Production line logistics beat correction method and device, electronic equipment and storage medium

Abstract

The present application provides a kind of production line logistics tempo correction method, device, electronic equipment and storage medium, belong to intelligent manufacturing technical field, including: obtaining historical task data set, effective task data set is obtained by deviation measurement screening;According to effective sample size, shunt adopts statistical quantile or target correction model to determine the corrected logistics tempo, and is used to update reference tempo.The present application is through the double filtering correction mechanism of constructing data pollution measurement cleaning and sample size evaluation and double branch prediction, first, the distortion data caused by insufficient transport capacity and other non-production factors is identified and removed by deviation measurement, to ensure the purity of data source, and then automatically switch to the statistical method with stronger robustness in small sample scene, avoids the risk of overfitting of correction model, effectively solves the tempo prediction distortion problem caused by logistics abnormal data pollution and small sample scene, realizes the high-precision, high-robustness adaptive correction of production line material consumption tempo.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method, apparatus, electronic device, and storage medium for correcting production line logistics cycle time. Background Technology

[0002] In the field of industrial automation, the material delivery efficiency of Automated Guided Vehicles (AGVs) directly affects the stable operation of the production line. To achieve efficient coordination between production and logistics, it is necessary to accurately predict the material consumption cycle time of each production line station as the triggering basis for AGV material delivery tasks.

[0003] To meet these requirements, existing technologies typically preset a theoretical baseline logistics cycle time for each production node in the Manufacturing Execution System (MES). This baseline logistics cycle time is based on a standard process flow setting. Some existing technologies attempt to use historical task data, such as task generation intervals or material requisition intervals, to statistically correct this baseline logistics cycle time in order to make it closer to actual production conditions.

[0004] However, the pre-set theoretical baseline logistics cycle time is static and cannot adapt to fluctuations in actual logistics cycle time caused by dynamic factors such as shifts and production line load, leading to biases in the prediction results. Furthermore, when using historical task data for correction, the collected data often contains a large amount of distorted data caused by non-productive logistics anomalies, such as AGV queuing and channel congestion. Existing statistical methods directly process such raw datasets containing distorted data, lacking an effective mechanism for verifying data usability. Summary of the Invention

[0005] This invention provides a production line logistics cycle time correction method, device, electronic device, and storage medium to solve the technical problem that the existing technology, when using historical data for cycle time correction, lacks an effective mechanism to identify the availability of data and directly processes distorted data containing logistics anomalies, resulting in insufficient reliability of the correction results. This invention achieves highly robust dynamic correction of material consumption cycle time at production line nodes.

[0006] This invention provides a production line logistics cycle time correction method, comprising the following steps: obtaining a historical task dataset of any target production node in the target production line, wherein the historical task dataset contains multiple historical task data, and determining the task generation interval corresponding to each historical task data; Based on the baseline logistics cycle time of each historical task data, the deviation measurement is performed on the task generation interval corresponding to each historical task data, so as to filter out the effective task dataset from the historical task dataset based on the measurement results. If the effective sample size in the effective task dataset is less than a preset sample size threshold, the corrected logistics cycle time of the target production node is determined based on the statistical quantiles of the effective task dataset. Otherwise, the effective task dataset is input into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node. The target correction model is obtained after training based on production factor samples and is used to characterize the nonlinear mapping relationship between multidimensional production factor features and logistics cycle time. Based on the corrected logistics cycle time, the baseline logistics cycle time of the target production node is updated.

[0007] According to a production line logistics cycle time correction method provided by the present invention, the step of measuring the deviation of the task generation interval in each historical task data according to the baseline logistics cycle time of each historical task data, and filtering the effective task dataset from the historical task dataset based on the measurement result, includes: The ratio of the task generation interval to the baseline logistics cycle time in each historical task data entry is calculated as the measurement result. If the measurement result is within a predetermined reasonable range, then the corresponding historical task data is determined to be valid task data. Construct the effective task dataset using all the aforementioned effective task data.

[0008] According to the production line logistics cycle time correction method provided by the present invention, the reasonable range is dynamically generated based on the following method: Arrange the multiple task generation intervals obtained during normal historical operation in ascending order of their numerical values; Obtain the first and third quartiles from the permutation results; Calculate the interquartile range between the first quartile and the third quartile; Based on the first quartile, the third quartile, the interquartile range, and the preset product coefficient, the lower and upper limits of the reasonable interval are calculated.

[0009] According to a production line logistics cycle time correction method provided by the present invention, after updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time, the method further includes: Obtain an online task data sequence containing multiple online task data, and perform time window sliding processing on the online task data sequence using a preset step size to obtain multiple sliding time windows; Within each sliding time window, the number of online task data whose measurement results are not within the reasonable range is counted to obtain the proportion of abnormal data; if the proportion of abnormal data within a consecutive preset number of sliding time windows is greater than a preset proportion threshold, a distribution offset signal is triggered. The target correction model is used to predict the beat of each online task data to obtain the prediction error corresponding to each online task data; the mean of each prediction error within the same sliding time window is calculated to obtain the mean prediction error; if the mean prediction error within a consecutive preset number of sliding time windows is greater than a preset error threshold, a model error signal is triggered. If the distribution offset signal and the model error signal are triggered simultaneously, the online task data is reacquired, and the effective task dataset is filtered and the model is reconstructed based on the reacquired online task data to obtain the corrected logistics cycle time for the next production period; if only the model error signal is triggered and the distribution offset signal is not triggered, the deviation measurement and validity determination of the online task data are re-performed to obtain the target task dataset.

[0010] According to a production line logistics cycle time correction method provided by the present invention, after obtaining the effective task dataset from the historical task dataset based on the measurement results, the method further includes: Obtain a historical task data sequence containing multiple historical task data, and perform time window sliding processing on the historical task data sequence using a preset step size to obtain multiple historical sliding time windows; Within each historical sliding time window, the number of historical task data whose measurement results are not within the reasonable range is statistically analyzed to obtain the percentage of data records. If the proportion of the data records within a consecutive first preset number of historical sliding time windows exceeds a preset proportion threshold, then the current sliding position is determined as a candidate point for production state transition. Based on the candidate points of production state transition, the time window sliding process is performed. If the proportion of the data records in the second consecutive preset number of historical sliding time windows is not greater than the preset proportion threshold, the starting historical task data in the first historical sliding time window in the second consecutive preset number of historical sliding time windows is extracted to obtain the modeling starting point. Based on the modeling starting point, the effective task dataset is divided into time periods to obtain multiple independent initial adaptive production time periods; the effective task data in each initial adaptive production time period is subjected to cycle time correction processing to obtain the corrected logistics cycle time corresponding to each initial adaptive production time period.

[0011] According to a production line logistics cycle time correction method provided by the present invention, the step of performing cycle time correction processing on the effective task data within each initial adaptive production period to obtain the corrected logistics cycle time corresponding to each initial adaptive production period includes: For any of the initial adaptive production periods, the task generation intervals in the effective task data are sorted in ascending order and quantiles are calculated to obtain a reference logistics cycle time. Based on the reference logistics cycle time and the current benchmark logistics cycle time of the target production node, the time period deviation for the current period is determined; If the time period deviation of the current time period is greater than a preset deviation threshold, then the valid task data of the current time period is subjected to cycle time correction processing to obtain the corrected logistics cycle time of the current time period; if the time period deviation of the current time period is less than or equal to the preset deviation threshold, then the current baseline logistics cycle time is determined as the corrected logistics cycle time of the current time period.

[0012] According to a production line logistics cycle time correction method provided by the present invention, when the effective sample size in the effective task dataset is less than a preset sample size threshold, determining the corrected logistics cycle time of the target production node based on the statistical quantiles of the effective task dataset includes: If the effective sample size is less than the preset sample size threshold, the generation intervals of all tasks in the effective task dataset are sorted in ascending order according to their numerical values ​​to obtain the duration sequence. The target task generation interval located in the middle of the duration sequence is extracted and used as the statistical quantile. The statistical quantile is determined as the corrected logistics cycle time for the target production node.

[0013] According to a production line logistics cycle time correction method provided by the present invention, the step of updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time includes: According to the preset time period division rules, obtain the actual logistics rhythm of the target production node in the current target time period; Calculate the absolute value of the deviation between the actual logistics cycle time and the current benchmark logistics cycle time, and calculate the percentage deviation between the absolute value of the deviation and the benchmark logistics cycle time; If the deviation percentage is greater than a preset deviation threshold, an operation is triggered to update the baseline logistics cycle time of the target production node within the target time period based on the corrected logistics cycle time.

[0014] According to a production line logistics cycle time correction method provided by the present invention, before inputting the effective task dataset into the target correction model for prediction, the effective task dataset is vectorized, specifically including: Extract multidimensional production factor features from the effective task dataset, the multidimensional production factor features including discrete features and continuous features; Convert the discrete features into binary sparse vectors; Map the continuous features to a preset numerical range; The input feature vector is generated by fusing the binary sparse vector and the mapped continuous features.

[0015] According to a production line logistics cycle time correction method provided by the present invention, the target correction model includes multiple independent regression sub-models, each of which contains multiple decision trees; the step of inputting the effective task dataset into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node includes: The input feature vectors are respectively input into the multiple independent regression sub-models; Within each regression sub-model, the basic predicted values ​​output by the multiple decision trees are obtained, and the internal average of all the basic predicted values ​​is calculated to obtain multiple independent predicted values ​​corresponding to each regression sub-model. Calculate the arithmetic mean of the multiple independent predicted values, and use the arithmetic mean as the corrected logistics cycle time of the target production node.

[0016] According to a production line logistics cycle time correction method provided by the present invention, after generating the input feature vector, it further includes: Calculate the mutual information entropy between each feature column in the input feature vector and the target predicted label; The feature columns are arranged in descending order according to the magnitude of the mutual information entropy. Redundant feature columns with mutual information entropy lower than a preset information threshold are removed to obtain the preferred feature matrix after dimensionality reduction; The preferred feature matrix is ​​used as the final input to the target correction model.

