A method and system for predicting fiber production manufacturing costs

Abstract

This invention discloses a method and system for predicting fiber manufacturing costs, comprising: collecting manufacturing data and establishing a timestamp mapping relationship between batch spinning spinneret groups; dividing phase boundary segments according to the process sequence and statistically analyzing quality loss and resource consumption to generate entropy-increasing units; based on energy coupling and material flow direction, performing tension mapping and energy consumption conversion to form an entropy-increasing field; constructing a potential field from spinneret to quality inspection and calculating tension quality level indicators to generate a closed loop; extracting entropy-increasing gradients and fracture propagation sequences to generate multi-risk quantile cost prediction intervals; comparing the relationship between risk quantiles and thresholds, and outputting the cost collapse level and corresponding closed loop results. This invention achieves accurate prediction of fiber manufacturing cost collapse risk and anomaly source location by constructing a marketable quality entropy-increasing field and combining it with a quantile regression neural network.

Description

Technical Field

[0001] This invention relates to the field of fiber production and manufacturing technology, and in particular to a method and system for predicting fiber production and manufacturing costs. Background Technology

[0002] Fiber production typically involves continuous processes such as spinning, cooling and curing, bundling and oiling, drawing and setting, winding into tubes, and quality inspection and grading. Manufacturing costs are influenced by factors including raw material input, energy consumption, equipment operation, waste fiber generation, rework, and changes in product grade. Existing methods for predicting fiber manufacturing costs are mostly based on historical output, raw material prices, energy costs, labor hours, and equipment depreciation, using statistical calculations or conventional forecasting models to obtain batch or unit costs.

[0003] However, fiber production is characterized by continuous forming and step-by-step quality transfer. Abnormalities in the spinneret assembly, monofilament tension fluctuations, drafting deviations, insufficient oiling, and abnormal winding structures do not necessarily manifest as immediate explicit costs. Instead, they can translate into increased waste yarn, increased rework, grade downgrading, or decreased effective output in subsequent processes. Existing methods typically predict quality losses on an entire batch or process basis, making it difficult to correlate quality losses with the spinneret assembly, spinning position, winding drum, and quality control points. Furthermore, it is challenging to identify the locations of breakages that occur during quality transfer.

[0004] Existing technologies suffer from problems such as coarse cost prediction granularity, unclear sources of quality loss, insufficient conversion of effective marketable quality, and weak ability to judge the upper limit of high-risk costs. In particular, they cannot judge the risk of unit effective manufacturing cost collapse caused by the failure of input materials and energy consumption to be converted into first-class marketable quality. As a result, the prediction results tend to remain at the average cost level, making it difficult to provide accurate basis for cost anomaly location, risk warning and process adjustment in fiber production.

[0005] Therefore, how to provide a method and system for predicting fiber production costs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method and system for predicting fiber manufacturing costs. This invention constructs a mass transfer potential field that runs through the spinneret group to the quality inspection node, and performs correlation analysis on mass loss, tension transfer instability, grade deviation, and mass breakage lock-in loops during the continuous fiber forming process. It also combines an improved quantile regression neural network to generate multi-risk quantile prediction intervals for the effective manufacturing cost per unit, thereby achieving early prediction of cost collapse risks and anomaly sources. This invention has the advantages of high cost prediction accuracy, strong anomaly location capability, timely risk warning, and high adaptability to the production process.

[0007] A method for predicting fiber production manufacturing costs according to an embodiment of the present invention includes:

[0008] Collect production data of the target fiber production batch and establish a production process mapping relationship based on batch number, spinning station number, spinneret group number and timestamp.

[0009] Based on the production process mapping relationship, the production data is divided into multiple process phase boundary segments. For each process phase boundary segment, the quality loss and resource consumption parameters are statistically analyzed to generate a marketable quality entropy increase unit.

[0010] Based on the energy coupling strength and material flow direction weight of adjacent processes, tension continuity mapping and energy consumption conversion are performed on all saleable quality entropy increase units to generate a hierarchical saleable quality entropy increase field. In each unit, the cost entropy increase intensity, saleable quality loss rate and entropy increase gradient direction are recorded simultaneously.

[0011] Construct a mass transfer potential field that runs through the spinneret group to the quality inspection node, calculate the tension potential difference, mass expansion and contraction rate and grade offset gradient of the continuous streamline, the conserved loop and the grade mapping surface, and mark the fracture critical point when any index exceeds the corresponding critical threshold and aggregate to form a mass fracture lock loop.

[0012] Extract the entropy increase gradient sequence from the layered saleable quality entropy increase field and the fracture propagation sequence from the quality fracture closed loop, input them into the improved quantile regression neural network, and perform dynamic quantile constraints and tail quantile amplification on different fracture propagation stages to generate a multi-risk quantile prediction interval for the unit effective manufacturing cost of the target batch.

