Method for real-time evaluation of order fulfillment capacity of tea leaf packaging box production factory

Abstract

The application discloses a tea leaf packaging box production factory order fulfillment capacity real-time evaluation method, relates to the production order fulfillment management technical field, and comprises the following steps: step one, constructing an order fulfillment time sequence data set; step two, obtaining an order fulfillment beat variation sequence; step three, generating an order fulfillment beat state sequence; step four, generating a fulfillment time sequence and a fulfillment state Token sequence; step five, executing a box order beat delivery mechanism to obtain a box order beat Token sequence; step six, a Chronos coding prediction unit generates future box order beat Tokens, and a fulfillment capacity inverse quantization output unit generates a future fulfillment beat prediction sequence; step seven, generating an order fulfillment capacity state; and step eight, outputting an order fulfillment capacity real-time evaluation result. The improved Chronos model and the discrete event fulfillment deducer are utilized to realize continuous and accurate evaluation of the order fulfillment capacity.

Description

Technical Field

[0001] This invention relates to the field of production order fulfillment management technology, and in particular to a method for real-time assessment of the order fulfillment capability of tea packaging box manufacturing plants. Background Technology

[0002] As tea packaging box production orders trend towards smaller batches, more diverse specifications, shorter lead times, and more frequent deliveries, real-time assessment technology for order fulfillment capabilities of tea packaging box manufacturing plants has garnered significant attention. Existing production management systems primarily rely on order delivery dates, scheduling, manual work reports, or static capacity tables for fulfillment assessment. However, these systems commonly suffer from the following problems in practical applications:

[0003] The lack of effective temporal organization in order fulfillment status during continuous production means that order production status, process waiting status, resource release status, and remaining delivery status are often recorded in a scattered manner. This leads the system to judge whether an order is delayed based on a single production point, making it difficult to reflect changes in the fulfillment takt time between adjacent production points. In the production process of tea packaging boxes, waiting, recovery, and blockage occur alternately in stages such as material cutting, printing, surface treatment, die cutting, box gluing, and warehousing. Existing scheduling methods usually only count the completed quantity or equipment occupancy, making it difficult to distinguish between normal fulfillment takt time, unrecoverable fulfillment takt time, and blocked fulfillment takt time. This can easily lead to misjudging short-term resource continuity as fulfillment risk or continuous blockage as recoverable waiting. Regarding the prediction of future fulfillment capacity, traditional order evaluation methods lack continuous modeling of takt breakpoints, takt reconnection points, and takt lockpoints. It is difficult to convert historical fulfillment status into a predictable box order takt time token sequence, and it is also impossible to form a stable judgment of order fulfillment capacity status through future fulfillment time prediction sequences and discrete event fulfillment inferencers, affecting the accuracy of order delivery confirmation and production scheduling adjustments.

[0004] Therefore, how to provide a real-time assessment method for the order fulfillment capability of tea packaging box manufacturing plants is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a real-time evaluation method for the order fulfillment capability of tea packaging box manufacturing plants. This invention utilizes an improved Chronos model and a discrete event fulfillment inference tool to perform time-series processing on the order fulfillment status during continuous production at the tea packaging box manufacturing plant. It extracts changes in order fulfillment rhythm and generates an order fulfillment rhythm status sequence. Through a box order rhythm delivery mechanism, it identifies rhythm breakpoints, rhythm reconnection points, and rhythm lockpoints, constructing a box order rhythm token sequence. Furthermore, it predicts future changes in fulfillment rhythm and generates the order fulfillment capability status, thereby achieving a continuous, accurate, and interpretable evaluation of order fulfillment capability, providing a reliable basis for order delivery confirmation and production scheduling adjustments.

[0006] The real-time evaluation method for order fulfillment capability of a tea packaging box manufacturing plant according to an embodiment of the present invention includes the following steps:

[0007] Step 1: Collect order fulfillment status at continuous production points and construct an order fulfillment time series dataset;

[0008] Step 2: Extract the order fulfillment cycle time variation between adjacent production time points based on the order fulfillment time series dataset to obtain the order fulfillment cycle time variation sequence;

[0009] Step 3: Mark the cycle time status according to the order fulfillment cycle time variation sequence to generate the order fulfillment cycle time status sequence;

[0010] Step 4: Input the order fulfillment cycle time status sequence into the improved Chronos model. The fulfillment time series processing unit generates a fulfillment time series based on the order fulfillment cycle time sequence, and the fulfillment status token construction unit generates a fulfillment status token sequence based on the fulfillment time series.

[0011] Step 5: Execute the box list beat delivery mechanism in the fulfillment status Token construction unit, identify the beat breakpoint, beat reconnection point and beat locking point, and construct the reconnection delivery Token and the locking delivery Token with the beat breakpoint as the delivery starting point. Embed the reconnection delivery Token and the locking delivery Token into the fulfillment status Token sequence to obtain the box list beat Token sequence.

[0012] Step 6: Input the box order beat token sequence into the Chronos encoding prediction unit to generate the future box order beat token, and then use the fulfillment capability inverse quantization output unit to restore the future box order beat token into the future fulfillment beat prediction sequence.

[0013] Step 7: Input the future fulfillment cycle prediction sequence into the discrete event fulfillment inference engine to generate the order fulfillment capability status;

[0014] Step 8: Output the real-time evaluation results of order fulfillment capability based on the order fulfillment capability status.

[0015] Optionally, step one specifically includes:

[0016] The order fulfillment status is collected at continuous production points in the tea packaging box production line. The order fulfillment status includes order production status, process waiting status, resource release status, and delivery remaining status.

[0017] Arrange the order fulfillment status of the same order according to the production time point to obtain the single order fulfillment sequence;

[0018] The order fulfillment time sequence corresponding to each order is summarized according to the order number to obtain the order fulfillment time sequence dataset.

[0019] Optionally, step two specifically includes:

[0020] Read the order fulfillment time sequence one by one from the order fulfillment time sequence dataset, and read the order fulfillment status of the same order at the previous production time point and the next production time point;

[0021] Extract the cumulative completed quantity from the order production status, and subtract the cumulative completed quantity of the previous production time from the cumulative completed quantity of the next production time to obtain the production progress quantity;

[0022] Extract the waiting process marker and waiting duration from the process waiting status. When the waiting process marker at the next production time point is the same as the waiting process marker at the previous production time point, use the waiting duration at the next production time point as the waiting dwell time. When the waiting process marker at the next production time point is different from the waiting process marker at the previous production time point, reset the waiting dwell time to the set base value.

[0023] Extract the resource availability marker from the resource release status. When the resource availability marker of the previous production time point is not released and the resource availability marker of the next production time point is released, mark the corresponding adjacent production time point as the resource continuation point.

[0024] Extract the remaining delivery time from the remaining delivery status, and subtract the remaining delivery time of the next production time from the remaining delivery time of the previous production time to obtain the delivery compression amount.

[0025] By binding production progress, waiting time, resource connection points, and delivery compression according to adjacent production time points, the order fulfillment rhythm changes can be obtained.

[0026] Arrange the changes in order fulfillment takt time of the target order at consecutive adjacent production time points in the order of production time points to obtain the order fulfillment takt time change sequence.

[0027] Optionally, step three specifically includes:

[0028] Read the order fulfillment cycle change sequence in order of production time, extract the production progress and resource continuity points in adjacent production time intervals, and mark the cycle status. The cycle status includes normal fulfillment cycle, fulfillment cycle to be restored, and fulfillment cycle that is blocked.

[0029] When the production progress is greater than zero, the interval between adjacent production time points is marked as the normal fulfillment cycle time.

