tasks and identify task attributes, progress status, and operational behaviors during the production execution process. These tags may include: batch identifiers, work order information, brand or product codes, production stages or batch status, event tags, and / or operation confirmation tags, etc.

[0043] Specifically, the aforementioned material process labels can be understood as labels used to coordinate and manage the real-time status of materials used in production, as well as the configuration, issuance, and verification of production formulas. These labels may include material formula information, formula issuance and readback verification, and / or material status.

[0044] Specifically, the aforementioned equipment status labels can be understood as labels used to characterize the real-time operating status and basic attribute information of the production line and various equipment units. These labels may include equipment or line status and / or equipment unit identification labels.

[0045] In a specific example, where the target equipment is a material processing device, the specific information of the operating condition label is shown in Table 1: Table 1 Specifically, the process of dividing a complete numerical time series segment into multiple time series segments based on the operating condition labels corresponding to the complete numerical time series segment may include: determining unstable operating condition segments based on the operating condition labels, removing unstable operating condition segments from the complete numerical time series segment, and dividing the remaining numerical segments of the complete numerical time series segment after removing the unstable operating condition segments into multiple continuous stable operating condition segments based on the operating condition labels.

[0046] Specifically, the process of dividing a complete numerical time series segment into multiple time series segments based on the operating condition labels corresponding to the complete numerical time series segment may include: determining the numerical segments corresponding to the running time of the corresponding operating condition labels, including: the start of plate change, the end of plate change, the start of batch change, the end of batch change, the start of line cleaning, the end of cleaning, and the trial run or manual intervention, as unstable operating condition segments.

[0047] Specifically, the process of dividing the complete numerical time series segment into multiple time series segments based on the operating condition label corresponding to the complete numerical time series segment may also include: if there is a change in batch identifier, brand, formula identifier, formula version, or work order number, the boundary time point of the unstable operating condition segment is determined based on the corresponding change start time point, and the unstable operating condition segment is determined based on the boundary time point of the unstable operating condition segment.

[0048] Specifically, the process of determining the boundary time points of the unstable operating condition segments based on the corresponding change start time points, and determining the unstable operating condition segments based on the boundary time points of the unstable operating condition segments, may further include: determining the time points of the first duration before and the second duration after the change start time points as the time point boundaries of the time series segments, and determining the numerical segments corresponding to the time periods between the first duration before and the second duration after as the unstable operating condition segments, so as to exclude the time periods before and after which the process is still in the stages of disturbance, line clearing, formula downloading, readback verification, and process re-stabilization.

[0049] Specifically, the first and second durations mentioned above can be set based on empirical data.

[0050] Specifically, the process of dividing the complete numerical time series segment into multiple time series segments based on the operating condition labels corresponding to the complete numerical time series segment may also include: if the production stage changes from stable operation to preparation, cleaning, evacuation, shutdown or startup, multiple operating condition labels change synchronously, the material status changes from feeding to material interruption, no load or reverse recovery, or the equipment status changes from automatic to manual, fault, maintenance or bypass, then the starting time point of the corresponding change is determined as the boundary time point of the unstable operating condition segment, and the above-mentioned unstable operating condition segment is determined based on the boundary time point of the unstable operating condition segment.

[0051] Specifically, the process of dividing a complete numerical time series into multiple time series segments based on the operating condition labels corresponding to the complete numerical time series can also include: directly identifying each continuous stable operating condition segment based on the operating condition labels.

[0052] Step 1013: Obtain the normal target item operating condition parameters within each continuous stable operating condition segment.

[0053] Optional, such as Figure 3 As shown, that is Figure 2Step 1013 may include the following steps: Step 1013A: For each continuous stable operating condition segment, calculate the median and median absolute deviation of the first difference quotient of the target item operating condition parameter sequence within the current continuous stable operating condition segment.

[0054] Specifically, the process of calculating the median and absolute deviation of the first-order difference quotient of the target item's operating condition parameter sequence within the current continuous stable operating condition segment can include: for each operating condition parameter, the median and absolute deviation of the first-order difference quotient of the current item's operating condition parameter sequence are calculated using the following formula: in, This represents the first-order difference quotient of the current operating condition parameter sequence; This represents the median of the first-order difference quotient; This represents the absolute deviation of the median of the first-order difference quotient; and This represents the value of the current operating condition parameter corresponding to two adjacent time points; and This indicates two adjacent moments.

[0055] Step 1013B: Calculate the anomalous jump score based on the first-order difference quotient, the median of the first-order difference quotient, and the absolute deviation of the median using the Z-score calculation formula.

[0056] Specifically, the first-order difference quotient can be used as the observed value, the median can be used instead of the mean, and the absolute deviation of the median can be used instead of the standard deviation. The Z-score formula can be used to calculate the outlier jump score.

[0057] Specifically, the process of calculating the outlier jump score using the Z-score formula based on the normal distribution scaling correction factor, the first-order difference quotient, and the median and absolute deviation of the first-order difference quotient can also include: using the first-order difference quotient, the median and absolute deviation of the first-order difference quotient, and the Z-score formula to make the absolute deviation of the median comparable to the standard deviation scale. This can be achieved using the following formula: in, represents the anomalous jump score, and 0.6745 represents the normal distribution scaling correction factor.