[0013] Compare the prediction intervals of each risk quantile with the preset unit manufacturing cost threshold. When the prediction interval of a high-risk quantile exceeds the corresponding threshold, the target fiber production batch is determined to enter the unit effective manufacturing cost collapse risk state, and the cost prediction result, cost collapse level and corresponding quality breakage lockout are output.

[0014] Optionally, the production and manufacturing data includes raw material input data, melt conveying data, spinneret forming data, cooling and solidification data, oiling treatment data, drawing and shaping data, winding into a cylinder data, equipment operation data, quality inspection data, energy consumption statistics, waste filament recycling data, rework processing data, and manufacturing cost data.

[0015] Optionally, establishing the production process mapping relationship includes:

[0016] Read the batch number, production line number, spinning station number, spinneret group number, metering pump number, winding drum number, and quality inspection number corresponding to the production batch of the target fiber;

[0017] Establish the process sequence relationship between spinneret group, spinning station, winding drum and quality inspection number according to the production time sequence;

[0018] Establish the quality transfer relationship between raw material input node, spinning node, drawing and shaping node, winding node and quality inspection and grading node according to the material flow direction;

[0019] Establish the equipment occupancy relationship among spinning equipment, drawing equipment, winding equipment, and quality inspection equipment based on equipment operating time;

[0020] By associating and mapping process succession relationships, quality transfer relationships, and equipment occupancy relationships, a production process mapping relationship corresponding to the target fiber production batch is generated.

[0021] Optionally, the unit for generating marketable quality entropy increase includes:

[0022] Based on the production process mapping relationship, read the mass loss and resource occupation parameters corresponding to the target process phase boundary segment, and write the mass loss and resource occupation parameters into the phase boundary data record in the order of incoming materials, effective output, waste filament generation, rework recycling, grade downgrade, energy consumption and equipment occupation.

[0023] Based on the quality inspection level, rework status and waste fiber status corresponding to the phase boundary segment of the target process, the saleable value of the output fibers in the phase boundary data record is converted. The first-grade output portion is determined as the complete saleable quality, the second-grade output portion, the substandard output portion and the reworked and recycled portion are converted into reduced saleable quality respectively, and the waste fiber portion is marked as unsaleable quality.

[0024] Align the incoming material in the target process phase segment with the complete marketable quality, reduced marketable quality, and unmarketable quality to identify the loss location where the input material has not been converted into marketable quality. Generate marketable quality loss marks based on the spinneret group, spinning position, drafting section, and winding drum corresponding to the loss location.

[0025] Read the tension fluctuation, stretching offset, oiling offset, winding offset and quality inspection level offset in the target process phase boundary segment, bind the marketable quality loss mark with the corresponding process offset, and generate a phase boundary entropy increase tag that records the source of loss, loss type, loss location and transmission direction;

[0026] By combining phase boundary data records, sales value conversion results, sales quality loss markers, and phase boundary entropy increase labels, a sales quality entropy increase unit corresponding to the target process phase boundary segment is generated.

[0027] Optionally, the generation of the hierarchical marketable quality entropy-increasing field, which synchronously records the cost entropy enhancement degree, marketable quality loss rate, and entropy increase gradient direction in each unit, includes:

[0028] Read the phase boundary number, process sequence, spinning position number, spinneret group number, winding drum number, effective output quality, energy consumption, equipment usage time, monofilament tension change and phase boundary entropy increase label from each available quality entropy increase unit, and establish the succession relationship between adjacent units according to the sequence of spinneret output, cooling and solidification, bundling and oiling, drawing and shaping, winding into a drum and quality inspection and grading.

[0029] Based on the connection relationship between adjacent units, the effective output quality, waste wire quality, rework quality and grade downgrade quality in the upstream available quality entropy increase unit are written into the material acceptance field of the downstream available quality entropy increase unit, and the material flow direction weight between adjacent processes is generated based on the material acceptance result.

[0030] Read the records of monofilament tension change, drafting offset, winding speed change and broken joint in adjacent saleable mass entropy increase units, align the tension change of the upstream unit with the tension change of the downstream unit, generate tension continuity mapping results, and write the tension instability position into the tension transmission field of the corresponding unit.

[0031] Read the energy consumption, equipment usage time, equipment continuous operation status and process continuity time in adjacent available quality entropy increase units, convert the energy consumption corresponding to the same spinning position, the same spinneret group and the same winding drum to the corresponding available quality entropy increase unit, and generate the energy coupling strength between adjacent processes.