[0030] When the production advance is zero and the resource succession point has been formed, the adjacent production time interval is marked as the fulfillment cycle to be restored.

[0031] When the production progress is zero and the resource succession point has not been formed, the interval between adjacent production time points is marked as a stalled performance cycle.

[0032] The normal fulfillment cycle time, the fulfillment cycle time to be restored, and the stalled fulfillment cycle time are arranged in the order of production time to generate an order fulfillment cycle time status sequence.

[0033] Optionally, step four specifically involves:

[0034] The order fulfillment cycle state sequence is input into the improved Chronos model, which includes a fulfillment time series processing unit, a fulfillment status token construction unit, a Chronos encoding prediction unit, and a fulfillment capability inverse quantization output unit.

[0035] The order fulfillment time sequence processing unit reads the normal fulfillment cycle, the fulfillment cycle to be recovered, and the stalled fulfillment cycle from the order fulfillment cycle status sequence, and reads the adjacent production time intervals corresponding to each cycle status;

[0036] According to the order of adjacent production time intervals in the order fulfillment cycle state sequence, write the initial interval position number for the first adjacent production time interval, and increment and update the interval position number of the subsequent adjacent production time intervals along the production time sequence to obtain the corresponding interval position number.

[0037] Normal performance timeframe, performance timeframe to be restored, and performance timeframe that is stalled are bound to the corresponding interval location number to generate a performance time series;

[0038] The performance time series is input into the performance status token construction unit. The performance status token construction unit encodes the normal performance beat, the performance beat to be recovered, and the performance beat that is blocked in the performance time series, and encodes the interval position number to obtain the beat status code and the interval position code respectively.

[0039] By concatenating the cycle state code and interval position code corresponding to the same adjacent production time interval, a single interval performance status Token is obtained.

[0040] Arrange the performance status tokens of each single interval according to the interval position number order to generate a performance status token sequence.

[0041] Optionally, step five specifically includes:

[0042] The box delivery mechanism is executed, the fulfillment status token sequence is read, and the fulfillment status tokens of adjacent single intervals are read in order of interval position number.

[0043] Subtract the previous order interval fulfillment status Token from the next order interval fulfillment status Token to obtain the beat difference Token. Then, concatenate the previous order interval fulfillment status Token, the next order interval fulfillment status Token, and the beat difference Token according to the channel dimension to obtain the beat migration Token.

[0044] The beat transition token is input into the Sigmoid activation function to obtain the beat transition gating value, and the beat transition token is then processed by GELU activation and LayerNorm normalization to obtain the beat transition response token.

[0045] Multiply the beat transition gating value element by element with the beat transition response Token to obtain the beat delivery discrimination Token;

[0046] When the cycle status code changes from the code corresponding to the normal performance cycle to the code corresponding to the performance cycle to be restored, the interval position number corresponding to the next single interval performance status Token is marked as the cycle breakpoint.

[0047] When the cycle status code changes from the code corresponding to the cycle to be restored to the code corresponding to the normal cycle, the interval position number corresponding to the next single interval cycle status Token is marked as the cycle reconnection point.

[0048] When the beat status code changes from the code corresponding to the beat to be restored to the code corresponding to the beat that is blocked, the interval position number corresponding to the next single interval performance status Token is marked as the beat lock point.

[0049] Starting from the beat breakpoint, the beat delivery discrimination tokens, including the beat breakpoint and the beat return point, are connected in order of interval position number to obtain the return delivery token.

[0050] Starting from the beat breakpoint, the beat delivery discrimination tokens, including the beat breakpoint and the beat lockpoint, are connected in order of interval position number to obtain the lock delivery token.

[0051] Write the return delivery token into the interval position number of the corresponding beat break point in the performance status token sequence, and mark the single interval performance status token, including the beat break point and the beat return point, as the returned delivery token.

[0052] Write the locked delivery token into the interval position number of the corresponding beat breakpoint in the performance status token sequence, and mark the single interval performance status token, including the beat breakpoint and the beat lockpoint, as a locked delivery token.

[0053] Retain the single-segment fulfillment status tokens that are not marked as received delivery tokens and not marked as locked delivery tokens, and arrange the retained single-segment fulfillment status tokens, received delivery tokens, and locked delivery tokens in order of segment position number to obtain the box order beat token sequence.

[0054] Optionally, the step of inputting the box order beat token sequence into the Chronos encoding prediction unit to generate future box order beat tokens specifically involves:

[0055] The Chronos coding prediction unit reads the box-single beat token sequence in the order of interval position numbering, and reads the interval position code corresponding to each box-single beat token in the box-single beat token sequence;

[0056] Each box's single beat token and its corresponding interval position code are concatenated according to the channel dimension to obtain the beat position token sequence;

[0057] Input the beat position token sequence into the Transformer encoding block, calculate the attention response between each beat position token through self-attention, and update the beat position token sequence with weights based on the attention response to obtain the beat context token sequence.

[0058] The beat context token sequence is input into the linear mapping layer, and then subjected to GELU activation function and LayerNorm normalization in sequence to obtain the beat prediction context token;

[0059] Input the beat prediction context token into the Softmax output layer to obtain the probability of the future beat state corresponding to the future production time point;

[0060] Select the beat state with the highest probability from the probabilities of future beat states as the future beat state, and read the interval position code corresponding to the future production time point;

[0061] The future beat state code is concatenated with the interval position code corresponding to the future production time point to generate the future box order beat token.

[0062] Optionally, the step of restoring the future order tachometer token to a future fulfillment tachometer prediction sequence through the fulfillment capability dequantization output unit specifically involves:

[0063] The fulfillment capability inverse quantification output unit reads the future box order beat token and separates the beat state code corresponding to the future beat state and the interval position code corresponding to the future production time point from the future box order beat token;

[0064] The corresponding normal performance cycle, performance cycle to be resumed, or performance cycle that is stalled is determined based on the cycle state code, and the corresponding future production time sequence is determined based on the interval position code.

[0065] Arrange the normal performance rhythm, the performance rhythm to be restored, and the performance rhythm that is hindered at each future production time point in the order of the future production time points to obtain the future performance rhythm prediction sequence.

[0066] Optionally, step seven specifically includes:

[0067] Input the future performance cycle prediction sequence into the discrete event performance inference engine, read the future performance cycle prediction sequence according to the future production time point sequence, convert the normal performance cycle into a production advancement event, convert the performance cycle to be restored into a resource succession event, and convert the stalled performance cycle into a performance stall event.

[0068] By connecting production advancement events, resource succession events, and fulfillment delay events in the order of future production time points, an order fulfillment event chain is obtained.

[0069] Count the number of production progress events, resource continuation events, and fulfillment delay events in the order fulfillment event chain, and read the last event in the order fulfillment event chain;

[0070] When the number of production advancement events exceeds the sum of the number of resource succession events and the number of performance delay events, and the last event is a production advancement event, a strong performance status is generated.

[0071] When the number of performance delay events exceeds the number of production progress events, or when the last event is a performance delay event, a weak performance status is generated.

[0072] When the conditions for generating a strong performance state and the conditions for generating a weak performance state are not met, a performance state to be restored is generated.

[0073] Strong performance status, pending recovery performance status, and weak performance status are used as the order fulfillment capability status.

[0074] Optionally, step eight specifically includes:

[0075] Read the order fulfillment capability status and the corresponding order number, and generate a fulfillment order tag based on the order fulfillment capability status;

[0076] When an order's fulfillment capability status is "strong fulfillment," the fulfillment order is marked as a "committable fulfillment order." When an order's fulfillment capability status is "pending restoration of fulfillment," the fulfillment order is marked as a "pending tracking fulfillment order." When an order's fulfillment capability status is "weak fulfillment," the fulfillment order is marked as a "risky fulfillment order."