[0058] Understandably, the median absolute deviation is a robust estimate of the scale of fluctuations and is less likely to be skewed by a few extreme points than the ordinary standard deviation.

[0059] Optionally, the process of calculating the abnormal jump score based on the first-order difference quotient, the median of the first-order difference quotient, and the median absolute deviation using the Z-score calculation formula includes: when the median absolute deviation of the first-order difference quotient is less than the first absolute deviation threshold, the abnormal jump score is calculated using the Z-score calculation formula based on the larger value between the median absolute deviation of the first-order difference quotient and the allowable benchmark value for rate of change fluctuation, the first-order difference quotient, and the median of the first-order difference quotient.

[0060] Specifically, the aforementioned first absolute deviation threshold can be set based on empirical data, for example, it can be set to 0.2 to 0.4.

[0061] Specifically, the aforementioned allowable benchmark value for rate of change fluctuation can be set based on the corresponding sensor resolution, engineering quantity range, historical normal fluctuation lower limit, interquartile range, or operating condition parameter rate of change threshold.

[0062] It can be understood that when the data is extremely stable, the median absolute deviation of the first-order difference quotient is close to 0, which will lead to abnormal jump scores, and thus cause normal data without abnormalities to be misjudged as outliers. Therefore, calculating abnormal jump scores based on the larger value between the median absolute deviation of the first-order difference quotient and the allowable benchmark value of the rate of change fluctuation can help reduce misjudgments.

[0063] Specifically, when the corresponding sampling interval is fixed, the abnormal jump score can also be calculated using the Z-score calculation formula based on the difference value of the target item working condition parameter sequence, the median of the difference value, and the absolute deviation of the median.

[0064] Step 1013C: Based on the abnormal jump score, the abnormal jump score threshold, and the physical limit range of the rate of change of the target item's operating condition parameters, determine the abnormal jump parameters in the target item's operating condition parameter sequence.

[0065] Specifically, the above-mentioned abnormal jump score threshold can be adjusted based on the on-site data, and can be set to 3.0 to 5.0.

[0066] Specifically, the aforementioned physical limit range of the rate of change can be understood as the maximum range of change in the value of the corresponding target operating condition parameter, which is constrained by the equipment's operating mechanism and process characteristics. Examples include the maximum opening and closing speed of a valve, the maximum acceleration of a belt speed, the maximum temperature rise rate, and the maximum response rate of a flow meter.

[0067] Specifically, the process of determining abnormal jump parameters in the target item operating condition parameter sequence based on abnormal jump scores, abnormal jump score thresholds, and the physical limit range of the rate of change of target item operating condition parameters may include: when the abnormal jump score is greater than the abnormal jump score threshold, the corresponding target item operating condition parameter is determined as a candidate abnormal operating condition parameter; when the candidate abnormal operating condition parameter exceeds the corresponding physical limit range of the rate of change, the candidate abnormal operating condition parameter is determined as an abnormal jump parameter.

[0068] It is understandable that anomalous jumps in fractional values ​​can only indicate statistical anomalies, not physical anomalies on their own. Therefore, superimposing engineering constraints can increase the credibility of jump anomalies.

[0069] Specifically, abnormal jump parameters in the target item's operating condition parameter sequence can be determined solely based on the abnormal jump score and the abnormal jump score threshold.

[0070] It is understandable that when operating parameters experience sudden spikes, jumps, or abrupt changes far exceeding their normal fluctuation scale within adjacent moments or short time windows—for example, in a stable production section, the outlet moisture content normally changes by only about 0.01% per second, but suddenly changes by 0.8% within a sampling interval; or the valve opening normally adjusts by only 0.2% per second, but suddenly jumps to 20%—these points are often not true smooth evolutions of the system, but rather atypical samples caused by abnormal events, cross-segment splicing, sensor problems, actuators, human intervention, sudden changes in setpoints, etc. Therefore, sample optimization can be performed by identifying and eliminating abnormally fluctuating parameters.

[0071] Step 1013D: Based on the standard deviation, standard deviation threshold, and average value of the target item operating condition parameters within the anomaly identification sliding window, as well as the physical boundary values ​​of the target item operating condition parameters, determine the abnormal static parameters in the target item operating condition parameter sequence.

[0072] Specifically, the window length of the above-mentioned anomaly detection sliding window can be set based on empirical data, for example, it can be set to 10s-60s.

[0073] Specifically, the process of determining abnormal static parameters in the target item operating condition parameter sequence based on the standard deviation, standard deviation threshold, and average value of the target item operating condition parameters within the anomaly identification sliding window, as well as the physical boundary value of the target item operating condition parameters, may include: when the standard deviation of the target item operating condition parameter within the anomaly identification sliding window is less than the standard deviation threshold, and the absolute deviation of the average value of the target item operating condition parameter within the anomaly identification sliding window relative to the physical boundary value of the target item operating condition parameter is less than the second absolute deviation threshold, the corresponding target item operating condition parameter within the anomaly identification sliding window is determined as an abnormal static parameter.