[0032] Based on the material flow direction weight, tension continuity mapping results, and energy coupling strength, each saleable quality entropy increase unit is divided into a spinneret forming layer, a cooling and curing layer, an oiling and bundling layer, a drawing and shaping layer, a winding and spinning layer, and a quality inspection and grading layer. In each layer, the cost entropy increase degree, saleable quality loss rate, and entropy increase gradient direction are recorded to generate a layered saleable quality entropy increase field.

[0033] Optionally, marking the fracture critical point and agglomerating it to form a mass fracture lock-in loop includes:

[0034] Read the available quality entropy increase units corresponding to each layer in the layered available quality entropy increase field, and extract the spinneret group number, spinning position number, single filament bundle identifier, drawing section identifier, winding drum number, quality inspection grade, material flow direction weight, tension continuity mapping result, energy coupling strength, phase boundary entropy increase label, cost entropy increase degree, and entropy increase gradient direction.

[0035] Using the spinneret group, monofilament bundle, drawing section, winding drum, and quality inspection grade as nodes of the mass transfer potential field, a potential field connection edge is established based on the filament output connection relationship from the spinneret group to the monofilament bundle, the tension transfer relationship from the monofilament bundle to the drawing section, the forming connection relationship from the drawing section to the winding drum, and the grade mapping relationship from the winding drum to the quality inspection grade, thereby generating a mass transfer potential field that runs through the spinneret group to the quality inspection node.

[0036] In the mass transfer potential field, the potential field connecting edges that are continuously connected along the spinneret group, monofilament bundle, drawing section and winding drum are combined into a continuous streamline. The closed correspondence between the mass of incoming material, effective output mass, waste yarn mass, rework mass and grade downgrade mass is combined into a conserved loop. The correspondence between fineness, strength, elongation, yarn unevenness and oil content and quality inspection grade is combined into a grade mapping surface.

[0037] For continuous streamlines, the tension difference is obtained by comparing the changes in monofilament tension and the tension continuity mapping results corresponding to adjacent potential field nodes. For conserved loops, the quality closure deviation between the quality of incoming material and effective output quality, waste filament quality, rework quality and grade downgrade quality is obtained. For grade mapping surfaces, the deviation direction and deviation magnitude between each quality inspection index and the target grade boundary are compared to obtain the grade offset gradient.

[0038] The tension potential difference, mass expansion and contraction rate, and grade offset gradient are compared with the corresponding preset critical thresholds. The fracture critical point is marked at the position that exceeds the corresponding preset critical threshold. According to the correlation relationship of the same spinneret group, the same spinning position, the same winding drum, and the continuous process phase boundary, the fracture critical point, the associated potential field connection edge, the over-limit index, the phase boundary entropy increase label, and the corresponding marketable mass entropy increase unit are aggregated to form a mass fracture lock-in loop.

[0039] Optionally, the generation of the multi-risk quantile prediction interval for the effective manufacturing cost per unit of the target batch includes:

[0040] Read the cost entropy enhancement degree, the rate of loss of marketable quality, the direction of entropy increase gradient, the weight of material flow direction, the result of tension continuity mapping and the energy coupling strength in the layered marketable quality entropy increase field, and generate an entropy increase gradient sequence in the order of spinneret forming layer, cooling and curing layer, oiling and bundling layer, stretching and shaping layer, winding and spinning layer and quality inspection and grading layer.

[0041] Read the fracture critical point, associated potential field connection edge, over-limit index, phase boundary entropy increase label, corresponding marketable quality entropy increase unit, spinneret group number, spinning position number and winding drum number in the quality fracture lock loop, and generate a fracture propagation sequence in the order of fracture initiation point, fracture transmission node, fracture closure node and quality inspection mapping node.

[0042] The entropy-increasing gradient sequence and the fracture propagation sequence are input into the improved quantile regression neural network, which includes an entropy-increasing input gate, a fracture stage gate, a quantile sharing mapping layer, a tail quantile amplification layer, and a monotonic quantile correction layer.

[0043] The entropy increase input gate reads the cost entropy increase degree, the rate of loss of marketable quality and the direction of entropy increase gradient. The fracture stage gate reads the number of fracture critical points, fracture propagation length, the coverage of the lock-in loop and the quality inspection level offset status. The reading results are input into the quantile sharing mapping layer to generate the basic quantile representation of the unit effective manufacturing cost.

[0044] Based on the fracture initiation stage, fracture propagation stage, fracture closure stage and quality inspection mapping stage of the fracture propagation sequence, dynamic quantile constraints are applied to the basic quantile representation, and risk amplification is performed on the cost representation corresponding to the 90th and 95th quantiles through the tail quantile amplification layer.