[0077] The order number, current production time, order fulfillment capability status, and fulfillment order tag are bound together to generate a real-time evaluation result of order fulfillment capability;

[0078] The real-time assessment results of order fulfillment capability are output to the order management and production planning ends for order delivery confirmation and production scheduling adjustment.

[0079] The beneficial effects of this invention are:

[0080] This invention addresses the problems in order fulfillment assessment faced by tea packaging box manufacturing plants. It addresses issues such as single-point-of-time judgment lag, difficulty in distinguishing between process waiting and resource release, and lack of continuous prediction of future delivery capacity by integrating order fulfillment cycle time variation extraction, improved Chronos model prediction, and discrete event fulfillment inference. It collects order fulfillment status data at continuous production points, constructs an order fulfillment time-series dataset, and extracts production progress, waiting time, resource connection points, and delivery compression from adjacent production points to form an order fulfillment cycle time variation sequence. This transforms the static record of order production status into a calculable time-series cycle time variation. During the cycle time status marking stage, normal fulfillment cycle time, unrecoverable fulfillment cycle time, and stalled fulfillment cycle time are generated based on production progress and resource connection points, thus distinguishing between continuous order progress, resources that have been connected but not yet resumed production, and fulfillment stalls caused by resource disconnection. The improved Chronos model... In the process, the order fulfillment time sequence is transformed into a fulfillment status token sequence through the fulfillment time sequence processing unit and the fulfillment status token construction unit. In the box order delivery mechanism, the breakpoint, the return point, and the lock point are identified, and the return delivery token and the lock delivery token are constructed. This enables the model to express the evolution process of the order from normal progress to waiting to be restored, re-returned, or locked. In the prediction output stage, the Chronos encoding prediction unit generates future box order tokens and restores them to the future fulfillment time sequence prediction sequence through the fulfillment capability dequantization output unit. Finally, the discrete event fulfillment inference unit converts the future fulfillment time sequence prediction sequence into an order fulfillment event chain, generating strong fulfillment status, waiting to be restored fulfillment status, and weak fulfillment status. This enables the output of real-time evaluation results of order fulfillment capability, providing a continuous and interpretable basis for fulfillment judgment for order delivery confirmation and production scheduling adjustment. Attached Figure Description

[0081] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0082] Figure 1 This is an overall flowchart of the real-time evaluation method for the order fulfillment capability of a tea packaging box manufacturing plant proposed in this invention;

[0083] Figure 2 This is a schematic diagram of the improved Chronos model for the real-time evaluation method of order fulfillment capability of tea packaging box manufacturing plants proposed in this invention.

[0084] Figure 3 This is a schematic diagram illustrating the process by which the discrete event fulfillment inferencer generates the order fulfillment capability status of a tea packaging box manufacturing plant, as proposed in this invention, in the real-time evaluation method for order fulfillment capability of the plant. Detailed Implementation

[0085] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0086] refer to Figures 1-3 The real-time assessment method for the order fulfillment capability of tea packaging box manufacturing plants includes the following steps:

[0087] Step 1: Collect order fulfillment status at continuous production points and construct an order fulfillment time series dataset;

[0088] Step 2: Extract the order fulfillment cycle time variation between adjacent production time points based on the order fulfillment time series dataset to obtain the order fulfillment cycle time variation sequence;

[0089] Step 3: Mark the cycle time status according to the order fulfillment cycle time variation sequence to generate the order fulfillment cycle time status sequence;

[0090] Step 4: Input the order fulfillment cycle time status sequence into the improved Chronos model. The fulfillment time series processing unit generates a fulfillment time series based on the order fulfillment cycle time sequence, and the fulfillment status token construction unit generates a fulfillment status token sequence based on the fulfillment time series.

[0091] Step 5: Execute the box list beat delivery mechanism in the fulfillment status Token construction unit, identify the beat breakpoint, beat reconnection point and beat locking point, and construct the reconnection delivery Token and the locking delivery Token with the beat breakpoint as the delivery starting point. Embed the reconnection delivery Token and the locking delivery Token into the fulfillment status Token sequence to obtain the box list beat Token sequence.

[0092] Step 6: Input the box order beat token sequence into the Chronos encoding prediction unit to generate the future box order beat token, and then use the fulfillment capability inverse quantization output unit to restore the future box order beat token into the future fulfillment beat prediction sequence.

[0093] Step 7: Input the future fulfillment cycle prediction sequence into the discrete event fulfillment inference engine to generate the order fulfillment capability status;

[0094] Step 8: Output the real-time evaluation results of order fulfillment capability based on the order fulfillment capability status.

[0095] In this embodiment, step one specifically includes:

[0096] The order fulfillment status is collected at continuous production points in the tea packaging box production line. The order fulfillment status includes order production status, process waiting status, resource release status, and delivery remaining status.

[0097] Arrange the order fulfillment status of the same order according to the production time point to obtain the single order fulfillment sequence;

[0098] The order fulfillment time sequence corresponding to each order is summarized according to the order number to obtain the order fulfillment time sequence dataset;

[0099] In practical implementation, continuous production time points correspond to the periodic records formed by the tea packaging box production factory within each production shift. Order fulfillment status is collected at each production time point, ensuring that the fulfillment changes of the same order at different production time points are continuously preserved. Order production status represents the production progress of the order at the current production time point; process waiting status represents the waiting status of the order in the stages of material cutting, printing, surface treatment, die-cutting, box gluing, or warehousing; resource release status represents whether production resources related to the continued fulfillment of the order have been released; and delivery remainder status represents the remaining fulfillment space of the order before it meets delivery requirements. Arranging the order fulfillment status of the same order according to production time points forms a single order fulfillment time sequence reflecting the order fulfillment process. Then, summarizing the single order fulfillment time sequences corresponding to each order according to order number yields an order fulfillment time sequence dataset. This processing method avoids the lag and bias caused by judging order fulfillment capability based solely on a single time point, ensuring that the extraction of order fulfillment rhythm changes has a continuous temporal basis.

[0100] In this embodiment, step two specifically includes:

[0101] Read the order fulfillment time sequence one by one from the order fulfillment time sequence dataset, and read the order fulfillment status of the same order at the previous production time point and the next production time point;

[0102] Extract the cumulative completed quantity from the order production status, and subtract the cumulative completed quantity of the previous production time from the cumulative completed quantity of the next production time to obtain the production progress quantity;

[0103] Extract the waiting process marker and waiting duration from the process waiting status. When the waiting process marker at the next production time point is the same as the waiting process marker at the previous production time point, use the waiting duration at the next production time point as the waiting dwell time. When the waiting process marker at the next production time point is different from the waiting process marker at the previous production time point, reset the waiting dwell time to the set base value.

[0104] Extract the resource availability marker from the resource release status. When the resource availability marker of the previous production time point is not released and the resource availability marker of the next production time point is released, mark the corresponding adjacent production time point as the resource continuation point.

[0105] Extract the remaining delivery time from the remaining delivery status, and subtract the remaining delivery time of the next production time from the remaining delivery time of the previous production time to obtain the delivery compression amount.