[0074] Specifically, if the standard deviation of the target item's operating condition parameter within multiple consecutive anomaly identification sliding windows is less than the standard deviation threshold, and the absolute deviation of the average value of the target item's operating condition parameter within the anomaly identification sliding window relative to the physical boundary value of the target item's operating condition parameter is less than the second absolute deviation threshold, the target item's operating condition parameter within the corresponding anomaly identification sliding window is determined as an abnormal static parameter, wherein the number of windows in multiple consecutive anomaly identification sliding windows is greater than the second number threshold.

[0075] Specifically, the aforementioned second quantity threshold can be set based on empirical data, for example, it can be set to 2 to 4.

[0076] In actual equipment operation scenarios, there are usually periods where actuator commands, actuator feedback, or related process variables remain almost unchanged, and operating parameters remain at a certain boundary or in a very small response range. This indicates that although the actuator is in the control link, it does not produce effective physical action. In control valves, dampers, baffles, feeders, servo mechanisms, and frequency converters, for example, one or more of the following four situations exist: (1) Dead zone / dead band: The input signal changes within a small range, but the output does not respond; (2) Hysteresis / backlash: When the signal is reversed, it needs to go through a "no-stroke" before it can drive the mechanism; (3) Stickiness / jamming: The static friction is too large, and small signals cannot push it. After accumulating to a certain extent, it suddenly slips; (4) Limit / saturation: The actuator is close to 0%, 100%, or mechanical limit, and it cannot respond to commands anymore. The embodiments of the present invention can eliminate abnormal parameters in the above operating scenarios, thereby optimizing the state evolution sample pairs.

[0077] In addition, since a single window meeting the condition may only be a normal transient state, the corresponding operating condition parameter can only be determined as an abnormal static parameter when multiple windows meet the condition consecutively, which can avoid misjudgment and omission of typical scenarios.

[0078] Step 1013E: Remove abnormal jump parameters and abnormal static parameters from the current continuous stable operating condition segment to obtain the normal target item operating condition parameters within the current continuous stable operating condition segment.

[0079] Optionally, abnormal jump parameters in the current continuous stable operating condition segment can be removed, and the normal target item operating condition parameters can be determined based on the remaining parameters after removing abnormal jump parameters in the current continuous stable operating condition segment.

[0080] Optionally, normal target item operating condition parameters can be determined based on the abnormal static parameters in the current continuous stable operating condition segment after removing the abnormal static parameters, and based on the remaining parameters after removing the abnormal static parameters in the current continuous stable operating condition segment.

[0081] Specifically, it can also directly identify and obtain the normal target item operating condition parameters in the current continuous stable operating condition segment.

[0082] Step 1014: For each continuous stable operating condition segment, determine the state evolution sample pair corresponding to the current continuous stable operating condition segment based on the normal target item operating condition parameters with adjacent sampling times within the current continuous stable operating condition segment.

[0083] Understandably, most material processing is done intermittently in multiple batches. Existing algorithms blindly pair data according to time sequence, forcibly linking "the last second before the previous batch was shut down and disconnected" with "the first second before the next batch was started," assuming a massive physical mutation occurred within that one second. This pseudo-mutation data across batches, once entering the least squares calculation pool, instantly skews the parameters of the entire Koopman matrix, causing the intelligent control system to make directional misjudgments and severe oscillations during normal predictions.

[0084] In this embodiment of the invention, pairing is performed only when the target item's operating parameters belong to the same continuous stable operating condition segment, which can optimize the state evolution sample pairs.

[0085] Optionally, the process of determining the state evolution sample pairs corresponding to the current continuous stable operating condition segment based on the normal target item operating condition parameters with adjacent sampling times within the current continuous stable operating condition segment includes: performing Z-score normalization on the target item operating condition parameters to obtain normalized operating condition parameters; and using the normalized operating condition parameters corresponding to each adjacent time pair corresponding to the current continuous stable operating condition segment to form the state evolution sample pairs corresponding to the respective adjacent time pairs.

[0086] Specifically, the process of forming state evolution sample pairs corresponding to each adjacent time pair using the normalized operating condition parameters corresponding to the current continuous stable operating condition segment includes: constructing state input vectors and control input vectors respectively using the normalized state parameters and normalized control parameters in the normalized operating condition parameters corresponding to the previous time of the adjacent time pair; and constructing state output vectors using the normalized state parameters in the normalized operating condition parameters corresponding to the next time of the adjacent time pair.

[0087] In a specific example, the target operating parameters include: exit tobacco moisture content, exit tobacco temperature, and hot air temperature. The above state input vector can be represented as: The above control input vector can be expressed as: Specifically, the normal target item operating parameters corresponding to each adjacent time pair of the current continuous stable operating condition segment can be directly used to form the state evolution sample pairs corresponding to the adjacent time pairs.

[0088] Understandably, due to the significant differences in scale between various operating parameters, such as the relatively small percentage concentration and the massively variable equipment temperature, it is necessary to force a unified calculation using all aggregated valid time-series data to determine the global static mean and global standard deviation before performing any nonlinear expansion. This eliminates the risk of imbalanced principal component extraction caused by the uneven input magnitudes during subsequent high-dimensional partial derivative gradient optimization calculations.