[0045] By sequentially correcting the cost representations corresponding to the 10th, 50th, 80th, 90th, and 95th quantiles through a monotonic quantile correction layer, the unit effective manufacturing cost corresponding to the subsequent quantile is not lower than the unit effective manufacturing cost corresponding to the preceding quantile, thereby generating a multi-risk quantile prediction interval for the unit effective manufacturing cost of the target fiber production batch.

[0046] Optionally, the output cost prediction results, cost collapse level, and corresponding quality fracture lock-in loop include:

[0047] Read the multi-risk quantile prediction interval of the unit effective manufacturing cost of the target fiber production batch, and extract the median quantile prediction interval, high-risk quantile prediction interval, extreme-risk quantile prediction interval and the corresponding upper boundary of the quantile interval;

[0048] Read the preset unit manufacturing cost threshold, preset high-risk cost threshold and preset extreme cost threshold, compare the median quantile prediction interval, high-risk quantile prediction interval and extreme-risk quantile prediction interval with the corresponding threshold respectively, and generate quantile exceedance markers;

[0049] Read the quantile over-limit marker, the number of quality breakage lock rings, the number of spinning positions covered by lock rings, the number of winding drums covered by lock rings, and the corresponding cost entropy enhancement. When the high-risk quantile prediction interval or the extreme-risk quantile prediction interval exceeds the corresponding threshold, mark the target fiber production batch as a batch with unit effective manufacturing cost collapse risk, and generate cost collapse level according to the over-limit quantile level.

[0050] The cost collapse level is correlated with the quality breakage lockout ring that triggers the over-limit, and the cost prediction results, unit effective manufacturing cost risk range, cost collapse level, trigger spinneret group, trigger spinning position, trigger winding drum and corresponding quality breakage lockout ring for the target fiber production batch are output.

[0051] A fiber production cost prediction system according to an embodiment of the present invention includes:

[0052] The data mapping module is used to collect production and manufacturing data and establish mapping relationships for the production process;

[0053] The entropy increase unit module is used to divide the process phase boundary segments according to the production process mapping relationship and generate marketable quality entropy increase units;

[0054] The entropy increase field module is used to perform tension continuity mapping and energy consumption conversion on the saleable mass entropy increase unit to generate a layered saleable mass entropy increase field.

[0055] The lock-in loop module is used to construct the mass transfer potential field and generate the mass break lock-in loop;

[0056] The quantile prediction module is used to extract the entropy increase gradient sequence and the fracture propagation sequence and generate multi-risk quantile prediction intervals.

[0057] The collapse determination module is used to determine the collapse risk status of the unit effective manufacturing cost based on the multi-risk quantile prediction interval and output the corresponding results.

[0058] The beneficial effects of this invention are:

[0059] This invention establishes a production process mapping relationship, associating and mapping continuous processes such as spinning, cooling and curing, bundling and oiling, drawing and shaping, winding into tubes, and quality inspection and grading. It also generates marketable quality entropy increase units for each process phase segment, enabling the unified recording of the correspondence between raw material input, waste fiber generation, rework, grade downgrading, energy consumption, and equipment usage during the production process. This allows for precise tracking of the location and source of marketable quality loss during fiber production, improving the consistency between manufacturing cost prediction and actual production status.

[0060] This invention constructs a mass transfer potential field that runs through the spinneret assembly to the quality inspection node, performs correlation analysis on continuous streamlines, conserved loops, and grade mapping surfaces, and generates a mass fracture lock-in loop by combining tension potential difference, mass expansion and contraction rate, and grade offset gradient. This enables continuous location of problems such as spinneret assembly anomalies, tension transfer instability, drafting offset, insufficient oiling, and winding structure anomalies along the process chain. It solves the problems of existing technologies that are difficult to identify the location of mass transfer fractures and cannot determine the propagation path of cost anomalies, thus improving the ability to identify production anomalies and locate the source of costs.

[0061] This invention extracts the entropy increase gradient sequence from the layered marketable mass entropy increase field and the fracture propagation sequence from the mass fracture lock loop, and inputs them into an improved quantile regression neural network to generate a multi-risk quantile prediction interval for unit effective manufacturing cost. This allows the cost prediction results to no longer be limited to a single average value, but to simultaneously output high-risk cost intervals and extreme risk intervals, thereby achieving early warning of the risk of unit effective manufacturing cost collapse and improving the accuracy of fiber production manufacturing cost prediction, risk identification capability, and production process control capability. Attached Figure Description

[0062] 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:

[0063] Figure 1 This is a flowchart of a fiber production cost prediction method proposed in this invention;

[0064] Figure 2 This is a schematic diagram of the structure of a fiber production cost prediction system proposed in this invention. Detailed Implementation

[0065] 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.