[0017] According to a production line logistics cycle time correction method provided by the present invention, the step of updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time includes: Obtain the preset smoothing weight coefficients; The first component is determined based on the modified logistics cycle time and the smoothing weight coefficient; The second component is determined based on the current benchmark logistics cycle time and the remaining weight coefficient, wherein the remaining weight coefficient is the difference between 1 and the smoothing weight coefficient; Based on the first component and the second component, the updated baseline logistics cycle time is obtained.

[0018] According to a production line logistics cycle time correction method provided by the present invention, before obtaining the preset smoothing weight coefficient, the method further includes: Obtain the percentage deviation between the corrected logistics cycle time and the current baseline logistics cycle time; The smoothing weight coefficient is dynamically adjusted based on the magnitude of the deviation percentage. The larger the percentage of deviation, the larger the value of the determined smoothing weight coefficient.

[0019] According to a production line logistics cycle time correction method provided by the present invention, before obtaining the historical task dataset of any target production node in the target production line, the method further includes: Obtain the raw task data stream; Remove empty task data, loop task data, and discontinuous task data with discontinuous task serial numbers from the original task data stream; The original task data stream after the removal process is normalized to a timestamp format to obtain the historical task dataset.

[0020] The present invention also provides a production line logistics cycle time correction device, comprising: The data acquisition unit is used to acquire the historical task dataset of any target production node in the target production line. The historical task dataset contains multiple historical task data and determines the task generation interval corresponding to each historical task data. The data filtering unit is used to measure the deviation of the task generation interval corresponding to each historical task data according to the baseline logistics cycle time of each historical task data, so as to filter the effective task dataset from the historical task dataset based on the measurement result. The cycle time determination unit is used to determine the corrected logistics cycle time of the target production node based on the statistical quantile of the effective task dataset when the effective sample size in the effective task dataset is less than a preset sample size threshold; otherwise, the effective task dataset is input into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node. The target correction model is obtained after training based on production factor samples and is used to characterize the nonlinear mapping relationship between multidimensional production factor characteristics and logistics cycle time. The baseline update unit is used to update the baseline logistics cycle time of the target production node based on the corrected logistics cycle time.

[0021] The present invention also provides 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 production line logistics cycle time correction method as described above.

[0022] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the production line logistics cycle time correction method as described above.

[0023] The production line logistics cycle time correction method, device, electronic equipment, and storage medium provided by this invention construct a dual filtering correction mechanism of data contamination measurement and cleaning, sample size assessment, and bi-branch prediction. First, it identifies and eliminates distorted data caused by non-production factors such as insufficient transportation capacity through deviation measurement, ensuring the purity of the data source. Then, in small sample scenarios, it automatically switches to a more robust statistical method, avoiding the risk of overfitting of the correction model. It effectively solves the problem of cycle time prediction distortion caused by abnormal logistics data contamination and small sample scenarios, and achieves high-precision and high-robust adaptive correction of production line material consumption cycle time. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating the production line logistics cycle time correction method provided by the present invention.

[0026] Figure 2 The present invention provides the cycle time values ​​of some production line logistics before and after the morning peak period correction.

[0027] Figure 3 These are the production line logistics cycle times before and after the morning production period correction provided by this invention.

[0028] Figure 4 These are the production line logistics cycle values ​​before and after the lunch break period correction provided by this invention.

[0029] Figure 5 These are the production line logistics cycle times before and after the afternoon production period correction provided by this invention.

[0030] Figure 6 These are the cycle times of some production line logistics before and after the evening shift correction provided by this invention.

[0031] Figure 7These are the cycle times of some production line logistics before and after nighttime periods provided by this invention.

[0032] Figure 8 This is a schematic diagram of the structure of a production line logistics cycle correction device provided by the present invention.

[0033] Figure 9 This is a schematic diagram of another production line logistics cycle correction device provided by the present invention.

[0034] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0036] It should be noted that, in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0037] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more.

[0038] This invention provides a production line logistics cycle time correction method, the execution entity of which can be a production line logistics cycle time correction device or an electronic device integrating a corresponding algorithm. In a specific embodiment, the method is executed by an industrial PC deployed on the production site, which interacts with the manufacturing execution system or equipment control system of the production line.

[0039] Figure 1This is a flowchart illustrating the production line logistics cycle time correction method provided by the present invention, as shown below. Figure 1 As shown, the main steps include, but are not limited to, the following: First, obtain the historical task dataset of any target production node in the target production line. The historical task dataset contains multiple historical task data, and determine the task generation interval corresponding to each historical task data.

[0040] In this embodiment, the target production line can be any production line that uses automated logistics equipment (such as AGVs) for material distribution, such as the final assembly line for home appliances. The target production node is a specific material consumption point or interaction point on the target production line. It can be an automated equipment station that automatically triggers material requisition tasks via a Programmable Logic Controller (PLC) signal, or it can be a manual station that requires production line workers to manually trigger material requisition tasks via a physical button box.

[0041] The historical task dataset is a collection of data containing multiple historical task data that occurred at the target production node within a preset time window (e.g., the past seven days). This historical task dataset can be obtained in various ways. For example, it can be extracted from the databases of the factory's Manufacturing Execution System (MES), Warehouse Management System (WMS), and / or logistics scheduling system via a data interface through timed polling or event-triggered extraction. Alternatively, in another optional embodiment, it can be obtained by directly parsing the underlying operation log files of the AGV scheduling system.

[0042] After obtaining the historical task dataset, multiple task generation intervals for the target production node can be extracted based on the timestamp information of the multiple historical task data. The task generation interval refers to the time interval between the current delivery task and the next delivery task, which directly reflects the actual material consumption rate and material calling cycle time of the production line workstation. It should be noted that the aforementioned delivery tasks can be generated by production line workers through manual buttons, photoelectric signals, or automatic material calling by a programmable logic controller, and are executed by logistics equipment such as automated guided vehicles and material bin robots.

[0043] For example, at the compressor installation station on a refrigerator assembly line, the PLC sends out the first material call signal (the first delivery task is generated) at 10:01:30 AM, and then sends out the second material call signal (the second delivery task is generated) at 10:03:45 AM. The interval between these two tasks is 135 seconds.

[0044] Next, based on the baseline logistics cycle time of each historical task data, the deviation measurement is performed on the task generation interval corresponding to each historical task data, so as to filter out the effective task dataset from the historical task dataset based on the measurement results.

[0045] In this embodiment, the logistics cycle time refers to the time interval for material flow (such as delivery, replenishment, and handling) set to match the production cycle time of the production line. The baseline logistics cycle time is a theoretical value preset for the target production node or updated in the previous correction cycle. It represents the standard replenishment cycle for the target production node to consume one unit of material (such as one pallet or one bin) under ideal, undisturbed conditions. Taking the compressor installation station mentioned above as an example, its baseline logistics cycle time may be preset to 120 seconds / pallet.

[0046] Deviation measurement is the process of quantitatively comparing the task generation interval of each historical task data with the aforementioned benchmark logistics cycle time. Its purpose is to assess whether the execution efficiency of the task is within the normal range.

[0047] The method of deviation measurement can be set according to actual needs. In one optional embodiment, the deviation measurement can be the ratio between the task generation interval and the baseline logistics cycle time; in another optional embodiment, it can be the absolute difference or percentage deviation between the two.

[0048] Based on deviation metrics, contaminated data caused by external disturbances unrelated to the material consumption rate of the workstation itself can be identified and eliminated. For example, if the task generation interval for a compressor delivery task is recorded as 800 seconds, far exceeding the baseline logistics cycle time of 120 seconds, deviation metrics reveal that the ratio or difference is extremely large. System log analysis shows that this abnormal delay is due to prolonged traffic congestion or queuing of the AGV on the main aisle, rather than a slowdown in production at that workstation. Failure to eliminate such data will seriously mislead subsequent cycle time adjustments. Therefore, the filtering process based on measurement results involves retaining historical task data with measurement results within a reasonable range and recombining them into a cleaner, more effective task dataset free of the aforementioned contaminated data.

[0049] Subsequently, this embodiment enters a conditional decision branch, specifically evaluating the effective sample size in the effective task dataset and comparing it with a preset sample size threshold. The preset sample size threshold is set to determine the most suitable correction method for both scenarios with sufficient and insufficient data, ensuring the robustness of the correction result. For example, this preset sample size threshold can be empirically set to 3000 records.

[0050] If the effective sample size in the effective task dataset is less than a preset sample size threshold, the corrected logistics cycle time of the target production node is determined based on the statistical quantiles of the effective task dataset.

[0051] Considering that complex models are prone to overfitting when the amount of data is small, while statistical methods are more robust, the statistical quantiles mentioned in this embodiment are statistics used to describe the distribution of data. For example, the median (i.e., the P50 quantile) can be used as this statistical quantile because it can effectively filter out a few extreme long-tail outliers in the dataset, such as interference from employees temporarily leaving their posts or temporary equipment debugging, thereby more realistically reflecting the typical consumption cycle of the target production node under normal operating conditions.

[0052] Otherwise, the effective task dataset is input into the target correction model for prediction to obtain the corrected logistics rhythm of the target production node. The target correction model is obtained after training based on production factor samples and is used to characterize the nonlinear mapping relationship between multidimensional production factor characteristics and logistics rhythm.

[0053] In this embodiment, when the number of valid samples in the effective task dataset is sufficient, a target correction model can be used to uncover deeper patterns. The target correction model can be a regression prediction model pre-trained on a large number of production factor samples, capable of representing the nonlinear mapping relationship between multidimensional production factor characteristics and logistics cycle time.

[0054] The production factor sample refers to data records that include historical task generation intervals and corresponding influencing factors. The multidimensional production factor characteristics can be time-related features (e.g., whether it's peak hours, weekends), spatial features (e.g., whether the workstation type is manual or automated), and business logic features (e.g., whether the delivery container type is pallet or bin, number of AGVs online). The nonlinear mapping relationship refers to the target correction model's ability to capture the complex, non-simple linear interactions between these features and their comprehensive impact on the final logistics cycle time.