[0089] The embodiments of the present invention can optimize state evolution sample pairs, which can help improve modeling accuracy, improve the prediction accuracy of subsequent control systems, and avoid oscillations.

[0090] The device operation modeling method provided in the embodiments of the present invention will be further described below.

[0091] like Figure 4 As shown, that is Figure 1 Step 102 may include the following steps: Step 1021: Obtain training sample pairs based on state evolution sample pairs and Koopman boosting dictionary.

[0092] In a specific instance, when the Koopman boosting dictionary includes constant terms, first-order terms, second-order terms, and cross terms, the process of obtaining training sample pairs based on state evolution sample pairs and the Koopman boosting dictionary includes: inputting the state vector... The dimension-upgrading representation is: .

[0093] Step 1022: Construct the boosted state matrix, control input matrix, and next-step boosted state matrix based on the training sample pairs, and construct the Koopman linear prediction equation based on the boosted state matrix, control input matrix, and next-step boosted state matrix.

[0094] In a specific instance, the above-mentioned boosted state matrix can be represented as: .

[0095] In a specific example, the above control input matrix can be represented as: In a specific instance, the aforementioned next-step boosting state matrix can be represented as: Specifically, the above Koopman linear prediction equation can be expressed as: Where K represents the coefficient matrix, .

[0096] Step 1023: Solve the coefficient matrix of the Koopman linear prediction equation using the ridge regression method.

[0097] Specifically, the process of solving the coefficient matrix of the Koopman linear prediction equation using the ridge regression method can include: using a regularized Koopman operator least squares solution to optimize the coefficient matrix of the Koopman linear prediction equation by minimizing the sum of the squared norm of the fitted residuals and the L2 regularization penalty term of the matrix. The regularized Koopman operator least squares solution can be expressed as: in, λ represents the ridge regression regularization coefficient. The larger the coefficient, the smoother the model, but it may underfit; the smaller the coefficient, the closer it fits the training data, but it is more prone to overfitting to noise. express Identity matrix.

[0098] Step 1024: Construct a dynamic model using the Koopman boosting dictionary and coefficient matrix.

[0099] Understandably, by superimposing regularization coefficients on the diagonal of the covariance matrix, it is possible to enhance the constraint on the inverse solution of the noise matrix without increasing the algorithm complexity. Combined with global dimensionless scaling, it is possible to weaken the parameter allocation of overfitting high-frequency perturbations while balancing and unifying the data metric basis. Together, they can facilitate the approximate analytical extraction of the slow dynamic manifold of the state core under low signal-to-noise ratio, thereby forcibly suppressing the model's over-catering to high-frequency amplified noise fragments and improving the modeling accuracy.

[0100] The following further explains the device operation modeling method provided in the embodiments of the present invention, such as... Figure 5 As shown, that is Figure 1 Step 103 may include the following steps: Step 1031: Based on multiple state evolution sample pairs, a multi-step time series prediction method is used to predict state parameters through a preliminary dynamic model to obtain multiple predicted values ​​of state parameters.

[0101] Specifically, other methods such as iterative multi-step prediction can also be used for prediction.

[0102] Specifically, the aforementioned multi-step time series forecasting method can be understood as a time series analysis and forecasting method that predicts the numerical changes of multiple consecutive time steps in the future at once, thereby enabling long-term multi-time state prediction.

[0103] Specifically, the number of steps for the aforementioned multiple consecutive time steps can be set based on empirical data, for example, it can be set to 5-10 steps.

[0104] Specifically, when the state parameters include multiple state parameters, the predicted values ​​of the above multiple state parameters can be the predicted values ​​corresponding to any one of the state parameters.

[0105] Specifically, before step 1031, multiple state evolution sample pairs corresponding to the second historical period can be obtained using the same state evolution sample pair acquisition method as described above.

[0106] Step 1032: Obtain the root mean square error of the predicted value of each state parameter relative to the true value of the corresponding state parameter through the error evaluation sliding window as the reconstruction error corresponding to the predicted value of the corresponding state parameter.

[0107] Specifically, the window length of the aforementioned error assessment sliding window can be set based on empirical data, for example, it can be set to 10s to 60s.

[0108] Specifically, the reconstruction error corresponding to the predicted value of each state parameter can be determined based on the absolute error of the predicted value of each state parameter relative to the actual value of the corresponding state parameter.

[0109] Step 1033: When the values ​​of multiple consecutive reconstruction errors are greater than the reconstruction error threshold, it is determined that the preliminary dynamic model has failed the reliability verification and the number of consecutive moving average root errors is greater than the first quantity threshold.

[0110] Specifically, the aforementioned reconstruction error threshold can be calculated using the following formula: ( ) Where θ represents the reconstruction error threshold, E represents the reconstruction error, and k represents the standard deviation relaxation factor. This represents the average reconstruction error of the model on normal, stable data. This represents the standard deviation of the model's reconstruction error on normal, stable data. and It can be calculated based on the reconstruction error on the normal training set or validation set.

[0111] Specifically, before step 1033, the value of the standard deviation relaxation factor can be determined based on the application scenario of this embodiment of the invention.