[0066] refer to Figure 1 A method for predicting fiber production manufacturing costs, comprising:

[0067] Collect production data of the target fiber production batch and establish a production process mapping relationship based on batch number, spinning station number, spinneret group number and timestamp.

[0068] Based on the production process mapping relationship, the production data is divided into multiple process phase boundary segments. For each process phase boundary segment, the quality loss and resource consumption parameters are statistically analyzed to generate a marketable quality entropy increase unit.

[0069] Based on the energy coupling strength and material flow direction weight of adjacent processes, tension continuity mapping and energy consumption conversion are performed on all saleable quality entropy increase units to generate a hierarchical saleable quality entropy increase field. In each unit, the cost entropy increase intensity, saleable quality loss rate and entropy increase gradient direction are recorded simultaneously.

[0070] Construct a mass transfer potential field that runs through the spinneret group to the quality inspection node, calculate the tension potential difference, mass expansion and contraction rate and grade offset gradient of the continuous streamline, the conserved loop and the grade mapping surface, and mark the fracture critical point when any index exceeds the corresponding critical threshold and aggregate to form a mass fracture lock loop.

[0071] Extract the entropy increase gradient sequence from the layered saleable quality entropy increase field and the fracture propagation sequence from the quality fracture closed loop, input them into the improved quantile regression neural network, and perform dynamic quantile constraints and tail quantile amplification on different fracture propagation stages to generate a multi-risk quantile prediction interval for the unit effective manufacturing cost of the target batch.

[0072] Compare the prediction intervals of each risk quantile with the preset unit manufacturing cost threshold. When the prediction interval of a high-risk quantile exceeds the corresponding threshold, the target fiber production batch is determined to enter the unit effective manufacturing cost collapse risk state, and the cost prediction result, cost collapse level and corresponding quality breakage lockout are output.

[0073] In this embodiment, the production data includes raw material input data, melt conveying data, spinneret forming data, cooling and solidification data, oiling treatment data, drawing and shaping data, winding into a cylinder data, equipment operation data, quality inspection data, energy consumption statistics, waste filament recycling data, rework processing data, and manufacturing cost data.

[0074] In this embodiment, establishing the production process mapping relationship includes:

[0075] Read the batch number, production line number, spinning station number, spinneret group number, metering pump number, winding drum number, and quality inspection number corresponding to the production batch of the target fiber;

[0076] Establish the process sequence relationship between spinneret group, spinning station, winding drum and quality inspection number according to the production time sequence;

[0077] Establish the quality transfer relationship between raw material input node, spinning node, drawing and shaping node, winding node and quality inspection and grading node according to the material flow direction;

[0078] Establish the equipment occupancy relationship among spinning equipment, drawing equipment, winding equipment, and quality inspection equipment based on equipment operating time;

[0079] By associating and mapping process succession relationships, quality transfer relationships, and equipment occupancy relationships, a production process mapping relationship corresponding to the target fiber production batch is generated.

[0080] In this embodiment, the unit for generating marketable quality entropy increase includes:

[0081] Based on the production process mapping relationship, read the mass loss and resource occupation parameters corresponding to the target process phase boundary segment, and write the mass loss and resource occupation parameters into the phase boundary data record in the order of incoming materials, effective output, waste filament generation, rework recycling, grade downgrade, energy consumption and equipment occupation.

[0082] Based on the quality inspection level, rework status and waste fiber status corresponding to the phase boundary segment of the target process, the saleable value of the output fibers in the phase boundary data record is converted. The first-grade output portion is determined as the complete saleable quality, the second-grade output portion, the substandard output portion and the reworked and recycled portion are converted into reduced saleable quality respectively, and the waste fiber portion is marked as unsaleable quality.

[0083] Align the incoming material in the target process phase segment with the complete marketable quality, reduced marketable quality, and unmarketable quality to identify the loss location where the input material has not been converted into marketable quality. Generate marketable quality loss marks based on the spinneret group, spinning position, drafting section, and winding drum corresponding to the loss location.

[0084] Read the tension fluctuations, stretching offsets, oiling offsets, winding offsets, and quality inspection grade offsets in the target process phase boundary segment. Bind the marketable quality loss markers to the corresponding process offsets to generate phase boundary entropy increase tags that record the loss source, loss type, loss location, and transmission direction. Specifically, the generation of these tags involves:

[0085] The root mean square value of the tension signal is calculated using a 30s sliding window and then compared with the reference tension. The absolute value of the difference is marked as T0 if it is less than 0.20cN, T1 if it is between 0.20cN and 0.50cN, and T2 if it is greater than 0.50cN.