[0106] By binding production progress, waiting time, resource connection points, and delivery reduction according to adjacent production time points, the change in order fulfillment rhythm can be obtained;

[0107] Arrange the changes in order fulfillment takt time of the target order under consecutive adjacent production time points in the order of production time points to obtain the order fulfillment takt time change sequence;

[0108] In the specific implementation process, the production time point is set to be generated every 30 minutes. Each record in the single order fulfillment sequence is stored using a field structure of "order number, production time point, cumulative completed quantity, waiting process mark, waiting duration, resource availability mark, and remaining delivery time". The cumulative completed quantity is counted in units of the number of tea packaging boxes that have been completed. The waiting process mark is recorded using a fixed code, with cutting recorded as A, printing as B, surface treatment as C, die cutting as D, box gluing as E, and warehousing as F. The waiting duration is recorded in minutes. When the waiting process mark changes, it means that the order has left the original waiting process, and the waiting duration is reset to the set base value of 0 minutes, thereby preventing the waiting time of the previous process from being carried over to the next process.

[0109] Resource availability is marked with a binary value: 0 for unreleased and 1 for released. When the current production time point is 0 and the next production time point is 1, the value of the resource continuation point is recorded as 1, and the values ​​of other adjacent production time points are recorded as 0. This indicates that the order has obtained the resources required for continued fulfillment within the production interval. The remaining delivery time is recorded in hours, and the delivery compression amount is used to characterize the delivery time consumed between adjacent production time points. The change in each order's fulfillment cycle time is stored in the structure of "production time point interval, production progress, waiting time, resource continuation point, and delivery compression amount". This allows the order fulfillment cycle time status sequence to simultaneously identify production progress, waiting accumulation, resource release, and delivery time consumption. This processing method can transform the order fulfillment status from a static record into a calculable change between continuous production time points, reducing misjudgments caused by judging fulfillment capability solely based on the current process or delivery date.

[0110] In this embodiment, step three specifically includes:

[0111] Read the order fulfillment cycle change sequence in order of production time, extract the production progress and resource continuity points in adjacent production time intervals, and mark the cycle status. The cycle status includes normal fulfillment cycle, fulfillment cycle to be restored, and fulfillment cycle that is blocked.

[0112] When the production progress is greater than zero, the interval between adjacent production time points is marked as the normal fulfillment cycle time.

[0113] When the production advance is zero and the resource succession point has been formed, the adjacent production time interval is marked as the fulfillment cycle to be restored.

[0114] When the production progress is zero and the resource succession point has not been formed, the interval between adjacent production time points is marked as a stalled performance cycle.

[0115] Arrange the normal fulfillment cycle time, the fulfillment cycle time to be restored, and the stalled fulfillment cycle time according to the production time sequence to generate an order fulfillment cycle time status sequence;

[0116] In practical implementation, production progress directly reflects whether an order has actually been produced within an adjacent production time interval, while resource continuity points reflect whether the resource conditions for continued production are available when an order has not been produced. Therefore, these two factors can divide the order fulfillment cycle status into three distinct states: "progressed," "pending recovery," and "blocked." Production progress is based on changes in the number of tea packaging boxes completed, while resource continuity points are based on the change from unreleased to released resources such as printing plates, die-cutting molds, outsourced returns, or available resources in the process. The reason for using this marking method is that an order not progressing does not necessarily indicate an order fulfillment anomaly. If a resource continuity point has been formed, it means that the order has the conditions to resume production and should be classified as a pending recovery fulfillment cycle. If an order has not progressed and a resource continuity point has not been formed, it means that the order is still in a waiting or restricted state and should be classified as a blocked fulfillment cycle. Through this processing, we can avoid misjudging short-term waiting as fulfillment blockage and enable the improved Chronos model to obtain a continuous and clear sequence of order fulfillment cycle statuses.

[0117] In this embodiment, step four specifically includes:

[0118] The order fulfillment cycle state sequence is input into the improved Chronos model. The improved Chronos model includes a fulfillment time series processing unit, a fulfillment status token construction unit, a Chronos encoding prediction unit, and a fulfillment capability inverse quantization output unit.

[0119] The order fulfillment time series processing unit reads the normal fulfillment cycle, the fulfillment cycle to be recovered, and the stalled fulfillment cycle from the order fulfillment cycle status sequence, and reads the adjacent production time intervals corresponding to each cycle status;

[0120] According to the order of adjacent production time intervals in the order fulfillment cycle state sequence, write the initial interval position number for the first adjacent production time interval, and increment and update the interval position number of the subsequent adjacent production time intervals along the production time sequence to obtain the corresponding interval position number.

[0121] Normal performance timeframe, performance timeframe to be restored, and performance timeframe that is stalled are bound to the corresponding interval location number to generate a performance time series;

[0122] The performance time series is input into the performance status token construction unit. The performance status token construction unit encodes the normal performance beat, the performance beat to be recovered, and the performance beat that is blocked in the performance time series, and encodes the interval position number to obtain the beat status code and the interval position code respectively.

[0123] By concatenating the cycle state code and interval position code corresponding to the same adjacent production time interval, a single interval performance status Token is obtained.

[0124] Arrange the performance status tokens of each single interval according to the interval position number order to generate a performance status token sequence;

[0125] In the specific implementation process, the initial interval position number is set to 0, and the interval position number of subsequent adjacent production time point intervals is incremented by a step size of 1, so that each beat state has a unique temporal position in the performance time sequence; the beat state encoding adopts a fixed integer encoding method, with normal performance beat encoding as 0, performance beat waiting to be recovered encoding as 1, and performance beat hindered encoding as 2; the interval position encoding adopts sine and cosine position encoding, dividing the interval position number by 1, 10, 100, and 1000 respectively to obtain four position phase values, performing sine and cosine operations on the four position phase values ​​respectively, and arranging the four sine values ​​and four cosine values ​​in sequence to form an eight-dimensional interval position encoding; the single interval performance state token is formed by concatenating the beat state encoding and the interval position encoding, which can simultaneously express the performance category and time sequence of the order within adjacent production time point intervals. This processing improves the Chronos model so that it receives not isolated beat tags, but a sequence of performance status tokens with continuous temporal positions. This helps to identify the temporal correlation between normal progress, recovery progress, and stall continuation, and improves the stability of future performance beat prediction sequences.

[0126] In this embodiment, step five specifically includes:

[0127] The box delivery mechanism is executed, the fulfillment status token sequence is read, and the fulfillment status tokens of adjacent single intervals are read in order of interval position number.

[0128] Subtract the previous order interval fulfillment status Token from the next order interval fulfillment status Token to obtain the beat difference Token. Then, concatenate the previous order interval fulfillment status Token, the next order interval fulfillment status Token, and the beat difference Token according to the channel dimension to obtain the beat migration Token.

[0129] The beat transition token is input into the Sigmoid activation function to obtain the beat transition gating value, and the beat transition token is then processed by GELU activation and LayerNorm normalization to obtain the beat transition response token.

[0130] Multiply the beat transition gating value element by element with the beat transition response Token to obtain the beat delivery discrimination Token;

[0131] When the cycle status code changes from the code corresponding to the normal performance cycle to the code corresponding to the performance cycle to be restored, the interval position number corresponding to the next single interval performance status Token is marked as the cycle breakpoint.

[0132] When the cycle status code changes from the code corresponding to the cycle to be restored to the code corresponding to the normal cycle, the interval position number corresponding to the next single interval cycle status Token is marked as the cycle reconnection point.

[0133] When the beat status code changes from the code corresponding to the beat to be restored to the code corresponding to the beat that is blocked, the interval position number corresponding to the next single interval performance status Token is marked as the beat lock point.

[0134] Starting from the beat breakpoint, the beat delivery discrimination tokens, including the beat breakpoint and the beat return point, are connected in order of interval position number to obtain the return delivery token.