[0055] Optionally, the aforementioned target correction model can be one of the following: Random Forest, Gradient Boosting Decision Tree, Support Vector Regression, or Deep Neural Network. In this embodiment, a valid task dataset containing multi-dimensional production factor characteristics is used as input, and the trained target correction model performs inference calculations. The output predicted value is called the corrected logistics cycle time.

[0056] Based on the corrected logistics cycle time, the baseline logistics cycle time of the target production node is updated.

[0057] In this embodiment, updating the baseline logistics cycle time is a closed-loop feedback step. Specifically, the corrected logistics cycle time obtained from the previous step, regardless of which branch it passed through, is used to update the currently stored original baseline logistics cycle time for use in the next correction cycle. The update method for the baseline logistics cycle time can be direct replacement, or a weighted average method can be used for smooth updates, that is, the newly obtained corrected logistics cycle time is weighted and summed with the old baseline logistics cycle time to prevent drastic fluctuations in the single correction result from impacting the system stability.

[0058] The production line logistics cycle time correction method provided by this invention ensures the authenticity and validity of the data source used for correction by pre-screening historical task datasets for contaminated data based on deviation metrics. Furthermore, by setting a sample size threshold, it adaptively allocates data to the most suitable correction strategy in both sparse and dense data scenarios, effectively avoiding the risk of overfitting with small samples. This dual filtering and adaptive correction mechanism enables the system to perform more accurate dynamic correction of material consumption cycle time at production nodes in complex industrial environments.

[0059] Based on the above embodiments, as an optional embodiment, the step of measuring the deviation of the task generation interval in each historical task data according to the baseline logistics cycle time of each historical task data, and filtering the effective task dataset from the historical task dataset based on the measurement results, may include, but is not limited to: First, the ratio of the task generation interval to the baseline logistics cycle time in each historical task data point is calculated as the measurement result. This calculation step is to convert the absolute value of the task generation interval into a standardized, dimensionless relative indicator, thereby facilitating unified threshold judgment.

[0060] Taking the compressor installation station on the aforementioned refrigerator assembly line as an example, its baseline logistics cycle time is assumed to be 120 seconds. If the task generation interval of a certain historical task data is 135 seconds, then the system calculates its corresponding metric result as: 135 ÷ 120 = 1.125. If the task generation interval of another historical task data is abnormally extended to 800 seconds due to logistics congestion, then its corresponding metric result is: 800 ÷ 120 ≈ 6.67.

[0061] Furthermore, if the measurement result is within a predetermined reasonable range, the corresponding historical task data is determined to be valid task data.

[0062] The reasonable range can be a preset numerical range used to define normal logistics and production fluctuations. As an optional embodiment, the upper and lower limits of the reasonable range can be set based on historical experience or expert knowledge, and can be dynamically adjusted according to the characteristics of different production lines or different time periods. For example, the reasonable range can be set to [0.5, 1.5].

[0063] This embodiment compares the measurement result calculated in the previous step with the reasonable range. Using the example above, the first measurement result of 1.125 falls within the range of [0.5, 1.5]. This indicates that although the 135-second task interval is slightly longer than the baseline logistics cycle time, it is still within the reasonable range caused by normal logistics cycle time fluctuations. Therefore, this historical task data can be considered valid task data. However, the second measurement result of 6.67 is much greater than the upper limit of the reasonable range of 1.5. This indicates that the excessively long task duration was caused by non-productive logistics anomalies, such as insufficient transport capacity, scheduling system failure, or severe traffic congestion. Therefore, this historical task data is distorted and cannot reflect the actual material consumption rate at the workstation. Consequently, this historical task data will not be considered valid task data.

[0064] Finally, the effective task dataset is constructed using all the aforementioned effective task data.

[0065] This embodiment iterates through every historical task data in the historical task dataset and repeats the above calculation and judgment process. Finally, it gathers all data records that are determined to be valid task data, thereby constructing a more clean valid task dataset that has undergone data pollution cleaning, for use in subsequent correction steps.

[0066] Through the above method, the present invention can establish a clear data availability judgment gate based on the quantitative multiple relationship between the task generation interval and the benchmark logistics cycle time. This not only effectively identifies and eliminates extreme outliers caused by systemic delays in the logistics system, but also allows for natural fluctuations in the logistics cycle time within a reasonable range, thereby significantly improving the quality and reliability of the data source when making subsequent cycle time corrections.

[0067] Based on the above embodiments, as an optional embodiment, since the operating rhythms of different production lines and workstations vary at different times, using a fixed interval threshold may not accurately identify true abnormal conditions. Therefore, the reasonable interval can be dynamically generated based on the following method: First, sort the multiple task generation intervals obtained during normal historical operation in ascending order of their numerical values.

[0068] In this embodiment, the historical normal operation period refers to a predetermined time period during which no major equipment failures or prolonged production line shutdowns or other extreme anomalies occur. Multiple task generation intervals corresponding to various logistics and delivery tasks that occurred within this time period can be extracted from the database and rearranged in ascending order of numerical values ​​to construct an ordered time series.

[0069] Next, obtain the first and third quartiles from the permutation results.

[0070] In the above ascending sequence, the first quartile, also known as the Q1 quartile in statistics, refers to the task generation interval located at the 25th percentile of the sequence; the third quartile, also known as the Q3 quartile in statistics, refers to the task generation interval located at the 75th percentile of the sequence. These two quartiles represent the boundaries of the concentrated distribution area of ​​the time consumption of the vast majority of routine logistics tasks.

[0071] Further, the interquartile range (IQR) between the first quartile and the third quartile is calculated. The IQR is a robust statistic for measuring the dispersion of data. Specifically, in this embodiment, the first quartile is subtracted from the obtained third quartile, and the difference between the two is taken as the IQR. The resulting IQR reflects the fluctuation range of task generation intervals in the middle 50%.

[0072] Finally, based on the first quartile, the third quartile, the interquartile range, and the preset product coefficient, the lower and upper limits of the reasonable interval are calculated.

[0073] Specifically, the preset product coefficient is an adjustable parameter used to control the strictness of outlier determination. As an optional embodiment, the preset product coefficient can be set to 1.5. In terms of specific calculation logic, since the measurement result calculated in the aforementioned steps is a ratio, the calculation process here can combine the time boundary after quantile conversion with the baseline logistics cycle time.

[0074] The following example illustrates the entire calculation and data flow.

[0075] Assume the baseline logistics cycle time for the target production node is 120 seconds. The system obtains multiple task generation intervals during historical normal operation and sorts them in ascending order, extracting the first quartile as 110 seconds and the third quartile as 130 seconds. The interquartile range is calculated to be 20 seconds. When calculating the lower limit of the reasonable range, the first quartile can be subtracted from the product of the interquartile range and a preset product coefficient (e.g., 1.5) to obtain a lower time boundary: 110 - 1.5 × 20 = 80 seconds. Then, this lower time boundary is divided by the baseline logistics cycle time of 120 seconds, thus calculating the lower limit of the reasonable range as 0.67.

[0076] Similarly, when calculating the upper limit of the reasonable range, the third quartile can be added to the product of the interquartile range and the preset product coefficient to obtain a time upper boundary: 130 + 1.5 × 20 = 160 seconds. This time upper boundary is then divided by the baseline logistics cycle time to calculate the upper limit of the reasonable range as 1.33. Thus, a reasonable range of [0.67, 1.33] can be dynamically generated.

[0077] By introducing a statistical algorithm based on quartiles and interquartile ranges to dynamically generate the reasonable range, this embodiment can adaptively characterize the normal fluctuation boundary with statistical significance according to the true probability distribution of the target production node's own historical operating data. This greatly enhances the scientific nature and anti-interference ability of identifying non-production abnormal pollution data. It not only makes the interception of extreme outliers more accurate, but also ensures that the reasonable range can dynamically evolve with changes in the production line environment, thereby improving the scenario adaptability and robustness of the entire correction system.

[0078] In an optional embodiment, in addition to the aforementioned implementation method of dynamically generating the reasonable interval based on quantiles and interquartile ranges, in order to adapt to production line scenarios with different data distribution characteristics, the reasonable interval can also be dynamically generated based on a normal distribution model of multiple task generation intervals obtained during historical normal operation.

[0079] Specifically, the average and standard deviation of multiple task generation intervals acquired during normal historical operation can be calculated first. Then, based on the normal distribution law in statistics, the average is subtracted from the standard deviation by a preset multiple (e.g., 3 times) to calculate the lower time boundary; the average is then added to the standard deviation by the preset multiple to calculate the upper time boundary. Finally, the lower and upper time boundaries are divided by the corresponding baseline logistics cycle time to calculate the lower and upper limits of the reasonable range.

[0080] As an optional implementation, the lower and upper limits of the reasonable range can be dynamically generated using formulas. Assume the first quartile is Q1, the third quartile is Q3, the interquartile range is IQR, and the preset product coefficient is k. First, calculate the interquartile range using the formula: IQR = Q3 - Q1. Then, calculate the boundaries of the reasonable range, which can be represented as [Q1 - k × IQR, Q3 + k × IQR]. The range calculated using this formula is the dynamically generated reasonable range. If the ratio of the task's interval to the baseline logistics cycle time falls within this range, it indicates normal logistics operation, and the data is considered usable. If it falls outside the range, it is considered contaminated data and discarded.

[0081] In another optional embodiment, in order to further meet the needs of refined logistics management in complex production line environments, the lower and upper limits of the reasonable range can also be adaptively adjusted based on the workstation type of the target production node or the material priority of the material currently requested for delivery by the target production node.

[0082] For example, attribute labels for each production node on the production line can be pre-stored in the database. Before performing deviation measurement, the attribute labels of the target production node are first read. If the target production node's workstation type is identified as a critical bottleneck workstation, meaning that the operating efficiency of this workstation directly determines the total capacity of the entire production line, or if the material priority of its requested delivery is the highest, the tolerance for logistics fluctuations at that target production node will be automatically reduced. In this case, the span of the reasonable range can be dynamically narrowed, for example, setting the reasonable range to [0.8, 1.2], to ensure that any minor abnormal delay data can be strictly identified and removed as contaminated data.