[0086] The measured values ​​of the draw ratio are taken as the range per minute and compared with the set value. A range less than 0.04 is marked as D0, between 0.04 and 0.10 is marked as D1, and greater than 0.10 is marked as D2.

[0087] The oiling rate is calculated by taking a 5-minute moving average and comparing it with the set value. If the absolute value of the deviation is less than 0.3g / 1000m, it is marked as O0; if it is between 0.3 and 0.8g / 1000m, it is marked as O1; and if it is greater than 0.8g / 1000m, it is marked as O2.

[0088] The frequency of winding speed compensation is counted every 10 minutes. If the frequency is no more than 4 times, it is marked as W0; if it is between 5 and 10 times, it is marked as W1; and if it is more than 10 times, it is marked as W2. At the same time, the quality inspection grade deviation is based on the first grade product. If it is downgraded by one grade, it is marked as Q1; if it is downgraded by two grades, it is marked as Q2; and if it is reworked, it is marked as Q3.

[0089] The tension, drafting, oiling, winding and grade offset codes are spliced ​​according to the five-segment character encoding format of T×D×O×W×Q. The source spinneret group, target winding cylinder position index and upstream and downstream transmission direction identifiers are added to generate a complete phase boundary entropy increase label.

[0090] The phase boundary data records, the sales value conversion results, the sales quality loss markers, and the phase boundary entropy increase labels are combined to generate the sales quality entropy increase unit corresponding to the target process phase boundary segment. Specifically, generating the sales quality entropy increase unit corresponding to the target process phase boundary segment involves:

[0091] Read the incoming material quality, effective output quality, waste wire quality, rework quality, grade degradation quality, energy consumption and equipment usage time of the target phase boundary segment, and write them into the seven-dimensional original feature vector in the order of incoming, output, waste wire, rework, degradation, energy consumption and equipment.

[0092] According to the conversion factors of first-class products, second-class products, substandard products and rework and recycling, the effective output quality, grade downgrade quality and rework quality are converted into a four-dimensional converted quality vector, which is then concatenated with the seven-dimensional original feature vector to obtain an eleven-dimensional converted value vector.

[0093] Calculate the quality difference between the quality of incoming materials and the converted saleable quality. Then, perform percentage processing on the proportion of the quality difference to the quality of incoming materials to obtain the quality loss rate. Write the quality loss rate, waste wire quality ratio, rework quality ratio, and grade downgrade quality ratio into a four-dimensional loss rate vector.

[0094] Read the tension continuity mapping results of the spinneret group-drawing section-winding drum, and use 0, 1, and 2 to represent stable, fluctuating, and unstable states for the tension continuity level; read the energy consumption conversion level and use 0, 1, and 2 to represent normal, high, and ultra-high states; write the tension level and energy consumption level into the two-dimensional resource state vector;

[0095] The eleven-dimensional saleable value conversion vector, the four-dimensional loss rate vector, and the two-dimensional resource state vector are sequentially concatenated into a seventeen-dimensional feature record. Phase boundary number, batch number, spinning position number, spinneret group number, winding drum number, and time segment are added to the end of the record to form a twenty-three-dimensional data entry. This entry is written into the saleable quality entropy increase unit table to generate the saleable quality entropy increase unit corresponding to the target process phase boundary segment.

[0096] In this embodiment, the generation of a hierarchical marketable quality entropy-increasing field, which synchronously records the cost entropy enhancement degree, marketable quality loss rate, and entropy increase gradient direction in each unit, includes:

[0097] Read the phase boundary number, process sequence, spinning position number, spinneret group number, winding drum number, effective output quality, energy consumption, equipment usage time, monofilament tension change and phase boundary entropy increase label from each available quality entropy increase unit, and establish the succession relationship between adjacent units according to the sequence of spinneret output, cooling and solidification, bundling and oiling, drawing and shaping, winding into a drum and quality inspection and grading.

[0098] Based on the connection relationship between adjacent units, the effective output quality, waste wire quality, rework quality, and grade downgrade quality in the upstream available quality entropy increase unit are written into the material acceptance field of the downstream available quality entropy increase unit. Material flow direction weights between adjacent processes are generated based on the material acceptance results. Specifically, the generation of material flow direction weights between adjacent processes based on the material acceptance results is as follows:

[0099] After writing the acceptance field, calculate the percentage of the four types of acceptance volume in the upstream effective output quality;

[0100] If the effective acceptance rate is above 97% and the total loss rate of waste wire, rework, and downgrade is not higher than 2%, the weight of material flow for adjacent process pairings will be set to 0.90.

[0101] If the effective acceptance rate is between 92% and 97% and the total loss rate is between 2% and 6%, the material flow weight will be set to 0.70.