[0135] Starting from the beat breakpoint, the beat delivery discrimination tokens, including the beat breakpoint and the beat lockpoint, are connected in order of interval position number to obtain the lock delivery token.

[0136] Write the return delivery token into the interval position number of the corresponding beat break point in the performance status token sequence, and mark the single interval performance status token, including the beat break point and the beat return point, as the returned delivery token.

[0137] Write the locked delivery token into the interval position number of the corresponding beat breakpoint in the performance status token sequence, and mark the single interval performance status token, including the beat breakpoint and the beat lockpoint, as a locked delivery token.

[0138] Retain the single-segment fulfillment status tokens that are not marked as received delivery tokens and not marked as locked delivery tokens, and arrange the retained single-segment fulfillment status tokens, received delivery tokens, and locked delivery tokens in order of segment position number to obtain the box order beat token sequence;

[0139] In the specific implementation process, the single-interval performance status tokens in the performance status token sequence are stored in ascending order according to the interval position number. Each single-interval performance status token is represented by a fixed-dimensional vector with a vector dimension of 16. When executing the box delivery mechanism, the first single-interval fulfillment state token does not have the previous single-interval fulfillment state token. Therefore, the beat difference token corresponding to the first single-interval fulfillment state token is set to a 16-dimensional zero vector. Starting from the second single-interval fulfillment state token, the operation of subtracting the previous single-interval fulfillment state token from the next single-interval fulfillment state token is performed one by one to obtain the beat difference token that can reflect the direction of the beat state change. The sigmoid activation function is used to compress the beat migration token to between 0 and 1 to form the beat migration gating value, so that weak beat changes are suppressed. GELU activation and LayerNorm normalization are used to retain the nonlinear changes in the beat migration token and stabilize the vector scale. By multiplying the beat migration gating value with the beat migration response token element by element, the beat delivery discrimination token is obtained, so that the beat breakpoint, beat reconnection point and beat lock point are not only determined by the beat state encoding change, but also retain the change intensity between adjacent fulfillment state tokens. After writing the reconnection delivery token or the lock delivery token to the corresponding position of the beat breakpoint, the single-interval fulfillment status tokens covered in the original interval will no longer be repeatedly entered into subsequent predictions. This reduces redundant inputs caused by continuous unrecoverable intervals or continuous blocked intervals, and makes the box order beat token sequence more concentrated in expressing the disconnection, reconnection and lock process of the order fulfillment beat.

[0140] Furthermore, when executing the box delivery mechanism, since the return delivery token and the lock delivery token are formed by connecting multiple time point delivery discrimination tokens within the abnormal window interval according to the interval position number order, their feature channel dimension will dynamically step expand as the span of the abnormal duration interval increases, resulting in inconsistency in channel dimension between them and the 16-dimensional single-interval fulfillment status token that has not been marked as discarded. In order to ensure the matrix homogeneity when the whole sequence is input into the improved Chronos model, in the specific implementation process, before embedding the return delivery token and the lock delivery token into the sequence, the fulfillment status token construction unit first inputs them into an adaptive linear projection layer. The weight matrix parameters of the adaptive linear projection layer are learned synchronously by the improved Chronos model during the training phase. The return delivery token and the lock delivery token are uniformly reduced in dimension and truncated to a fixed 16-dimensional space that is completely consistent with the single-interval fulfillment status token through matrix multiplication. Then, they are concatenated with the corresponding interval position code, thereby completely closing the loop in mathematical arithmetic and ensuring that the Transformer coding block can perform uniform scale self-attention response matrix calculation for event block tokens of heterogeneous lengths.

[0141] In this embodiment, the box order beat token sequence is input into the Chronos encoding prediction unit to generate future box order beat tokens, specifically as follows:

[0142] The Chronos coding prediction unit reads the box-single beat token sequence in the order of interval position numbering, and reads the interval position code corresponding to each box-single beat token in the box-single beat token sequence;

[0143] Each box's single beat token and its corresponding interval position code are concatenated according to the channel dimension to obtain the beat position token sequence;

[0144] Input the beat position token sequence into the Transformer encoding block, calculate the attention response between each beat position token through self-attention, and update the beat position token sequence with weights based on the attention response to obtain the beat context token sequence.

[0145] The beat context token sequence is input into the linear mapping layer, and then subjected to GELU activation function and LayerNorm normalization in sequence to obtain the beat prediction context token;

[0146] Input the beat prediction context token into the Softmax output layer to obtain the probability of the future beat state corresponding to the future production time point;

[0147] Select the beat state with the highest probability from the probabilities of future beat states as the future beat state, and read the interval position code corresponding to the future production time point;

[0148] The future beat state code is concatenated with the interval position code corresponding to the future production time point to generate the future box order beat token.

[0149] In the specific implementation process, each box-by-box beat token includes a single-interval fulfillment status token, a reconnection delivery token, and a locked delivery token that are not marked as reconnected delivery tokens and not marked as locked delivery tokens. The single-interval fulfillment status token is used to retain the original fulfillment status of a single adjacent production time interval. The reconnection delivery token is used to represent the recovery delivery process formed from the beat breakpoint to the beat reconnection point. The locked delivery token is used to represent the stalled delivery process formed from the beat breakpoint to the beat lockpoint. Each box-by-box beat token is concatenated with the corresponding interval position code to form a beat position token, so that the Chronos coding prediction unit can simultaneously obtain the beat content and the temporal position.

[0150] After the beat position token sequence is input into the Transformer encoding block, each beat position token is first input into the query mapping matrix, key mapping matrix, and value mapping matrix to obtain the corresponding query vector, key vector, and value vector. For any current beat position token, its query vector is multiplied by the key vectors of each beat position token in the beat position token sequence to obtain the association score of the current beat position token relative to the total number of beat position tokens in the sequence. The association score is then divided by the square root of the key vector dimension and input into the Softmax function to obtain the normalized attention response, making the sum of all attention responses equal to 1. 1. Next, multiply each attention response by its corresponding value vector and sum the products to obtain the weighted update result of the current beat position token. After updating all beat position tokens in the beat position token sequence in the above manner, a beat context token sequence is formed. Subsequently, the beat context token sequence is input into a linear mapping layer to map each beat context token to a unified prediction dimension. The GELU activation function is used to preserve the nonlinear response in the beat changes, and LayerNorm normalization is used to stabilize the token numerical scale under different interval position numbers, finally obtaining the beat prediction context token. Through this processing, the dependencies between continuous normal fulfillment, return delivery, and locked delivery can be expressed in the same prediction context, avoiding the need to judge future fulfillment beats based solely on the most recent production time point, improving the continuity of future box order beat token generation and the stability of real-time evaluation results of order fulfillment capabilities.

[0151] In this embodiment, the future order tachometer token is restored to a future fulfillment tachometer prediction sequence through the fulfillment capability inverse quantification output unit, specifically as follows:

[0152] The fulfillment capability inverse quantification output unit reads the future box order beat token and separates the beat state code corresponding to the future beat state and the interval position code corresponding to the future production time point from the future box order beat token;

[0153] The corresponding normal performance cycle, performance cycle to be resumed, or performance cycle that is blocked is determined according to the cycle state code, and the corresponding future production time sequence is determined according to the interval position code.

[0154] Arrange the normal performance rhythm, the performance rhythm to be restored, and the performance rhythm that is hindered at each future production time point in the order of the future production time points to obtain the future performance rhythm prediction sequence.