[0083] Conversely, if the target production node's workstation type is identified as a non-critical workstation, such as an auxiliary material addition workstation or an empty pallet recycling workstation in a non-main production flow, or if the material priority of its requested delivery is low, the tolerance will be increased to avoid excessive cleaning leading to a sharp decrease in the sample size of the effective task dataset. In this case, the span of the reasonable interval can be dynamically expanded, for example, the reasonable interval can be set to [0.4, 2.0].

[0084] The dynamic generation method based on normal distribution and the adaptive adjustment method based on workstation type or material priority provided in this embodiment enable the determination of reasonable intervals to be rationally selected based on actual production needs. It is no longer limited to a single pure mathematical statistical method, but can deeply integrate the importance of the physical topology of the production site and the priority of production scheduling. This not only greatly reduces the risk of takt prediction distortion caused by abnormal data infiltration in key core workstations and ensures the absolute accuracy of main production line scheduling, but also gives non-critical auxiliary workstations higher fault tolerance flexibility, making the overall data screening mechanism more in line with the complex and ever-changing flexible manufacturing needs of actual industrial sites.

[0085] In another embodiment of the present invention, after updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time, the method further includes: obtaining an online task data sequence containing multiple online task data, performing time window sliding processing on the online task data sequence using a preset step size to obtain multiple sliding time windows; and within each sliding time window, performing quantity statistics on online task data whose measurement results are not within a reasonable range to obtain the proportion of abnormal data.

[0086] Abnormal data percentage refers to the proportion of tasks within the currently selected sliding time window whose task generation interval does not fall within the reasonable range of the interquartile range.

[0087] For example, an online task data sequence can be a batch of new automated guided vehicle (AGV) logistics handling records generated during the online operation phase. Processing the sequence using a preset step size can be achieved by sliding the time window backward once each time according to a fixed number of data entries, thereby continuously acquiring the latest batch of production line material requisition records.

[0088] If the proportion of abnormal data within a consecutive preset number of sliding time windows is greater than a preset proportion threshold, a distribution offset signal is triggered; the target correction model is used to predict the beat of each online task data to obtain the prediction error corresponding to each online task data; the mean of each prediction error within the same sliding time window is calculated to obtain the mean prediction error; if the mean prediction error within a consecutive preset number of sliding time windows is greater than a preset error threshold, a model error signal is triggered.

[0089] The distribution offset signal is a notification indicating that new data characteristics generated during the current production line operation have deviated from the historical data characteristics used when the target correction model was established. For example, the distribution offset signal can be a Boolean variable or a status flag. When the production line speeds up overall production due to temporary production acceleration or product model change, the reasonable range calculated using historical data is no longer applicable to the newly inflowing data. As long as the proportion of abnormal data calculated for three consecutive sliding time windows exceeds a preset threshold of 30%, the status of this flag will change, thus triggering the distribution offset signal.

[0090] Model error signals are used to indicate that the currently used target correction model is inaccurate in predicting new incoming tasks. For example, prediction error can be the absolute difference between the predicted cycle time value output by the target correction model and the actual cycle time value of the task. Suppose the target correction model predicts that a certain workstation will take 120 seconds, but the task actually takes 150 seconds, the prediction error for this task record is 30 seconds. Model error signals can also be a status flag. By adding up the prediction errors of all online task data within the same sliding time window and dividing by the total number of tasks to calculate the average, if the average prediction error calculated for three consecutive sliding time windows is greater than a preset 20-second error threshold, it indicates that the current model's ability to predict real-time cycle time has decreased, thus triggering the model error signal.

[0091] Specifically, after being put into online operation, these two signals will be maintained in parallel, forming a dual-signal joint monitoring mechanism. Calculating and judging the data using two independent signals is to decouple the increase in error from the distribution shift, in order to distinguish whether it is a genuine change in production status or simply a data quality issue. For example, in a complex actual workshop environment, if only the single case of increased model prediction error is considered, it is very likely due to packet loss in the underlying network causing the loss of some features, or the underlying log records encountering occasional random noise. Using a single signal-driven mechanism could easily lead to incorrect time periods and model segmentation. Now, by simultaneously calculating the proportion of abnormal data and the mean prediction error within the same sliding time window, using these two dimensions as the basis for subsequent status judgment, we can filter out short-term data disturbances and achieve adaptive perception of continuous dynamic changes in production status.

[0092] If the distribution offset signal and the model error signal are triggered simultaneously, the online task data is reacquired, and the effective task dataset is filtered and the model is reconstructed based on the reacquired online task data to obtain the corrected logistics cycle time for the next production period. If only the model error signal is triggered and the distribution offset signal is not triggered, the deviation measurement and validity determination of the online task data are performed again to obtain the target task dataset.

[0093] Specifically, the joint judgment logic of the two signals can identify whether the production line has experienced a real change in production status or a temporary data anomaly. When the distribution offset signal and the model error signal are triggered simultaneously, it is confirmed that a change in online production status has occurred.

[0094] For example, if a production line was originally producing an older product but switched to producing a new one, and the model's average prediction error exceeded 20 seconds for several consecutive seconds, while the percentage of outlier data also exceeded 30%, then the current production cycle would be terminated. Next, a new batch of online task data generated during this period would be acquired, and contaminated data caused by queuing or waiting would be removed using the aforementioned deviation measurement method. The clean, valid task data would then be used to re-evaluate the sample size and retrain the random forest model, thereby establishing a model suitable for the new product's production rhythm and obtaining the corrected logistics cycle time for the next production period.

[0095] If only the model error signal is triggered and the distribution offset signal is not, it means that the overall numerical distribution of the newly arrived batch of task records is still within a reasonable range, the actual production rhythm of the production line has not changed, and the reason for inaccurate prediction is determined to be a data quality problem or missing features. For example, if the proportion of abnormal data is only 15%, and the distribution offset signal is not triggered, but the average prediction error reaches 25 seconds, the model error signal is triggered. In this case, instead of splitting into new time periods or rebuilding the model, the data quality self-check process is triggered. The relationship between the task generation interval and the baseline logistics cycle time will be recalculated for this batch of online task data. The data will be cleaned and its validity will be judged again through a reasonable range, and erroneous data caused by occasional network jitter or sensor failure will be removed to obtain a cleaner target task dataset.

[0096] It's important to note that after obtaining the corrected logistics takt time, a revised logistics takt time report can be generated and output. This report may include the adaptive time period identifier, container type, the original logistics takt time, and the revised logistics takt time. The report will also record the correction magnitude, which is the difference between the revised logistics takt time and the original baseline logistics takt time. Recording this information provides data for the backend to retain, facilitating subsequent tracking and performance evaluation of logistics takt times under different production conditions.

[0097] The production line logistics cycle time correction method provided by this invention decouples error increase from distribution offset through a joint determination method, forming a closed loop that includes initial time period division, modeling, dual signal monitoring, status confirmation, re-division, and re-modeling. This allows for automatic updates to time periods and model reconstruction during production line changes or accelerations, while avoiding erroneous cuts caused by accidental random noise or temporary channel congestion, thus maintaining the stability of the time period division and cycle time prediction process.

[0098] Production status transition candidate points refer to the points on the historical data timeline where the actual production rhythm of a production line may change. Because the new status is not yet fully stable, these points are considered as candidate nodes awaiting confirmation. For example, a production status transition candidate point could be the specific timestamp when an assembly line switches from producing standard-sized refrigerators to producing large-capacity refrigerators midway through the morning shift. After this, the original material consumption rate changes, causing a large number of subsequent tasks to deviate from their original normal distribution range.

[0099] Specifically, historical data was previously divided by manually setting fixed morning, noon, and night shifts. However, fixed shifts cannot detect real rhythm changes that cross shift boundaries, such as mid-shift changes or accelerated production, easily mixing data from different production states. By reusing the reasonable intervals established earlier based on interquartile ranges, a time window is slid across the historical task data sequence with a fixed step size, for example, sliding 50 task records each time. Each slide picks out historical data that does not fall within the reasonable interval and divides it by the total amount of data in the window to calculate the percentage of data records.

[0100] For example, if the proportion of data records calculated by the current historical sliding time window reaches 40%, which is greater than the preset proportion threshold of 30%, and the proportion calculated by the three consecutive historical sliding time windows exceeds this threshold, it means that the increase in data deviating from the reasonable range is not a coincidence, but rather that the overall logistics rhythm of the production line has shifted. At this time, the current sliding position will be designated as a candidate point for production status change.

[0101] Based on the candidate points of production state transition, a time window sliding process is performed. If the proportion of data records within a consecutive second preset number of historical sliding time windows is not greater than a preset proportion threshold, the starting historical task data within the first historical sliding time window of the consecutive second preset number of historical sliding time windows is extracted to obtain the modeling starting point. Based on the modeling starting point, the effective task dataset is divided into time periods to obtain multiple independent initial adaptive production time periods. The effective task data within each initial adaptive production time period is subjected to cycle time correction processing to obtain the corrected logistics cycle time corresponding to each initial adaptive production time period.

[0102] Specifically, after identifying candidate points for production status transitions, it indicates that the production line is in a transitional period of model change or adjustment. During this time, data fluctuations are significant and patterns are inconsistent, making it unsuitable for direct model building. Therefore, monitoring continues along the historical task data sequence, sliding forward through time windows until the data record percentage falls back down. If the data record percentage calculated over several consecutive historical sliding time windows is no greater than a preset threshold, it indicates that the transitional adjustment period of the production line has ended, and the new production rhythm has fully stabilized. At this point, the first historical task data point within the first historical sliding time window of this stable period is extracted and used as the starting point for the next stage of modeling.