[0112] Specifically, in offline scenarios, k can be set to 2 to improve sensitivity to fragments.

[0113] Specifically, in scenarios where online prediction is possible, k can be set to be greater than or equal to 3 to capture only very obvious anomalous disturbances and reduce false alarms.

[0114] Specifically, the first quantity threshold mentioned above can be set based on empirical data, for example, it can be set to 3-5.

[0115] The examples of this invention can help improve the accuracy of model self-diagnosis.

[0116] The following further explains the device operation modeling method provided in the embodiments of the present invention, such as... Figure 6 As shown, that is Figure 1 Step 104 may include the following steps: Step 1041: Determine the state evolution sample pairs corresponding to multiple consecutive moving root mean square errors as abnormal state evolution sample pairs.

[0117] Step 1042: Remove abnormal state evolution sample pairs from multiple state evolution sample pairs to obtain optimized state evolution sample pairs.

[0118] It is understandable that when multiple consecutive reconstruction errors exceed the reconstruction error threshold, it usually means that the corresponding state parameters have been affected by non-critical external physical events that were not detected in the original data screening stage and were not modeled, such as material blockage fluctuations. Therefore, removing the corresponding abnormal state evolution sample pairs can optimize the state evolution sample pairs.

[0119] Step 1043A: When the proportion of the optimized state evolution sample pair relative to the initial multiple state evolution sample pairs is not less than the first proportion threshold, the preliminary dynamic model of the target device is re-executed based on the optimized state evolution sample pair using the Koopman-Extended Dynamic Mode Decomposition algorithm based on multiple state evolution sample pairs, until the preliminary dynamic model passes the reliability verification.

[0120] Specifically, the aforementioned first percentage threshold can be set based on empirical data, for example, it can be set to 10%-15%.

[0121] Step 1043B: When the proportion of the optimized state evolution sample pairs relative to the initial multiple state evolution sample pairs is less than the first proportion threshold, the parameter items of the target item operating condition parameters are amplified, and the execution is restarted based on the amplified target item operating condition parameters to obtain multiple state evolution sample pairs based on the target item operating condition parameters collected in real time during the historical running period of the target device. The target item operating condition parameters include state parameters and control parameters, until the preliminary dynamic model passes the reliability verification.

[0122] Optionally, the process of the parameter items of the above-mentioned amplification target item operating condition parameters includes: the state parameter items in the amplification target item operating condition parameters.

[0123] Specifically, the process of expanding the state parameter items in the above-mentioned target operating condition parameters may include expanding the state parameter items from three state parameter items, namely: outlet tobacco moisture content, outlet tobacco temperature and hot air temperature, to eight state parameter items, namely: outlet tobacco moisture content, hot air temperature, outlet tobacco temperature, feeding rate, inlet tobacco temperature, inlet air volume, heating steam pressure and system negative pressure.

[0124] Understandably, if the system detects that the proportion of removed anomalous bands is too high, it indicates that the state variables used in the current model are insufficient to explain key disturbances in the production process. In this case, the system should not simply repeatedly delete anomalous data, but should proactively introduce additional sensor channels to expand the state vector, input vector, or disturbance vector, allowing the next round of modeling to explain the key disturbance events that were not originally modeled. Therefore, this embodiment of the invention can improve the model's self-healing capability.

[0125] In a specific example, for tobacco drying equipment, the dynamic operation model constructed by the equipment operation modeling method provided in this embodiment of the invention performs better than the existing dynamic operation model based on the Koopman-Extended Dynamic Mode Decomposition algorithm when predicting the moisture content of the exit tobacco shreds. Figure 7 and Figure 8 As shown.

[0126] in, Figure 7 To compare the results of multi-step forward prediction, from Figure 7 As can be seen, the existing dynamic model can closely approximate the true value in the early stages, but as the prediction step size increases, the error accumulates rapidly and exhibits an exponential divergence trend due to the influence of noise and the exponential cross-amplification effect. In the later stages, it oscillates violently and exceeds the 25% safety control limit, entering the physical collapse failure zone. In contrast, the dynamic model constructed in this invention effectively resists the cascading noise amplification effect throughout the entire process. The predicted trajectory smoothly and stably follows the true value. Even with a long prediction step size, there is no obvious deviation or divergence. It always stays within the safe range, demonstrating excellent multi-step forward prediction robustness and noise resistance.

[0127] Figure 8 To compare the simulation results of closed-loop control of moisture content in exported tobacco, from Figure 8As can be seen from the existing dynamic operation model, its anti-disturbance capability is weak, and the adjustment process is subject to significant oscillations. In the early stages, when it repeatedly approaches the tolerance boundary, it fails to adjust under sudden strong operating conditions (such as valve jamming or gas network pressure drop), and the control quantity falls below the lower limit, resulting in process defects. The process capability index CPK is only 0.28, which does not meet the qualification standard. However, the dynamic operation model constructed by the embodiment of the present invention can quickly track the target value, effectively suppress fluctuations in the strong disturbance range, and always keep the control quantity stable within the process tolerance zone. The process capability index CPK reaches 1.21, which is at the industry-recognized excellent level, verifying its excellent disturbance suppression capability and control accuracy.