[0102] If the effective acceptance rate is less than 92% or the total loss rate is greater than 6%, the material flow weight will be set to 0.45, and an inefficiency mark will be added to the pairing record.

[0103] When there are multiple upstream-downstream pairings in the same time slice, they are sorted from high to low according to the weight of the material flow direction. The weights of the first two are recorded for tension continuity mapping and energy consumption conversion. The remaining pairings only retain the acceptance volume and do not participate in weight propagation.

[0104] Read the records of monofilament tension changes, drafting offsets, winding speed changes, and broken joints from adjacent marketable mass entropy increase units. Align the tension changes of the upstream unit with those of the downstream unit to generate a tension continuity mapping result. Write the tension instability location into the tension transfer field of the corresponding unit. Specifically, the generation of the tension continuity mapping result is as follows:

[0105] Read the mean, maximum, minimum and standard deviation of the single filament tension of the upstream available quality entropy increase unit in the same time slice, and generate an upstream tension quadruple in the order of mean, maximum, minimum and standard deviation;

[0106] Read the mean, maximum, minimum and standard deviation of the single filament tension in the downstream marketable quality entropy increase unit within the same time slice and one subsequent sampling time slice, and generate downstream tension quadruples in the same order;

[0107] Calculate the absolute value of the difference between the corresponding position values ​​in the upstream tension quadruple set and the downstream tension quadruple set, and sum the four absolute values ​​of the difference to obtain the tension position difference;

[0108] Read the draw ratio offset, winding speed compensation times, number of head breaks, and number of joints of the downstream unit. When the tension alignment difference does not exceed 0.40 and the number of head breaks and joints does not exceed 2, mark the tension continuity mapping result as C0; when the tension alignment difference is greater than 0.40 but not more than 0.90, or the sum of the number of head breaks and joints is between 3 and 6, mark the tension continuity mapping result as C1; when the tension alignment difference is greater than 0.90, or the sum of the number of head breaks and joints is greater than 6, mark the tension continuity mapping result as C2.

[0109] Write the tension continuity mapping result, the corresponding upstream unit number, the corresponding downstream unit number, the tension alignment difference, the draw ratio offset, the number of winding speed compensations, the number of breakages, and the number of joints into the tension transmission field of the downstream saleable mass entropy increase unit, and determine the downstream unit position marked as C2 as the tension instability position.

[0110] Read the energy consumption, equipment usage time, continuous operation status, and process continuity time from adjacent available-quality entropy increase units. Convert the energy consumption corresponding to the same spinning position, the same spinneret group, and the same winding drum to the corresponding available-quality entropy increase unit to generate the energy coupling strength between adjacent processes. Specifically, the generation of the energy coupling strength between adjacent processes involves:

[0111] Read the energy consumption, equipment usage time, and continuous operation status of the upstream available quality entropy increase unit; read the energy consumption, equipment usage time, and continuous operation status of the downstream available quality entropy increase unit; and confirm that the two correspond to the same spinning position, the same spinneret group, or the same winding drum.

[0112] Add the upstream energy consumption to the downstream energy consumption to get the total energy consumption of adjacent process pairings; add the upstream equipment usage time to the downstream equipment usage time to get the total usage time of adjacent process pairings.

[0113] Read the end time of the upstream process and the start time of the downstream process, calculate the connection interval between them, mark it as continuous connection when the connection interval is no more than 5 minutes, mark it as delayed connection when the connection interval is greater than 5 minutes but no more than 15 minutes, and mark it as discontinuous connection when the connection interval is greater than 15 minutes.

[0114] When the total energy consumption of adjacent process pairing does not exceed 105% of the historical average of the same process section in the production line, and the total occupancy time of adjacent process pairing does not exceed 105% of the historical average, and the connection interval is marked as continuous connection, the energy coupling strength is marked as E0.

[0115] When the total energy consumption of adjacent process pairings is greater than 105% but not more than 120% of the historical average, or the total time occupied by adjacent process pairings is greater than 105% but not more than 120% of the historical average, or the connection interval is marked as delayed connection, the energy coupling strength is marked as E1.

[0116] When the total energy consumption of adjacent process pairing is greater than 120% of the historical average, or the total time occupied by adjacent process pairing is greater than 120% of the historical average, or the connection interval is marked as discontinuous connection, the energy coupling strength is marked as E2, and the corresponding upstream unit number, downstream unit number, total pairing energy consumption, total pairing time occupied, connection interval and energy coupling strength are written into the resource transfer field.