[0103] For example, an appliance production line starts changing production models at 10:00 AM. An abnormal increase in data at this time triggers a candidate point for a production status transition. After the model change is completed and the production line begins operating according to the new logistics rhythm, after a period of time, the proportion of data records calculated over three consecutive historical sliding time windows drops to 20%, not exceeding the preset 30% threshold. This indicates that from the starting point of these three windows, the production line has entered a new normalized production phase. Subsequently, the initial historical task data of this first window is used as the starting point for modeling. This mechanism automatically divides the originally continuous long string of historical records into multiple independent initial adaptive production periods, completely replacing the traditional fixed morning, noon, and evening shifts that rely on manual configuration. For the valid task data in each independent period, the subsequent rhythm correction process is executed separately to calculate the corresponding corrected logistics rhythm, thus ensuring that data from different production states are not mixed in the modeling and that the logistics rhythm estimates for different periods do not interfere with each other.

[0104] The production line logistics cycle time correction method provided by this invention automatically extracts the modeling starting point at the position where the proportion of data records continuously exceeds a preset threshold and then falls back to a stable position. Then, it divides the effective task dataset into time periods. This can get rid of the limitations of traditional data division based on manually preset fixed shifts. It can automatically sense the real production rhythm changes such as mid-production line changes or production acceleration, ensuring that the logistics cycle time estimation of each time period does not interfere with each other, and improving the accuracy of the logistics cycle time correction results.

[0105] In another embodiment of the present invention, the effective task data within each initial adaptive production period is subjected to cycle time correction processing to obtain the corrected logistics cycle time corresponding to each initial adaptive production period. This includes: for any initial adaptive production period, sorting the task generation intervals in the effective task data in ascending order and calculating the quantiles to obtain a reference logistics cycle time; determining the time period deviation of the current period based on the reference logistics cycle time and the current benchmark logistics cycle time of the target production node; if the time period deviation of the current period is greater than a preset deviation threshold, then performing cycle time correction processing on the effective task data of the current period to obtain the corrected logistics cycle time of the current period; if the time period deviation of the current period is less than or equal to the preset deviation threshold, then determining the current benchmark logistics cycle time as the corrected logistics cycle time of the current period.

[0106] Reference logistics takt time refers to a typical statistical value used to represent the normal material consumption rhythm of a production line within a specific time period. For example, the reference logistics takt time could be the 50th percentile, or median, extracted after sorting the intervals of all effective tasks in the current time period from smallest to largest.

[0107] Time-of-use deviation refers to the degree of deviation between the actual reference logistics cycle time and the pre-set baseline logistics cycle time. For example, time-of-use deviation can be calculated as the percentage difference between the reference logistics cycle time and the baseline logistics cycle time, calculated by subtracting the baseline logistics cycle time from the reference cycle time.

[0108] Specifically, after segmenting multiple independent initial adaptive production periods, deviation identification is performed separately for each period to avoid blind intervention. For a specific segmented period, all valid task generation intervals are first sorted in ascending order of numerical value, and the 25th, 50th, and 75th quantiles are calculated. The 50th quantile is selected as the reference logistics cycle time because the median can better filter out the interference of individual long-tail outliers compared to the arithmetic mean.

[0109] For example, if the current baseline logistics cycle time for a certain workstation is 120 seconds, and the 50th percentile calculated for this time period is 135 seconds, the absolute difference between the two is 15 seconds. Dividing 15 seconds by 120 seconds yields a time period deviation of 12.5%. Assuming a pre-set deviation threshold of 10%, since 12.5% ​​is greater than 10%, this indicates that the material consumption cycle time for this time period has indeed drifted. Therefore, it is determined that the current time period needs to be corrected. Subsequently, subsequent model predictions or small sample statistical processes are performed on the valid task data within this time period to calculate the true corrected logistics cycle time for the current time period.

[0110] If the calculated time-period deviation is only 5%, which is less than the preset deviation threshold of 10%, it means that the production rhythm fluctuation during this period is within the normal range of minor adjustments. Therefore, no further correction calculations will be made, and the original 120-second baseline logistics cycle time will be used as the corrected logistics cycle time for this period. By using this hierarchical statistical and deviation identification logic, unnecessary corrections can be avoided for production lines or time periods with small deviations, improving the targeting of correction actions and the overall execution efficiency.

[0111] The production line logistics cycle time correction method provided by this invention determines whether the time deviation between the reference logistics cycle time and the benchmark logistics cycle time exceeds a preset deviation threshold. Subsequent cycle time correction processing is only performed when the time deviation is greater than the threshold. This avoids blind intervention in normal periods with small fluctuations in production rhythm, reduces unnecessary model calculations and system update actions, and improves the pertinence and overall execution efficiency of cycle time correction.

[0112] Based on the above embodiments, as an optional embodiment, when the effective sample size in the effective task dataset is less than a preset sample size threshold, determining the corrected logistics cycle time of the target production node based on the statistical quantiles of the effective task dataset may include, but is not limited to: If the effective sample size is less than the preset sample size threshold, the intervals of all tasks in the effective task dataset are sorted in ascending order of their numerical values ​​to obtain a duration sequence.

[0113] For example, the preset sample size threshold is set to 3000. If the system counts that the number of valid samples in the currently acquired valid task dataset is only 150 (less than 3000), it will extract the task generation intervals corresponding to these 150 valid task data and reorder them from the minimum to the maximum value, thereby generating a duration sequence consisting of 150 values.

[0114] Next, the target task generation interval located in the middle of the duration sequence is extracted as the statistical quantile.

[0115] Specifically, the statistical quantile here is represented by the P50 quantile (i.e., the median). Optionally, during the extraction process, if the effective sample size is odd, the middle task generation interval in the time sequence is directly extracted as the target task generation interval; if the effective sample size is even, the two middle task generation intervals in the time sequence are extracted, and their arithmetic mean is calculated, which is then used as the target task generation interval. Subsequently, the extracted target task generation interval is directly assigned to the statistical quantile.

[0116] Finally, the statistical quantiles are determined as the corrected logistics cycle time for the target production node.

[0117] Taking the sequence of the above 150 data points as an example, the task generation intervals at the 75th and 76th positions are 130 seconds and 132 seconds respectively. The arithmetic mean of these two intervals, 131 seconds, is calculated as the target task generation interval, and 131 seconds is directly determined as the corrected logistics cycle time.

[0118] Based on the above embodiments, as an optional embodiment, updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time may include, but is not limited to: According to the preset time period division rules, the actual logistics rhythm of the target production node within the current target time period is obtained.

[0119] Continuing with the example of the compressor installation station, assuming that the current baseline logistics cycle time of the target production node is 120 seconds, and the corrected logistics cycle time obtained through the aforementioned steps is 131 seconds, the system performs subtraction or absolute value operation to calculate the difference between 131 seconds and 120 seconds, obtaining the absolute value of the deviation as 11 seconds.

[0120] Next, the absolute value of the deviation between the actual logistics cycle time and the current benchmark logistics cycle time is calculated, and the percentage deviation between the absolute value of the deviation and the benchmark logistics cycle time is calculated.

[0121] For example, in this embodiment, the absolute value of the deviation of 11 seconds obtained in the previous step is divided by the current benchmark logistics cycle time of 120 seconds to calculate the deviation percentage of approximately 9.17%. This deviation percentage objectively quantifies the severity of the deviation of the actual material consumption cycle time from the theoretical design benchmark.

[0122] Finally, if the deviation percentage is greater than a preset deviation threshold, an operation is triggered to update the baseline logistics takt of the target production node within the target time period based on the corrected logistics takt.

[0123] In this embodiment, the preset deviation threshold is a warning line used to determine whether the beat drift has reached a level requiring intervention. Optionally, the preset deviation threshold can be set to 10% or 15%.

[0124] If the calculated deviation percentage is 9.17%, and the preset deviation threshold is set to 10%, since 9.17% is not greater than the preset deviation threshold of 10%, it can be determined that the current cycle fluctuation is a small normal drift, thus not triggering the update operation and maintaining the original baseline logistics cycle unchanged.

[0125] If, in another time period, the calculated deviation percentage reaches 12.5%, which is greater than the preset deviation threshold of 10%, then the deviation can be determined to be significant, thereby triggering the subsequent update write-back operation.

[0126] To further improve the precision of the correction, this embodiment can also independently execute the above triggering logic in conjunction with different production periods. For example... Figures 2 to 7 As shown, logistics cycle time can be statistically analyzed by time period and the logistics cycle time values ​​before and after correction can be compared for different production lines in each time period.

[0127] in, Figure 2 This invention provides the corrected cycle times for some production line logistics before and after the morning peak period. Figure 3 This invention provides the cycle time values ​​for some production line logistics before and after the morning production period correction. Figure 4 This invention provides the cycle time values ​​of some production line logistics before and after the lunch break period correction. Figure 5 These are the production line logistics cycle times before and after the afternoon production period provided by this invention. Figure 6 This invention provides the corrected production line logistics cycle time values ​​before and after the evening shift. Figure 7 These are the cycle times of some production line logistics before and after nighttime periods provided by this invention.

[0128] Figures 2 to 7 The differences between the corrected logistics cycle time of each production line and the original baseline logistics cycle time are shown for the morning peak, morning production, lunch break, afternoon production, evening shift, and night shift, respectively. This provides a clear understanding of the correction comparison results. By independently calculating the deviation percentage for different time periods, precise positioning of specific time periods and specific production lines can be achieved.

[0129] This invention constructs an anti-jitter mechanism that does not intervene in non-significant deviations by calculating the absolute value and percentage of the deviation and setting thresholds for interception triggers. This effectively avoids frequent updates and wasted computing power in the underlying manufacturing execution system caused by minor or occasional fluctuations in the production line's cycle time, thus improving the stability of the entire factory logistics scheduling system. Simultaneously, combined with precise identification at the hierarchical level, it ensures that limited computing power can be concentrated on correcting production nodes that experience real and significant drift, as well as production lines and time periods composed of multiple production nodes, significantly improving the targeting and execution efficiency of correction actions.

[0130] Based on the above embodiments, as an optional embodiment, before inputting the effective task dataset into the target correction model for prediction, the effective task dataset is vectorized, specifically including: First, multidimensional production factor features are extracted from the effective task dataset. These multidimensional production factor features include discrete features and continuous features.