[0128] Figure 9 This is a structural diagram of a device for modeling equipment operation provided in an embodiment of the present invention. This device is suitable for executing the device operation modeling method provided in an embodiment of the present invention. Figure 7 As shown, the device may specifically include: The sample pair acquisition module 901 is used to acquire and obtain multiple state evolution sample pairs based on the target item operating condition parameters collected in real time during the historical operation period of the target device. The target item operating condition parameters include state parameters and control parameters. This module can facilitate the construction of a preliminary dynamic model of the target device based on multiple state evolution sample pairs.

[0129] Optionally, the sample pair acquisition module 901 can be specifically used to: acquire the complete numerical time series of the target item operating condition parameters, wherein the complete numerical time series is a numerical sequence of the target item operating condition parameters that are continuously complete and temporally ordered and collected in real time during historical operation; divide the complete numerical time series into multiple time series segments based on the operating condition labels corresponding to the complete numerical time series, wherein the multiple time series segments include one or more continuous stable operating condition segments and one or more unstable operating condition segments; acquire the normal target item operating condition parameters within each continuous stable operating condition segment; and for each continuous stable operating condition segment, determine the state evolution sample pair corresponding to the current continuous stable operating condition segment based on the normal target item operating condition parameters with adjacent sampling times within the current continuous stable operating condition segment.

[0130] Optionally, the sample acquisition module 901 can be specifically used to: calculate the median and absolute deviation of the first-order difference quotient of the target item operating condition parameter sequence within each continuous stable operating condition segment; calculate the abnormal jump score using the Z-score calculation formula based on the first-order difference quotient, the median of the first-order difference quotient, and the absolute deviation of the median; determine the abnormal jump parameters in the target item operating condition parameter sequence based on the abnormal jump score, the abnormal jump score threshold, and the physical limit range of the rate of change of the target item operating condition parameters; determine the abnormal static parameters in the target item operating condition parameter sequence based on the standard deviation, standard deviation threshold, and average value of the target item operating condition parameters within the anomaly identification sliding window, as well as the physical boundary values ​​of the target item operating condition parameters; and remove the abnormal jump parameters and abnormal static parameters in the current continuous stable operating condition segment to obtain the normal target item operating condition parameters within the current continuous stable operating condition segment.

[0131] Optionally, the sample pair acquisition module 901 can be specifically used to calculate the abnormal jump score using the Z-score calculation formula when the median absolute deviation of the first-order difference quotient is less than the first absolute deviation threshold, based on the larger value between the median absolute deviation of the first-order difference quotient and the allowable benchmark value of the rate of change fluctuation, the first-order difference quotient, and the median of the first-order difference quotient.

[0132] Optionally, the sample pair acquisition module 901 can be specifically used to perform Z-score normalization on the target item operating condition parameters to obtain normalized operating condition parameters; and to use the normalized operating condition parameters corresponding to each adjacent time pair corresponding to the current continuous stable operating condition segment to form the state evolution sample pair corresponding to the corresponding adjacent time pair.

[0133] The preliminary dynamic model acquisition module 902 is used to construct a preliminary dynamic model of the target device based on multiple state evolution sample pairs using the Koopman-Extended Dynamic Mode Decomposition algorithm. This module facilitates the verification of the preliminary dynamic model and, based on the verification results, obtains the operational dynamic model of the target device.

[0134] Optionally, the aforementioned preliminary dynamic model acquisition module 902 can be specifically used to construct a Koopman lifting dictionary, including constant terms, first-order terms, second-order terms, and cross terms, before constructing the preliminary dynamic model of the target device based on multiple state evolution sample pairs using the Koopman-extended dynamic mode decomposition algorithm.

[0135] Optionally, the aforementioned preliminary dynamic model acquisition module 902 can be specifically used to: acquire training sample pairs based on state evolution sample pairs and the Koopman boosting dictionary; construct a boosting state matrix, a control input matrix, and a next-step boosting state matrix based on the training sample pairs; construct the Koopman linear prediction equation based on the boosting state matrix, the control input matrix, and the next-step boosting state matrix; solve the coefficient matrix of the Koopman linear prediction equation using the ridge regression method; and construct the running dynamic model using the Koopman boosting dictionary and the coefficient matrix.

[0136] The reliability verification module 903 is used to acquire and verify the reliability of the preliminary dynamic model based on the reconstruction error. This module can facilitate the acquisition of the target device's operational dynamic model based on the reliability verification results of the preliminary dynamic model.

[0137] Optionally, the aforementioned reliability verification module 903 can be specifically used to: obtain multiple state parameter prediction values ​​by using a multi-step time-series prediction method based on multiple state evolution sample pairs to predict state parameters through a preliminary dynamic model; obtain the moving root mean square error of each state parameter prediction value relative to the corresponding state parameter true value through an error evaluation sliding window as the reconstruction error corresponding to the corresponding state parameter prediction value; and determine that the preliminary dynamic model has failed reliability verification when the values ​​of multiple consecutive reconstruction errors are greater than the reconstruction error threshold, and the number of moving root mean square errors of multiple consecutive moving root mean square errors is greater than the first quantity threshold.