[0155] In this invention, the output target of the Chronos coding prediction unit is the future order beat token. The future order beat token retains the concatenation structure of beat state code and interval position code to maintain the consistency of the coding system between the model prediction result and the previous order beat token sequence. If the future fulfillment beat prediction sequence is directly output in the Chronos coding prediction unit, the model prediction stage will directly enter the business state output, which will make it impossible to uniformly organize the future production time sequence and will not be convenient to standardize the connection with the discrete event fulfillment inference device. The fulfillment capability inverse quantization output unit is set up separately to restore the beat state code in the future order beat token to the normal fulfillment beat, the fulfillment beat to be restored, or the fulfillment beat that is blocked, and redetermine the future production time sequence according to the interval position code, thereby forming a future fulfillment beat prediction sequence that can be directly called by the subsequent order fulfillment capability status generation step. Through this separation design, the prediction expression within the model and the performance evaluation expression on the business side are independent of each other. This ensures the stability of the output structure of the Chronos coding prediction unit and improves the interpretability and temporal consistency of the future performance beat prediction sequence when it enters the discrete event performance inferencer.

[0156] The improved Chronos model is based on the existing Chronos time series forecasting model. It retains the core structure of predicting future sequence tokens based on historical sequence tokens, and makes business-oriented improvements for the order fulfillment cycle data of tea packaging box production factories. Existing Chronos models typically receive general time series values ​​and output future time series. This invention changes the input object to an order fulfillment beat state sequence consisting of normal fulfillment beats, fulfillment beats to be recovered, and stalled fulfillment beats. It sets up a fulfillment time series processing unit, a fulfillment status token construction unit, a Chronos encoding prediction unit, and a fulfillment capability inverse quantization output unit. During data transmission, the order fulfillment beat state sequence is first bound to interval position numbers by the fulfillment time series processing unit to form a fulfillment time series, and then the fulfillment status token construction unit generates a fulfillment status token sequence. The single-interval fulfillment status token is formed by concatenating beat state codes and interval position codes, with an input dimension of 16. The box-order beat delivery mechanism further compresses beat breakpoints, beat reconnection points, and beat locking points into reconnection delivery tokens and locking delivery tokens, forming a box-order beat token sequence. The Chronos encoding prediction unit uses Transformer encoding blocks, a linear mapping layer, a GELU activation function, LayerNorm normalization processing, and a Softmax output layer to output future box-order beat tokens. The training data comes from the factory's historical order fulfillment time-series dataset. The annotation method involves labeling normal fulfillment cycles, pending recovery cycles, and stalled fulfillment cycles according to production time intervals. The training batch size is set to 64, the learning rate to 0.0001, and the training epochs to 80. Compared with the existing Chronos model, the improvement lies in introducing order fulfillment cycle state encoding, interval position encoding, and a box order cycle delivery mechanism. This enables the model to identify the fulfillment changes of orders from normal progress to pending recovery, reconnection, or locking, improving the interpretability of future fulfillment cycle prediction sequences and the stability of real-time evaluation of order fulfillment capabilities.

[0157] In this embodiment, step seven specifically includes:

[0158] Input the future performance cycle prediction sequence into the discrete event performance inference engine, read the future performance cycle prediction sequence according to the future production time point sequence, convert the normal performance cycle into a production advancement event, convert the performance cycle to be restored into a resource succession event, and convert the stalled performance cycle into a performance stall event.

[0159] By connecting production advancement events, resource succession events, and fulfillment delay events in the order of future production time points, an order fulfillment event chain is obtained.

[0160] Count the number of production progress events, resource continuation events, and fulfillment delay events in the order fulfillment event chain, and read the last event in the order fulfillment event chain;

[0161] When the number of production advancement events exceeds the sum of the number of resource succession events and the number of performance delay events, and the last event is a production advancement event, a strong performance status is generated.

[0162] When the number of performance delay events exceeds the number of production progress events, or when the last event is a performance delay event, a weak performance status is generated.

[0163] When the conditions for generating a strong performance state and the conditions for generating a weak performance state are not met, a performance state to be restored is generated.

[0164] Strong performance status, pending recovery performance status, and weak performance status are used as the order fulfillment capability status;

[0165] In this invention, a discrete event fulfillment inference tool is used to convert the future fulfillment cycle prediction sequence from a cycle state expression to an event chain expression. This allows the order fulfillment capability status to no longer depend solely on the prediction results of a single future production point. Normal fulfillment cycles correspond to production progress events, indicating that the order is undergoing continuous processing or an increase in completion volume at a future production point. Fulfillment cycles awaiting recovery correspond to resource continuation events, indicating that although the order has not yet achieved stable progress, subsequent resources are available for continuation. Fulfillment cycles that are hindered correspond to fulfillment stagnation events, indicating that the order faces the risk of being unable to continue at a future production point. By connecting these three types of events in the order of future production points, the order can be preserved. The evolution of the fulfillment status over a future timeline; by statistically analyzing the number of production advancement events, resource continuation events, and fulfillment delay events, and combining this with the last event in the order fulfillment event chain, the overall trend and final destination of the future fulfillment process can be simultaneously reflected; when production advancement events dominate and the last event is a production advancement event, it indicates that the order's future fulfillment process will continue to move forward; when fulfillment delay events dominate or the last event is a fulfillment delay event, it indicates that the order's future fulfillment process will eventually fall into delay; when neither of the above two conditions is met, it indicates that the order is still in a recoverable but not fully stable state. This processing can improve the continuity and interpretability of the judgment of the order fulfillment capability status.

[0166] In this embodiment, step eight specifically includes:

[0167] Read the order fulfillment capability status and the corresponding order number, and generate a fulfillment order tag based on the order fulfillment capability status;

[0168] When an order's fulfillment capability status is "strong fulfillment," the fulfillment order is marked as a "committable fulfillment order." When an order's fulfillment capability status is "pending restoration of fulfillment," the fulfillment order is marked as a "pending tracking fulfillment order." When an order's fulfillment capability status is "weak fulfillment," the fulfillment order is marked as a "risky fulfillment order."

[0169] The order number, current production time, order fulfillment capability status, and fulfillment order tag are bound together to generate a real-time evaluation result of order fulfillment capability;

[0170] The real-time assessment results of order fulfillment capability are output to the order management and production planning ends for order delivery confirmation and production scheduling adjustment.

[0171] Example 1: To verify the feasibility of this invention in practice, it was applied to an order fulfillment assessment scenario in a tea packaging box manufacturing factory. This factory mainly undertakes orders for gift-boxed tea packaging boxes, drawer-type tea boxes, and hinged-lid tea boxes. The production process includes material preparation, printing, surface treatment, die-cutting, box gluing, and warehousing. Due to the large number of order batches, frequent box type changes, and significant reuse of printing plates and die-cutting molds, situations frequently arise on the production floor where orders are completed in the previous process but resources for the next process have not yet been released. Traditional scheduling systems can only assess order risk based on delivery dates and manual work reports, often only discovering delivery pressure after orders have already entered a stalled state, making it difficult to adjust production plans in a timely manner.

[0172] In this embodiment, the implementer continuously collects the order fulfillment status over 10 production days, forming a production time point every 30 minutes, and collects a total of 18,640 valid order fulfillment status records, covering 126 orders. Based on the order production status, process waiting status, resource release status, and delivery remaining status of the same order at adjacent production time points, the production progress, waiting time, resource continuation points, and delivery compression amounts are extracted to form an order fulfillment takt variation sequence. Further, adjacent production time point intervals are marked as normal fulfillment takt, pending recovery fulfillment takt, and stalled fulfillment takt, resulting in an order fulfillment takt status sequence. Subsequently, the improved Chronos model organizes the order fulfillment takt status sequence into a fulfillment time sequence, constructs a fulfillment status token sequence, and identifies takt breakpoints, takt reconnection points, and takt lockpoints through a box-order takt delivery mechanism, forming a box-order takt token sequence, which then predicts the future fulfillment takt prediction sequence. Finally, the discrete event fulfillment inferrer converts the future fulfillment takt prediction sequence into an order fulfillment event chain, outputting strong fulfillment status, pending recovery fulfillment status, and weak fulfillment status, and outputting real-time evaluation results of order fulfillment capability based on the order fulfillment capability status.