[0131] This embodiment parses the context information of each record in the valid task dataset. Among them, discrete features usually refer to state attributes that have finite values ​​and do not have a continuous mathematical relationship. For example, container type features (such as bins, pallets, and buckets), workstation type features (such as manual button box workstations and PLC automatic material calling workstations), and time cycle rhythm features (such as morning shift, afternoon shift, night shift, or Monday to Sunday) are extracted by parsing task data.

[0132] The continuous feature refers to a variable that has continuous numerical attributes and whose magnitude has actual physical meaning, such as the number of AGVs online on the production line when the task occurs, the current load rate of the production line, or the number of temporary workstations at the target production node.

[0133] Next, the discrete features are converted into binary sparse vectors.

[0134] To enable subsequent models to understand the aforementioned classification states, the system digitally encodes the discrete features. As an optional implementation, one-hot encoding can be used for this conversion. For example, if the container type includes three states: "bin," "pallet," and "bucket," when a data point corresponds to the container type "pallet," the system converts it into a binary sparse vector such as [0,1,0]. This vector contains only one 1 element, with the rest being 0. This sparse representation preserves category information while avoiding introducing false numerical relationships into the model, such as preventing the model from mistakenly considering "pallet" to be greater than "bin."

[0135] Next, the continuous features are mapped to a preset numerical range. Since the physical dimensions and numerical ranges of different continuous features vary greatly—for example, the number of AGVs online may be dozens, while the production line load rate may be between 0 and 1—direct calculation would cause large numerical features to mask the effect of small numerical features. Therefore, this embodiment performs dimensionless processing. In one optional embodiment, the system can use a min-max scaling algorithm to scale the continuous features proportionally and map them to the preset numerical range [0, 1]. In another optional embodiment, the system can also use the Z-score normalization method to map them to a distribution range with a mean of 0 and a variance of 1.

[0136] Finally, the binary sparse vector and the mapped continuous features are fused to generate the input feature vector. In this embodiment, all the transformed binary sparse vectors are concatenated bit-by-bit with all the mapped and scaled continuous feature values ​​to reconstruct the initial input feature vector. It should be noted that in an optional embodiment without feature filtering, this initial input feature vector can be directly used as input to the subsequent target correction model; while in an optional embodiment that further includes dimensionality reduction processing, the initial input feature vector will also undergo subsequent mutual information entropy filtering to remove redundant features.

[0137] In an optional embodiment, the target correction model is specifically implemented as an ensemble learning model based on the RandomForest algorithm, which includes multiple independent regression sub-models. Based on this, inputting the effective task dataset into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node may include, but is not limited to: The input feature vectors are respectively input into the multiple independent regression sub-models; Within each regression sub-model, the basic predicted values ​​output by the multiple decision trees are obtained, and the internal average of all the basic predicted values ​​is calculated to obtain multiple independent predicted values ​​corresponding to each regression sub-model. Calculate the arithmetic mean of the multiple independent predicted values, and use the arithmetic mean as the corrected logistics cycle time of the target production node.

[0138] Specifically, in this embodiment, the input feature vector generated in the previous step, which contains multi-dimensional information such as time period, container type, and number of online AGVs, is distributed in parallel to each independent regression sub-model within the target correction model. This parallel input method enables each regression sub-model to perform independent inference calculations based on the same current operating condition data.

[0139] Within any of the aforementioned regression sub-models, since it contains multiple decision trees, each decision tree will make layer-by-layer judgments on the input feature vector based on the nonlinear bifurcation rules it learned during the training phase, and finally output a basic prediction value at its leaf node.

[0140] Subsequently, the regression sub-model sums and averages the multiple base predictions output by all its internal decision trees to obtain an independent prediction that represents the overall judgment conclusion of the regression sub-model. Thus, multiple independent predictions can be collected from multiple regression sub-models.

[0141] Finally, all the collected independent predictions are summed and divided by the total number of regression sub-models to obtain the final arithmetic mean. This arithmetic mean is then assigned as the output target production node to correct the logistics cycle time.

[0142] This invention employs a random forest architecture comprising multiple regression sub-models and decision trees for prediction, fully leveraging the principles of ensemble learning. Through a non-linear, tree-like branching structure, it significantly enhances the ability to capture the combined effects of complex, multi-dimensional features, such as the interplay between night shifts and the overlapping of material bins and containers on cycle time. Simultaneously, a two-stage smoothing process—internal averaging and external arithmetic averaging—effectively offsets the prediction variance and overfitting tendency that may exist with a single decision tree. This results in a final output of highly accurate logistics cycle time predictions, while also exhibiting high robustness in the face of complex and ever-changing production line dynamics.

[0143] In a specific embodiment, to further improve the training and inference efficiency of the target correction model and prevent overfitting risks caused by noisy features in complex industrial field data, after generating the aforementioned input feature vector, the following operations can be further performed: Calculate the mutual information entropy between each feature column in the input feature vector and the target predicted label; The feature columns are arranged in descending order according to the magnitude of the mutual information entropy. Redundant feature columns with mutual information entropy lower than a preset information threshold are removed to obtain the preferred feature matrix after dimensionality reduction; The preferred feature matrix is ​​used as the final input to the target correction model.

[0144] In this context, each feature column in the input feature vector represents a multi-dimensional feature attribute, such as time period, container type, and number of AGVs online, after the aforementioned encoding and mapping processing. The target prediction label refers to the actual physical results recorded in historical data, such as the actual task generation interval or the actual production line cycle time. The mutual information entropy is a statistical indicator used to measure the non-linear dependency and information sharing degree between two random variables. This embodiment calculates the mutual information entropy between each feature column and the target prediction label one by one using relevant algorithms to quantify the actual influence of each feature on the final cycle time fluctuation.

[0145] Furthermore, all the calculated mutual information entropy values ​​can be reordered from largest to smallest. The larger the mutual information entropy value and the higher the ranking, the more significant the influence of the corresponding feature column on the target prediction label and the stronger the information gain it can provide; conversely, the lower the ranking, the weaker the correlation between the feature and the actual logistics rhythm.

[0146] A preset information threshold can be set in advance (e.g., 0.05 or 0.1). Then, the feature set after descending order is traversed. If it is found that the mutual information entropy corresponding to some feature columns is less than the preset information threshold, such as the ambient humidity, AGV body color or status code of unrelated workstations that are unintentionally mixed in the data acquisition system, these feature columns are determined to have no substantial contribution to the cycle prediction and belong to redundant feature columns.

[0147] Next, all redundant feature columns will be directly pruned and deleted from the original feature library, and only the key core features that are greater than or equal to the preset information threshold will be retained. These retained core features will be combined to form the dimensionality-reduced preferred feature matrix.

[0148] Finally, the refined, lower-dimensional, and higher-information-density optimized feature matrix is ​​passed to the subsequent target correction model, such as a random forest, for training or inference prediction.

[0149] This invention implements feature filtering and dimensionality reduction by introducing mutual information entropy, which can automatically identify and remove irrelevant interference items and weakly correlated features mixed in the massive data of industrial sites. This not only significantly reduces the dimensionality distribution and computational complexity of the input data of the target correction model, but also effectively eliminates the overfitting phenomenon caused by the target correction model learning invalid features, thereby improving the prediction accuracy and generalization ability in complex and ever-changing actual production line environments.

[0150] Based on the above embodiments, as an optional embodiment, updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time may specifically include: Obtain the preset smoothing weight coefficients; The first component is determined based on the modified logistics cycle time and the smoothing weight coefficient; The second component is determined based on the current benchmark logistics cycle time and the remaining weight coefficient, wherein the remaining weight coefficient is the difference between 1 and the smoothing weight coefficient; Based on the first component and the second component, the updated baseline logistics cycle time is obtained.

[0151] The smoothing weight coefficient can be a value between 0 and 1, primarily used to measure the confidence level in the latest prediction during the update process. This smoothing weight coefficient can be an empirically fixed value preset by the system administrator, such as 0.3.

[0152] Next, this embodiment will multiply the corrected logistics cycle time output from the aforementioned steps with the smoothing weight coefficient. For example, if the corrected logistics cycle time is 131 seconds and the smoothing weight coefficient is 0.3, then the first component is calculated to be 39.3 seconds. This first component represents the proportion of the latest correction result in the final updated value.

[0153] Furthermore, the remaining weight coefficient can be calculated first, which is 1 minus 0.3 to get 0.7. Then, the current baseline logistics cycle time of 120 seconds is multiplied by the remaining weight coefficient of 0.7 to calculate the second component as 84 seconds. The resulting second component represents the proportion of the historical baseline retained in the final updated value.

[0154] Finally, the calculated first component and the second component are summed: 39.3 seconds + 84 seconds = 123.3 seconds. This sum is used as the updated baseline logistics cycle time and then written back to the scheduling system.

[0155] As an optional embodiment, when updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time, the final result can be calculated using a smooth weighted update formula. The specific update formula is as follows: ; in, This indicates the updated baseline logistics cycle time. This represents the corrected logistics cycle time obtained from this calculation or model prediction. This indicates the current baseline logistics cycle time of the target production node. This represents the smoothing weighting coefficient, which is dynamically adjusted according to the percentage of deviation. The larger the percentage of deviation, the smoother the smoothing. The larger the value, the faster the benchmark responds to actual changes in the logistics cycle time; the smaller the percentage deviation, the better. The smaller the value, the smoother the fine-tuning, and the less drastic the benchmark fluctuations.

[0156] In another optional embodiment, regarding the process of obtaining the preset smoothing weight coefficients, since fixed weight coefficients cannot flexibly cope with different degrees of cycle time drift, a dynamic adjustment mechanism is introduced into the provided production line logistics cycle time correction method, specifically including: The percentage deviation between the corrected logistics cycle time and the current baseline logistics cycle time is obtained. As described in the previous embodiment, this embodiment calculates the relative magnitude of the difference between the two. For example, the current deviation percentage is calculated to be 15%.