[0138] The dynamic model acquisition module 904 is used to determine optimized state evolution sample pairs based on reconstruction errors when the preliminary dynamic model fails reliability verification, and to update the preliminary dynamic model based on the optimized state evolution sample pairs to obtain the operational dynamic model of the target equipment. This module, combined with modules 901 to 903, can train an operational dynamic model of the equipment with self-diagnosis and self-healing capabilities, thereby significantly improving the control performance of automated control and enhancing the stability of the automated control system.

[0139] Optionally, the aforementioned dynamic model acquisition module 904 can be specifically used to: identify multiple consecutive sliding root mean square errors corresponding to state evolution sample pairs as abnormal state evolution sample pairs; remove abnormal state evolution sample pairs from multiple state evolution sample pairs to obtain optimized state evolution sample pairs; when the proportion of optimized state evolution sample pairs relative to the initial multiple state evolution sample pairs is not less than a first proportion threshold, restart the execution of constructing a preliminary dynamic model of the target device based on multiple state evolution sample pairs using the Koopman-Extended Dynamic Mode Decomposition algorithm, until the preliminary dynamic model passes reliability verification; when the proportion of optimized state evolution sample pairs relative to the initial multiple state evolution sample pairs is less than the first proportion threshold, amplify the parameter items of the target item operating condition parameters, and restart the execution of acquiring multiple state evolution sample pairs based on the amplified target item operating condition parameters and the target item operating condition parameters collected in real time during the historical operation period of the target device, the target item operating condition parameters including state parameters and control parameters, until the preliminary dynamic model passes reliability verification.

[0140] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is merely an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the functional modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0141] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the device operation modeling method provided in any of the above embodiments.

[0142] This invention also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the device operation modeling method provided in any of the above embodiments.

[0143] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the device operation modeling method as described in any of the embodiments of this invention.

[0144] The following is for reference. Figure 10 It shows a schematic diagram of the structure of a computer system 1000 suitable for implementing an electronic device according to embodiments of the present invention. Figure 10 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0145] like Figure 10 As shown, the computer system 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage section 1008 into a random access memory (RAM) 1003. The RAM 1003 also stores various programs and data required for the operation of the system 1000. The CPU 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0146] The following components are connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. A removable medium 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 1010 as needed so that computer programs read from it can be installed into storage section 1008 as needed.

[0147] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1009, and / or installed from removable medium 1011. When the computer program is executed by central processing unit (CPU) 1001, it performs the functions defined in the system of this invention.

[0148] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0149] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0150] The modules and / or units described in the embodiments of the present invention can be implemented in software or hardware. The described modules and / or units can also be housed in a processor; for example, a processor can be described as including a sample pair acquisition module, a preliminary dynamic model acquisition module, a reliability verification module, and a dynamic model acquisition running module. The names of these modules do not necessarily limit the module itself.

[0151] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to implement a device operation modeling method comprising the following steps: acquiring and obtaining multiple state evolution sample pairs based on target item operating condition parameters collected in real time during historical operation periods, the target item operating condition parameters including state parameters and control parameters; constructing a preliminary dynamic model of the target device based on the multiple state evolution sample pairs using the Koopman-Extended Dynamic Mode Decomposition algorithm; acquiring and performing reliability verification on the preliminary dynamic model based on the reconstruction error; and when the preliminary dynamic model fails the reliability verification, determining optimized state evolution sample pairs based on the reconstruction error, and updating the preliminary dynamic model based on the optimized state evolution sample pairs to obtain the operating dynamic model of the target device.

[0152] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for modeling equipment operation, characterized in that, include: Multiple state evolution sample pairs are obtained based on the target item operating condition parameters collected in real time during the historical running period of the target device. The target item operating condition parameters include state parameters and control parameters. Based on the multiple state evolution sample pairs, a preliminary dynamic model of the target device is constructed using the Koopman-Extended Dynamic Mode Decomposition algorithm; The reliability of the preliminary dynamic model is verified by acquiring and using the reconstruction error of the preliminary dynamic model. If the preliminary dynamic model fails the reliability verification, an optimized state evolution sample pair is determined based on the reconstruction error, and the preliminary dynamic model is updated based on the optimized state evolution sample pair to obtain the operating dynamic model of the target device.

2. The equipment operation modeling method according to claim 1, characterized in that, The process of acquiring and obtaining multiple state evolution sample pairs based on the target item operating condition parameters collected in real time by the target device during historical operating periods includes: Obtain the complete numerical time series of the target item operating condition parameters, wherein the complete numerical time series is a numerical sequence of the target item operating condition parameters that are continuously and completely collected in real time during the historical running period and are ordered in time. Based on the operating condition label corresponding to the complete numerical time series, the complete numerical time series is divided into multiple time series segments, and the multiple time series segments include one or more continuous stable operating condition segments and one or more unstable operating condition segments. Obtain the normal target item operating parameters within each continuous stable operating condition segment; For each continuous stable operating condition segment, the state evolution sample pair corresponding to the current continuous stable operating condition segment is determined based on the normal target item operating condition parameters with adjacent sampling times within the current continuous stable operating condition segment.