[0173] To verify the practical effectiveness of this invention, three comparative schemes were set up. Comparative scheme one is a manual scheduling and evaluation scheme, where planners assess performance risks based on order delivery dates and work reports; comparative scheme two is a traditional ERP rule-based evaluation scheme, which uses rules to determine performance based on the remaining order duration, current process, and planned completion time; comparative scheme three is a standard Chronos prediction scheme, which only uses the order performance cycle state sequence to predict future cycle times without introducing a packing slip delivery mechanism or a discrete event performance inference tool. The test data included 126 orders, of which 48 were manually verified as strongly fulfilled, 39 were pending resumption of fulfillment, and 39 were weakly fulfilled. The comparison results are shown in Table 1.

[0174] Table 1. Comparison of Order Fulfillment Capability Status Recognition Results

[0175] plan Accuracy rate of strong performance identification / % Accuracy rate of identifying pending performance issues / % Weak performance identification accuracy rate / % Average state determination delay / h Comparison Option 1 78.6 61.5 69.2 5.5 Comparison Option 2 83.3 69.2 74.4 4.3 Comparison Option 3 89.6 79.5 84.6 2.6 Method of the present invention 95.8 92.3 94.9 1.2

[0176] As shown in Table 1, the method of this invention performs best in all three aspects: strong performance identification, performance pending recovery identification, and weak performance identification. This indicates that it can not only determine whether an order can proceed stably, but also distinguish between a recoverable state and a state of continuous blockage after resource continuation. In contrast, Scheme 1 mainly relies on manual scheduling experience, with a strong performance identification accuracy of 78.6% and a performance pending recovery identification accuracy of only 61.5%, indicating that the manual method has a weak ability to identify orders in the intermediate transitional state. In contrast, Scheme 2 uses ERP rules for judgment, which improves the overall results, but still cannot characterize adjacent production. The change in beat between time points; compared with Scheme 3 using ordinary Chronos prediction, the accuracy of weak performance identification reaches 84.6%, and the average state judgment delay is reduced to 2.6 hours, indicating that time series prediction can improve the performance status judgment; the method of this invention further introduces a box list beat delivery mechanism and a discrete event performance inferencer, which makes the accuracy of strong performance identification reach 95.8%, the accuracy of performance to be recovered identification reach 92.3%, the accuracy of weak performance identification reach 94.9%, and the average state judgment delay is reduced to 1.2 hours, indicating that it can detect performance changes earlier and reduce state misjudgment.

[0177] This embodiment verifies the feasibility and effectiveness of the present invention in order fulfillment assessment at a tea packaging box manufacturing plant. By continuously collecting order fulfillment status data and extracting changes in order fulfillment cycle time, and combining an improved Chronos model incorporating a box order cycle time delivery mechanism with a discrete event fulfillment inference tool, the present invention can more accurately distinguish between strong fulfillment status, fulfillment status awaiting recovery, and weak fulfillment status. Compared with manual scheduling, traditional ERP rules, and ordinary Chronos prediction schemes, the present invention has significant advantages in identification accuracy, timeliness of status judgment, and order delivery support capabilities, providing a reliable basis for order delivery confirmation and production scheduling adjustments.

[0178] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for real-time assessment of the order fulfillment capability of tea packaging box manufacturing plants, characterized in that, Includes the following steps: Step 1: Collect order fulfillment status at continuous production points and construct an order fulfillment time series dataset; Step 2: Extract the order fulfillment cycle time variation between adjacent production time points based on the order fulfillment time series dataset to obtain the order fulfillment cycle time variation sequence; Step 3: Mark the cycle time status according to the order fulfillment cycle time change sequence to generate an order fulfillment cycle time status sequence; wherein, the cycle time status includes normal fulfillment cycle time, fulfillment cycle time to be recovered, and stalled fulfillment cycle time. The normal fulfillment cycle time is the adjacent production time point interval where the production progress is greater than zero, the fulfillment cycle time to be recovered is the adjacent production time point interval where the production progress is equal to zero and the resource continuation point has been formed, and the stalled fulfillment cycle time is the adjacent production time point interval where the production progress is equal to zero and the resource continuation point has not been formed. Step 4: Input the order fulfillment cycle time status sequence into the improved Chronos model. The improved Chronos model includes a fulfillment time series processing unit, a fulfillment status token construction unit, a Chronos encoding prediction unit, and a fulfillment capability inverse quantization output unit. The fulfillment time series processing unit generates a fulfillment time series based on the order fulfillment cycle time status sequence, and the fulfillment status token construction unit generates a fulfillment status token sequence based on the fulfillment time series. Step 5: Execute the order book delivery mechanism within the fulfillment status Token construction unit. Identify the order book breakpoint, order book reconnection point, and order book lock point. Construct a reconnection delivery Token and a lock delivery Token using the order book breakpoint as the delivery starting point. Embed the reconnection delivery Token and the lock delivery Token into the fulfillment status Token sequence to obtain the order book order book order book Token sequence. Specifically, when the order book status code changes from the code corresponding to the normal fulfillment order book to the code corresponding to the order book order book to be restored, it is marked as an order book breakpoint. When the order book status code changes from the code corresponding to the order book order book to the code corresponding to the normal fulfillment order book, it is marked as an order book reconnection point. When the order book status code changes from the code corresponding to the order book order book to the code corresponding to the order book order book to be restored, it is marked as an order book lock point. Step 6: Input the box order beat token sequence into the Chronos encoding prediction unit to generate the future box order beat token, and then use the fulfillment capability inverse quantization output unit to restore the future box order beat token into the future fulfillment beat prediction sequence. Step 7: Input the future fulfillment cycle prediction sequence into the discrete event fulfillment inference engine to generate the order fulfillment capability status; Step 8: Output the real-time evaluation results of order fulfillment capability based on the order fulfillment capability status.

2. The method for real-time evaluation of order fulfillment capability of tea packaging box manufacturing plants according to claim 1, characterized in that, Step one specifically involves: The order fulfillment status is collected at continuous production points in the tea packaging box production line. The order fulfillment status includes order production status, process waiting status, resource release status, and delivery remaining status. Arrange the order fulfillment status of the same order according to the production time point to obtain the single order fulfillment sequence; The order fulfillment time sequence corresponding to each order is summarized according to the order number to obtain the order fulfillment time sequence dataset.

3. The method for real-time evaluation of order fulfillment capability of a tea packaging box manufacturing plant according to claim 1, characterized in that, Step two specifically involves: Read the order fulfillment time sequence one by one from the order fulfillment time sequence dataset, and read the order fulfillment status of the same order at the previous production time point and the next production time point; Extract the cumulative completed quantity from the order production status, and subtract the cumulative completed quantity of the previous production time from the cumulative completed quantity of the next production time to obtain the production progress quantity; Extract the waiting process marker and waiting duration from the process waiting status. When the waiting process marker at the next production time point is the same as the waiting process marker at the previous production time point, use the waiting duration at the next production time point as the waiting dwell time. When the waiting process marker at the next production time point is different from the waiting process marker at the previous production time point, reset the waiting dwell time to the set base value. Extract the resource availability marker from the resource release status. When the resource availability marker of the previous production time point is not released and the resource availability marker of the next production time point is released, mark the corresponding adjacent production time point as the resource continuation point. Extract the remaining delivery time from the remaining delivery status, and subtract the remaining delivery time of the next production time from the remaining delivery time of the previous production time to obtain the delivery compression amount. By binding production progress, waiting time, resource connection points, and delivery compression according to adjacent production time points, the order fulfillment rhythm changes can be obtained. Arrange the changes in order fulfillment takt time of the target order at consecutive adjacent production time points in the order of production time points to obtain the order fulfillment takt time change sequence.