[0157] Next, the smoothing weight coefficient is dynamically adjusted according to the magnitude of the deviation percentage. The specific logic is as follows: the larger the deviation percentage, the larger the value of the determined smoothing weight coefficient.

[0158] In this embodiment, a mapping table or a piecewise function can be preset. For example, when the deviation percentage is in the range of [10%, 20%), it indicates that the cycle time has deviated moderately, and the smoothing weight coefficient can be dynamically adjusted to 0.4; when the deviation percentage is greater than or equal to 20%, it indicates that the production line has experienced a major change in operating conditions, such as changing molds or switching product models, and the cycle time has deviated drastically. At this time, a new cycle time is urgently needed to guide the scheduling, so the smoothing weight coefficient will be dynamically adjusted to a larger value, such as 0.7 or 0.8.

[0159] In this embodiment of the invention, the smoothing weight coefficient is dynamically adjusted according to the percentage of deviation. When the cycle drift is small, more historical benchmarks are retained. When the cycle changes drastically, the weight of the newly corrected cycle is quickly assigned. This ensures scheduling stability while significantly improving the response speed to major production line condition changes.

[0160] In a specific embodiment, to ensure the quality of the input data source, the provided production line logistics cycle time correction method may further include the following data pre-cleaning steps before obtaining the historical task dataset of any target production node in the target production line: First, the raw task data stream is obtained, specifically by continuously receiving the unfiltered raw task data stream directly from the underlying device's operation logs or sensor interfaces.

[0161] Next, empty task data, cyclic task data, and discontinuous task data with discontinuous task serial numbers in the original task data stream are identified and removed.

[0162] In actual industrial settings, underlying information systems may generate logical error records due to network jitter or packet loss. This embodiment parses the original task data stream. If a record is found to have the same task start time and end time, it is determined to be empty task data. If a task end time is found to be earlier than the task start time on the timeline, it is determined to be cyclic task data with timestamp inversion. If two adjacent task numbers have a large gap, such as jumping directly from Task_001 to Task_105, it is determined to be discontinuous task data with discontinuous task serial numbers. The above three types of data that violate basic physical and business logic can be directly eliminated and not retained.

[0163] Finally, the original task data stream after the removal process is normalized to a timestamp format to obtain the historical task dataset. Since the time formats reported by different subsystems (such as WMS and AGV scheduling systems) may be inconsistent—for example, strings containing year, month, day, hour, minute, and second, or millisecond-level integers—all valid records can be uniformly converted and mapped to a standard second-level timestamp format, thus ultimately outputting a unified and logically correct historical task dataset.

[0164] This invention, through regular pre-processing of anomaly removal and format normalization on the underlying raw data, effectively prevents dirty data pollution caused by pure logical errors in the underlying system before entering the complex multiple deviation measurement and model prediction, thus reducing the computational burden on subsequent core algorithms.

[0165] Figure 8 This is a schematic diagram of the structure of a production line logistics cycle time correction device provided by the present invention, as shown below. Figure 8 As shown, it mainly includes, but is not limited to: Data acquisition unit 11 is used to acquire the historical task dataset of any target production node in the target production line. The historical task dataset contains multiple historical task data and determines the task generation interval corresponding to each historical task data. The data filtering unit 12 is used to measure the deviation of the task generation interval corresponding to each historical task data according to the baseline logistics cycle time of each historical task data, so as to filter out the effective task dataset from the historical task dataset based on the measurement result. The cycle time determination unit 13 is used to determine the corrected logistics cycle time of the target production node based on the statistical quantile of the effective task dataset when the effective sample size in the effective task dataset is less than a preset sample size threshold; otherwise, the effective task dataset is input into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node. The target correction model is obtained after training based on production factor samples and is used to characterize the nonlinear mapping relationship between multidimensional production factor characteristics and logistics cycle time. The baseline update unit 14 is used to update the baseline logistics cycle time of the target production node based on the corrected logistics cycle time.

[0166] It should be noted that the production line logistics cycle time correction device provided in the embodiments of the present invention can execute the production line logistics cycle time correction method provided in any of the above embodiments during specific operation, which will not be described in detail here.

[0167] Figure 9 This is a schematic diagram of another production line logistics cycle time correction device provided by the present invention, as shown below. Figure 9As shown, as an optional embodiment, its specific execution process may include the following stages: First, the upstream data sources of the system mainly include historical AGV task logs and current production line business logic (including baseline logistics cycle time). The system continuously captures these raw materials from the underlying scheduling system and manufacturing execution system.

[0168] Subsequently, the data flow enters the data preparation and preprocessing module. In this module, the rules for determining the production line monitoring workstations are first established; then, the statistical criteria for logistics cycle time are designed, and preliminary cleaning and timestamp normalization are performed on underlying empty tasks and recurring tasks. After processing, this data preparation and preprocessing module outputs structured production event sequence data, namely the aforementioned historical task dataset used for subsequent analysis.

[0169] Next, the structured data flows into the data availability assessment module. In this stage, the aforementioned core deviation measurement operation is performed. By calculating the deviation between the task-generated interval and the baseline logistics cycle time (e.g., calculating the ratio), the reasonableness of the time consumption for each logistics task is quantitatively assessed. Subsequently, based on this deviation result, each data point can be tagged with data availability, data contaminated by logistics anomalies can be removed, thereby filtering and allowing truly valid records to be released, constructing a valid task dataset.

[0170] Subsequently, the data enters the hierarchical statistics and deviation identification module. To avoid blind intervention, this embodiment slices the effective task dataset according to time periods (such as morning shift, night shift, etc.), calculates the P25 and P50 quantiles of the logistics cycle time, and assesses the degree of drift by calculating the percentage deviation between the actual cycle time and the benchmark logistics cycle time. When the percentage deviation exceeds a preset threshold, the statistics and deviation identification module outputs a list of scenarios that need correction, accurately triggering the subsequent update mechanism.

[0171] For the scenarios to be corrected in the correction scenario list, the process will enter the core logistics cycle time adaptive correction module, where a two-branch adaptive strategy will be executed: First, the effective sample size is checked. In specific scenarios with sparse data (effective sample size less than the preset sample size threshold), a safety mechanism is activated, i.e., when the sample is insufficient, the P50 value is rolled back, and the statistical quantile is directly determined as the corrected logistics cycle time to avoid the risk of overfitting. When the sample is sufficient, the target correction model is activated, which uses time period, container type, production line load rate and other features to adaptively correct the logistics cycle time, and outputs accurate prediction values ​​by capturing the nonlinear mapping relationship of multidimensional production factor features.

[0172] Finally, the corrected results flow to the output and application module, where the final update operation is performed, updating the corrected values ​​back to the production line business logic configuration table (which can be automatically or manually updated after manual confirmation), replacing the original baseline logistics cycle time. Simultaneously, this output and application module continues to be responsible for the closed-loop operation of the entire system, monitoring performance, providing data feedback, and re-feeding the latest operating status back to the system input.

[0173] The production line logistics cycle correction device provided in this embodiment of the invention perfectly integrates data cleaning, deviation interception, adaptive prediction and business closed loop, effectively ensuring high cohesion and low coupling in actual industrial software systems, and ultimately achieving high-precision and high-robust dynamic correction of production line material consumption cycle.

[0174] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 10 As shown, the electronic device may include: a processor 110, a communications interface 120, a memory 130, and a communications bus 140, wherein the processor 110, the communications interface 120, and the memory 130 communicate with each other through the communications bus 140. The processor 110 can call logic instructions in the memory 130 to execute a production line logistics cycle time correction method. This method includes: acquiring a historical task dataset for any target production node in the target production line, the historical task dataset containing multiple historical task data points, and determining the task generation interval corresponding to each historical task data point; measuring the deviation of the task generation interval corresponding to each historical task data point based on the baseline logistics cycle time of each historical task data point, and filtering out an effective task dataset from the historical task dataset based on the measurement result; if the effective sample size in the effective task dataset is less than a preset sample size threshold, determining the corrected logistics cycle time of the target production node based on the statistical quantiles of the effective task dataset; otherwise, inputting the effective task dataset into a target correction model for prediction to obtain the corrected logistics cycle time of the target production node, the target correction model being obtained after training based on production factor samples and used to characterize the nonlinear mapping relationship between multidimensional production factor characteristics and logistics cycle time; and updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time.

[0175] Furthermore, the logical instructions in the aforementioned memory 130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0176] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer is able to execute the production line logistics cycle time correction method provided in the above embodiments. The method includes: acquiring a historical task dataset of any target production node in the target production line, the historical task dataset containing multiple historical task data, determining the task generation interval corresponding to each historical task data; and measuring the deviation of the task generation interval corresponding to each historical task data according to the baseline logistics cycle time of each historical task data. Valid task datasets are obtained by filtering from the historical task dataset based on the measurement results. If the number of valid samples in the valid task dataset is less than a preset sample size threshold, the corrected logistics takt time of the target production node is determined based on the statistical quantiles of the valid task dataset. Otherwise, the valid task dataset is input into the target correction model for prediction to obtain the corrected logistics takt time of the target production node. The target correction model is obtained by training based on production factor samples and is used to characterize the nonlinear mapping relationship between multidimensional production factor characteristics and logistics takt time. Based on the corrected logistics takt time, the baseline logistics takt time of the target production node is updated.

[0177] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the production line logistics cycle time correction method provided in the above embodiments. The method includes: acquiring a historical task dataset of any target production node in a target production line, the historical task dataset containing multiple historical task data, and determining the task generation interval corresponding to each historical task data; measuring the deviation of the task generation interval corresponding to each historical task data based on the baseline logistics cycle time of each historical task data, and filtering out an effective task dataset from the historical task dataset based on the measurement result; if the effective sample size in the effective task dataset is less than a preset sample size threshold, determining the corrected logistics cycle time of the target production node based on the statistical quantile of the effective task dataset; otherwise, inputting the effective task dataset into a target correction model for prediction to obtain the corrected logistics cycle time of the target production node, the target correction model being obtained after training based on production factor samples and used to characterize the nonlinear mapping relationship between multidimensional production factor characteristics and logistics cycle time; and updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time.