3. The equipment operation modeling method according to claim 2, characterized in that, The acquisition of normal target item operating condition parameters within each continuous stable operating condition segment includes: For each continuous stable operating condition segment, calculate the median and median absolute deviation of the first difference quotient of the target item operating condition parameter sequence within the current continuous stable operating condition segment; The anomalous jump score is calculated using the Z-score formula based on the first-order difference quotient, the median of the first-order difference quotient, and the absolute deviation of the median. Based on the abnormal jump score, the abnormal jump score threshold, and the physical limit range of the rate of change of the target item operating condition parameters, the abnormal jump parameters in the target item operating condition parameter sequence are determined; Based on the standard deviation, standard deviation threshold, and average value of the target item operating condition parameters within the anomaly identification sliding window, as well as the physical boundary values ​​of the target item operating condition parameters, abnormal static parameters in the target item operating condition parameter sequence are determined. By removing abnormal jump parameters and abnormal static parameters from the current continuous stable operating condition segment, the normal target item operating condition parameters within the current continuous stable operating condition segment are obtained.

4. The equipment operation modeling method according to claim 3, characterized in that, The calculation of the anomalous jump score based on the first-order difference quotient, the median of the first-order difference quotient, and the absolute deviation of the median using the Z-score formula includes: When the median absolute deviation of the first-order difference quotient is less than the first absolute deviation threshold, the abnormal jump score is calculated using the Z-score calculation formula based on the larger value between the median absolute deviation of the first-order difference quotient and the allowable benchmark value for rate of change fluctuation, the first-order difference quotient, and the median of the first-order difference quotient.

5. The equipment operation modeling method according to claim 2, characterized in that, The determination of the state evolution sample pair corresponding to the current continuous stable operating condition segment based on the normal target item operating condition parameters with adjacent sampling times within the current continuous stable operating condition segment includes: The target item's operating parameters are normalized using Z-score normalization to obtain normalized operating parameters; and Using the normalized operating parameters corresponding to each adjacent time pair of the current continuous stable operating condition segment, the state evolution sample pairs corresponding to the adjacent time pairs are formed.

6. The equipment operation modeling method according to claim 1, characterized in that, Before constructing the preliminary dynamic model of the target device using the Koopman-Extended Dynamic Mode Decomposition algorithm based on the multiple state evolution sample pairs, the method further includes: Construct a Koopman lifting dictionary that includes constant terms, linear terms, quadratic terms, and cross terms; The preliminary dynamic model of the target device is constructed using the Koopman-Extended Dynamic Mode Decomposition algorithm based on the multiple state evolution sample pairs, including: Training sample pairs are obtained based on the state evolution sample pairs and the Koopman boosting dictionary; Based on the training samples, a boosted state matrix, a control input matrix, and a next-step boosted state matrix are constructed, and a Koopman linear prediction equation is constructed based on the boosted state matrix, the control input matrix, and the next-step boosted state matrix. The coefficient matrix of the Koopman linear prediction equation is solved using the ridge regression method. The running dynamic model is constructed using the Koopman lifting dictionary and the coefficient matrix.

7. The equipment operation modeling method according to claim 1, characterized in that, The step of obtaining and verifying the reliability of the preliminary dynamic model based on the reconstruction error of the preliminary dynamic model includes: Based on the multiple state evolution sample pairs, a multi-step temporal prediction method is used to predict the state parameters through the preliminary dynamic model to obtain multiple predicted values ​​of state parameters; The moving root mean square error of each state parameter prediction value relative to the corresponding state parameter true value is obtained by using an error evaluation sliding window as the reconstruction error corresponding to the predicted state parameter value. When the values ​​of multiple consecutive reconstruction errors are greater than the reconstruction error threshold, the preliminary dynamic model is determined to have failed the reliability verification, and the number of consecutive moving root mean square errors is greater than the first quantity threshold.

8. The equipment operation modeling method according to claim 7, characterized in that, The step of determining optimized state evolution sample pairs based on the reconstruction error, and updating the preliminary dynamic model based on the optimized state evolution sample pairs to obtain the operational dynamic model of the target device, includes: The state evolution sample pairs corresponding to the consecutive multiple sliding root mean square errors are determined as abnormal state evolution sample pairs; The optimized state evolution sample pair is obtained by removing the abnormal state evolution sample pair from the plurality of state evolution sample pairs; When the proportion of the optimized state evolution sample pair relative to the initial plurality of state evolution sample pairs is not less than a first proportion threshold, the preliminary dynamic model of the target device constructed by the Koopman-Extended Dynamic Mode Decomposition algorithm based on the optimized state evolution sample pair is restarted, until the preliminary dynamic model passes the reliability verification. When the proportion of the optimized state evolution sample pair relative to the initial multiple state evolution sample pairs is less than a first proportion threshold, the parameter items of the target item operating condition parameters are amplified, and the process of acquiring and obtaining multiple state evolution sample pairs based on the target item operating condition parameters collected in real time by the target device during historical operation periods is restarted based on the amplified target item operating condition parameters, the target item operating condition parameters including state parameters and control parameters, until the preliminary dynamic model passes reliability verification.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the device operation modeling method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the device operation modeling method as described in any one of claims 1 to 8.

Images