4. The method for real-time evaluation of order fulfillment capability of tea packaging box manufacturing plants according to claim 1, characterized in that, Step four specifically involves: The order fulfillment time sequence processing unit reads the normal fulfillment cycle, the fulfillment cycle to be recovered, and the stalled fulfillment cycle from the order fulfillment cycle status sequence, and reads the adjacent production time intervals corresponding to each cycle status; According to the order of adjacent production time intervals in the order fulfillment cycle state sequence, write the initial interval position number for the first adjacent production time interval, and increment and update the interval position number of the subsequent adjacent production time intervals along the production time sequence to obtain the corresponding interval position number. Normal performance timeframe, performance timeframe to be restored, and performance timeframe that is stalled are bound to the corresponding interval location number to generate a performance time series; The performance time series is input into the performance status token construction unit. The performance status token construction unit encodes the normal performance beat, the performance beat to be recovered, and the performance beat that is blocked in the performance time series, and encodes the interval position number to obtain the beat status code and the interval position code respectively. By concatenating the cycle state code and interval position code corresponding to the same adjacent production time interval, a single interval performance status Token is obtained. Arrange the performance status tokens of each single interval according to the interval position number order to generate a performance status token sequence.

5. The method for real-time evaluation of order fulfillment capability of a tea packaging box manufacturing plant according to claim 1, characterized in that, Step five specifically involves: The box delivery mechanism is executed, the fulfillment status token sequence is read, and the fulfillment status tokens of adjacent single intervals are read in order of interval position number. Subtract the previous order interval fulfillment status Token from the next order interval fulfillment status Token to obtain the beat difference Token. Then, concatenate the previous order interval fulfillment status Token, the next order interval fulfillment status Token, and the beat difference Token according to the channel dimension to obtain the beat migration Token. The beat transition token is input into the Sigmoid activation function to obtain the beat transition gating value, and the beat transition token is then processed by GELU activation and LayerNorm normalization to obtain the beat transition response token. Multiply the beat transition gating value element by element with the beat transition response Token to obtain the beat delivery discrimination Token; Starting from the beat breakpoint, the beat delivery discrimination tokens, including the beat breakpoint and the beat return point, are connected in order of interval position number to obtain the return delivery token. Starting from the beat breakpoint, the beat delivery discrimination tokens, including the beat breakpoint and the beat lockpoint, are connected in order of interval position number to obtain the lock delivery token. Write the return delivery token into the interval position number of the corresponding beat break point in the performance status token sequence, and mark the single interval performance status token, including the beat break point and the beat return point, as the returned delivery token. Write the locked delivery token into the interval position number of the corresponding beat breakpoint in the performance status token sequence, and mark the single interval performance status token, including the beat breakpoint and the beat lockpoint, as a locked delivery token. Retain the single-segment fulfillment status tokens that are not marked as received delivery tokens and not marked as locked delivery tokens, and arrange the retained single-segment fulfillment status tokens, received delivery tokens, and locked delivery tokens in order of segment position number to obtain the box order beat token sequence.

6. The method for real-time evaluation of order fulfillment capability of a tea packaging box manufacturing plant according to claim 1, characterized in that, The step of inputting the box order beat token sequence into the Chronos encoding prediction unit to generate future box order beat tokens is specifically as follows: The Chronos coding prediction unit reads the box-single beat token sequence in the order of interval position numbering, and reads the interval position code corresponding to each box-single beat token in the box-single beat token sequence; Each box's single beat token and its corresponding interval position code are concatenated according to the channel dimension to obtain the beat position token sequence; Input the beat position token sequence into the Transformer encoding block, calculate the attention response between each beat position token through self-attention, and update the beat position token sequence with weights based on the attention response to obtain the beat context token sequence. The beat context token sequence is input into the linear mapping layer, and then subjected to GELU activation function and LayerNorm normalization in sequence to obtain the beat prediction context token; Input the beat prediction context token into the Softmax output layer to obtain the probability of the future beat state corresponding to the future production time point; Select the beat state with the highest probability from the probabilities of future beat states as the future beat state, and read the interval position code corresponding to the future production time point; The future beat state code is concatenated with the interval position code corresponding to the future production time point to generate the future box order beat token.

7. The method for real-time evaluation of order fulfillment capability of a tea packaging box manufacturing plant according to claim 1, characterized in that, The process of restoring the future order tachometer token into a future fulfillment tachometer prediction sequence through the fulfillment capability inverse quantification output unit is specifically as follows: The fulfillment capability inverse quantification output unit reads the future box order beat token and separates the beat state code corresponding to the future beat state and the interval position code corresponding to the future production time point from the future box order beat token; The corresponding normal performance cycle, performance cycle to be resumed, or performance cycle that is blocked is determined according to the cycle state code, and the corresponding future production time sequence is determined according to the interval position code. Arrange the normal performance rhythm, the performance rhythm to be restored, and the performance rhythm that is hindered at each future production time point in the order of the future production time points to obtain the future performance rhythm prediction sequence.

8. The method for real-time evaluation of order fulfillment capability of a tea packaging box manufacturing plant according to claim 1, characterized in that, Step seven specifically involves: Input the future performance cycle prediction sequence into the discrete event performance inference engine, read the future performance cycle prediction sequence according to the future production time point sequence, convert the normal performance cycle into a production advancement event, convert the performance cycle to be restored into a resource succession event, and convert the stalled performance cycle into a performance stall event. By connecting production advancement events, resource succession events, and fulfillment delay events in the order of future production time points, an order fulfillment event chain is obtained. Count the number of production progress events, resource continuation events, and fulfillment delay events in the order fulfillment event chain, and read the last event in the order fulfillment event chain; When the number of production advancement events exceeds the sum of the number of resource succession events and the number of performance delay events, and the last event is a production advancement event, a strong performance status is generated. When the number of performance delay events exceeds the number of production progress events, or when the last event is a performance delay event, a weak performance status is generated. When the conditions for generating a strong performance state and the conditions for generating a weak performance state are not met, a performance state to be restored is generated. Strong performance status, pending recovery performance status, and weak performance status are used as the order fulfillment capability status.

9. The method for real-time evaluation of order fulfillment capability of a tea packaging box manufacturing plant according to claim 1, characterized in that, Step eight specifically involves: Read the order fulfillment capability status and the corresponding order number, and generate a fulfillment order tag based on the order fulfillment capability status; When the order fulfillment capability status is strong fulfillment status, the fulfilled order will be marked as a fulfillable order; When the order fulfillment capability status is "pending resumption of fulfillment", the fulfilled order will be marked as a fulfillment order to be tracked. When the order fulfillment capability status is weak, the fulfillment order will be marked as a risky fulfillment order; The order number, current production time, order fulfillment capability status, and fulfillment order tag are bound together to generate a real-time evaluation result of order fulfillment capability; The real-time assessment results of order fulfillment capability are output to the order management and production planning ends for order delivery confirmation and production scheduling adjustment.

Images