[0178] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0179] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for correcting production line logistics cycle time, characterized in that, include: Obtain the historical task dataset of any target production node in the target production line. The historical task dataset contains multiple historical task data. Determine the task generation interval corresponding to each historical task data. Based on the baseline logistics cycle time of each historical task data, the deviation measurement is performed on the task generation interval corresponding to each historical task data, so as to filter out the effective task dataset from the historical task dataset based on the measurement results. If the effective sample size in the effective task dataset is less than a preset sample size threshold, the corrected logistics cycle time of the target production node is determined based on the statistical quantiles of the effective task dataset. Otherwise, the effective task dataset is input into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node. The target correction model is obtained after training based on production factor samples and is used to characterize the nonlinear mapping relationship between multidimensional production factor features and logistics cycle time. Based on the corrected logistics cycle time, the baseline logistics cycle time of the target production node is updated.

2. The production line logistics cycle time correction method according to claim 1, characterized in that, The step of measuring the deviation of task generation intervals in each historical task data based on the baseline logistics cycle time of each historical task data, and filtering out valid task datasets from the historical task dataset based on the measurement results, includes: The ratio of the task generation interval to the baseline logistics cycle time in each historical task data entry is calculated as the measurement result. If the measurement result is within a predetermined reasonable range, then the corresponding historical task data is determined to be valid task data. Construct the effective task dataset using all the aforementioned effective task data.

3. The production line logistics cycle time correction method according to claim 2, characterized in that, The reasonable range is dynamically generated based on the following method: Arrange the multiple task generation intervals obtained during normal historical operation in ascending order of their numerical values; Obtain the first and third quartiles from the permutation results; Calculate the interquartile range between the first quartile and the third quartile; Based on the first quartile, the third quartile, the interquartile range, and the preset product coefficient, the lower and upper limits of the reasonable interval are calculated.

4. The production line logistics cycle time correction method according to claim 2, characterized in that, After updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time, the method further includes: Obtain an online task data sequence containing multiple online task data, and perform time window sliding processing on the online task data sequence using a preset step size to obtain multiple sliding time windows; Within each sliding time window, the number of online task data whose measurement results are not within the reasonable range is counted to obtain the proportion of abnormal data; if the proportion of abnormal data within a consecutive preset number of sliding time windows is greater than a preset proportion threshold, a distribution offset signal is triggered. The target correction model is used to predict the beat of each online task data to obtain the prediction error corresponding to each online task data; the mean of each prediction error within the same sliding time window is calculated to obtain the mean prediction error; if the mean prediction error within a consecutive preset number of sliding time windows is greater than a preset error threshold, a model error signal is triggered. If the distribution offset signal and the model error signal are triggered simultaneously, the online task data is reacquired, and the effective task dataset is filtered and the model is reconstructed based on the reacquired online task data to obtain the corrected logistics cycle time for the next production period; if only the model error signal is triggered and the distribution offset signal is not triggered, the deviation measurement and validity determination of the online task data are re-performed to obtain the target task dataset.

5. The production line logistics cycle time correction method according to claim 2, characterized in that, After obtaining the valid task dataset from the historical task dataset based on the metric results, the process further includes: Obtain a historical task data sequence containing multiple historical task data, and perform time window sliding processing on the historical task data sequence using a preset step size to obtain multiple historical sliding time windows; Within each historical sliding time window, the number of historical task data whose measurement results are not within the reasonable range is statistically analyzed to obtain the percentage of data records. If the proportion of the data records within a consecutive first preset number of historical sliding time windows exceeds a preset proportion threshold, then the current sliding position is determined as a candidate point for production state transition. Based on the candidate points of production state transition, the time window sliding process is performed. If the proportion of the data records in the second consecutive preset number of historical sliding time windows is not greater than the preset proportion threshold, the starting historical task data in the first historical sliding time window in the second consecutive preset number of historical sliding time windows is extracted to obtain the modeling starting point. Based on the modeling starting point, the effective task dataset is divided into time periods to obtain multiple independent initial adaptive production time periods; the effective task data in each initial adaptive production time period is subjected to cycle time correction processing to obtain the corrected logistics cycle time corresponding to each initial adaptive production time period.

6. The production line logistics cycle time correction method according to claim 5, characterized in that, The step of performing cycle time correction processing on the effective task data within each initial adaptive production period to obtain the corrected logistics cycle time corresponding to each initial adaptive production period includes: For any of the initial adaptive production periods, the task generation intervals in the effective task data are sorted in ascending order and quantiles are calculated to obtain a reference logistics cycle time. Based on the reference logistics cycle time and the current benchmark logistics cycle time of the target production node, the time period deviation for the current period is determined; If the time period deviation of the current time period is greater than a preset deviation threshold, then the valid task data of the current time period is subjected to cycle time correction processing to obtain the corrected logistics cycle time of the current time period; if the time period deviation of the current time period is less than or equal to the preset deviation threshold, then the current baseline logistics cycle time is determined as the corrected logistics cycle time of the current time period.

7. The production line logistics cycle time correction method according to claim 1, characterized in that, When the effective sample size in the effective task dataset is less than a preset sample size threshold, the corrected logistics cycle time of the target production node is determined based on the statistical quantiles of the effective task dataset, including: If the effective sample size is less than the preset sample size threshold, the generation intervals of all tasks in the effective task dataset are sorted in ascending order according to their numerical values ​​to obtain the duration sequence. The target task generation interval located in the middle of the duration sequence is extracted and used as the statistical quantile. The statistical quantile is determined as the corrected logistics cycle time for the target production node.

8. The production line logistics cycle time correction method according to claim 1, characterized in that, The step of updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time includes: According to the preset time period division rules, obtain the actual logistics rhythm of the target production node in the current target time period; Calculate the absolute value of the deviation between the actual logistics cycle time and the current benchmark logistics cycle time, and calculate the percentage deviation between the absolute value of the deviation and the benchmark logistics cycle time; If the deviation percentage is greater than a preset deviation threshold, an operation is triggered to update the baseline logistics cycle time of the target production node within the target time period based on the corrected logistics cycle time.

9. The production line logistics cycle time correction method according to claim 1, characterized in that, Before inputting the effective task dataset into the target correction model for prediction, the effective task dataset is vectorized, specifically including: Extract multidimensional production factor features from the effective task dataset, the multidimensional production factor features including discrete features and continuous features; Convert the discrete features into binary sparse vectors; Map the continuous features to a preset numerical range; The input feature vector is generated by fusing the binary sparse vector and the mapped continuous features.

10. The production line logistics cycle time correction method according to claim 9, characterized in that, The target correction model includes multiple independent regression sub-models, each containing multiple decision trees; the step of inputting the effective task dataset into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node includes: The input feature vectors are respectively input into the multiple independent regression sub-models; Within each regression sub-model, the basic predicted values ​​output by the multiple decision trees are obtained, and the internal average of all the basic predicted values ​​is calculated to obtain multiple independent predicted values ​​corresponding to each regression sub-model. Calculate the arithmetic mean of the multiple independent predicted values, and use the arithmetic mean as the corrected logistics cycle time of the target production node.

11. The production line logistics cycle time correction method according to any one of claims 9-10, characterized in that, After generating the input feature vector, the process further includes: Calculate the mutual information entropy between each feature column in the input feature vector and the target predicted label; The feature columns are arranged in descending order according to the magnitude of the mutual information entropy. Redundant feature columns with mutual information entropy lower than a preset information threshold are removed to obtain the preferred feature matrix after dimensionality reduction; The preferred feature matrix is ​​used as the final input to the target correction model.

12. The production line logistics cycle time correction method according to claim 1, characterized in that, The step of updating the baseline logistics cycle time of the target production node based on the corrected logistics cycle time includes: Obtain the preset smoothing weight coefficients; The first component is determined based on the modified logistics cycle time and the smoothing weight coefficient; The second component is determined based on the current benchmark logistics cycle time and the remaining weight coefficient, wherein the remaining weight coefficient is the difference between 1 and the smoothing weight coefficient; Based on the first component and the second component, the updated baseline logistics cycle time is obtained.

13. The production line logistics cycle time correction method according to claim 12, characterized in that, Before obtaining the preset smoothing weight coefficients, the method further includes: Obtain the percentage deviation between the corrected logistics cycle time and the current baseline logistics cycle time; The smoothing weight coefficient is dynamically adjusted based on the magnitude of the deviation percentage. The larger the percentage of deviation, the larger the value of the determined smoothing weight coefficient.

14. The production line logistics cycle time correction method according to claim 1, characterized in that, Before obtaining the historical task dataset of any target production node in the target production line, the following steps are also included: Obtain the raw task data stream; Remove empty task data, loop task data, and discontinuous task data with discontinuous task serial numbers from the original task data stream; The original task data stream after the removal process is normalized to a timestamp format to obtain the historical task dataset.

15. A production line logistics cycle time correction device, characterized in that, include: The data acquisition unit is used to acquire the historical task dataset of any target production node in the target production line. The historical task dataset contains multiple historical task data and determines the task generation interval corresponding to each historical task data. The data filtering unit is used to measure the deviation of the task generation interval corresponding to each historical task data according to the baseline logistics cycle time of each historical task data, so as to filter the effective task dataset from the historical task dataset based on the measurement result. The cycle time determination unit is used to determine the corrected logistics cycle time of the target production node based on the statistical quantiles of the effective task dataset when the effective sample size in the effective task dataset is less than a preset sample size threshold. Otherwise, the effective task dataset is input into the target correction model for prediction to obtain the corrected logistics cycle time of the target production node. The target correction model is obtained after training based on production factor samples and is used to characterize the nonlinear mapping relationship between multidimensional production factor features and logistics cycle time. The baseline update unit is used to update the baseline logistics cycle time of the target production node based on the corrected logistics cycle time.

16. 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 computer program, it implements the production line logistics cycle time correction method as described in any one of claims 1 to 14.

17. A non-transitory 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 production line logistics cycle time correction method as described in any one of claims 1 to 14.

Images