[0117] Based on material flow direction weights, tension continuity mapping results, and energy coupling strength, each marketable quality entropy increase unit is divided into a spinneret forming layer, a cooling and curing layer, an oiling and bundling layer, a drawing and shaping layer, a winding and spinning layer, and a quality inspection and grading layer. In each layer, the cost entropy increase intensity, marketable quality loss rate, and entropy increase gradient direction are recorded to generate a layered marketable quality entropy increase field, where:

[0118] Each marketable quality entropy-increasing unit is divided into a spinneret forming layer, a cooling and curing layer, an oiling and bundling layer, a drawing and shaping layer, a winding and bobbin forming layer, and a quality inspection and grading layer, specifically:

[0119] Read the phase boundary number, process equipment number, sampling time period, and phase boundary entropy increase label of each marketable quality entropy increase unit; when the phase boundary number corresponds to the spinneret, metering pump, or melt pressure sampling section, write the unit into the spinneret forming layer; when the phase boundary number corresponds to the side blowing, ring blowing, or cooling wind speed sampling section, write the unit into the cooling and curing layer; when the phase boundary number corresponds to the oil nozzle, oil roller, or oiling rate sampling section, write the unit into the oiling bundle layer; when the phase boundary number corresponds to the drawing roller, hot roller, or drawing ratio sampling section, write the unit into the drawing and shaping layer; when the phase boundary number corresponds to the winding head, bobbin, or winding speed sampling section, write the unit into the winding bobbin layer; when the phase boundary number corresponds to the fineness, strength, elongation, evenness, oil content, filament quantity, or finished product grade detection section, write the unit into the quality inspection and grading layer.

[0120] Generate a hierarchical, marketable quality entropy-increasing field, specifically as follows:

[0121] According to the same batch number, same spinning position number, same spinneret group number, and same winding drum number, the corresponding marketable quality entropy increase units in the spinneret forming layer, cooling and curing layer, oiling and bundling layer, drawing and shaping layer, winding and grading layer, and quality inspection layer are connected in chronological order of sampling time. The material flow direction weight, tension continuity mapping result, and energy coupling strength between two adjacent marketable quality entropy increase units are read. When the material flow direction weight is not less than 0.70, the tension continuity mapping result is C0 or C1, and the energy coupling strength is E0 or E1, an entropy increase transmission edge is established between the two adjacent units. When the material flow direction weight is 0.45, the tension continuity mapping result is C2, or the energy coupling strength is E2, an entropy increase risk transmission edge is established between the two adjacent units. The cost entropy enhancement degree, marketable quality loss rate, and entropy increase gradient direction are written into each unit. The six process layers, entropy increase transmission edges, and entropy increase risk transmission edges are combined to generate a layered marketable quality entropy increase field.

[0122] In this embodiment, marking the fracture critical point and aggregating to form a mass fracture lock-up loop includes:

[0123] Read the available quality entropy increase units corresponding to each layer in the layered available quality entropy increase field, and extract the spinneret group number, spinning position number, single filament bundle identifier, drawing section identifier, winding drum number, quality inspection grade, material flow direction weight, tension continuity mapping result, energy coupling strength, phase boundary entropy increase label, cost entropy increase degree, and entropy increase gradient direction.

[0124] Using the spinneret group, monofilament bundle, drawing section, winding drum, and quality inspection grade as nodes of the mass transfer potential field, potential field connection edges are established based on the filament output connection relationship from the spinneret group to the monofilament bundle, the tension transfer relationship from the monofilament bundle to the drawing section, the forming connection relationship from the drawing section to the winding drum, and the grade mapping relationship from the winding drum to the quality inspection grade. This generates a mass transfer potential field penetrating from the spinneret group to the quality inspection node. Specifically, the generation of this mass transfer potential field penetrating from the spinneret group to the quality inspection node is as follows:

[0125] Read the spinneret group number, single filament bundle identifier, drafting section identifier, winding drum number, quality inspection grade number, sampling time period, effective output quality, average single filament tension, drafting ratio, winding speed, and quality inspection grade corresponding to each unit in the layered saleable quality entropy increase field. Convert the filament output records with non-zero effective output quality within a continuous sampling time period under the same spinneret group number into spinneret group nodes, and establish a filament output receiving edge between the spinneret group node and the corresponding single filament bundle identifier; read the average single filament tension corresponding to the single filament bundle identifier and the drafting ratio corresponding to the drafting section identifier. When the two belong to the same spinning position and the sampling time period is continuous, establish a tension transfer edge between the single filament bundle node and the drafting section node; read the effective output quality corresponding to the drafting section identifier. Output quality and winding speed corresponding to the winding drum number. When the effective output quality is written to the corresponding winding drum and the winding speed has continuous records, establish a forming receiving edge between the drafting section node and the winding drum node; read the fineness, strength, elongation, evenness, oil content, filament quantity and finished product grade corresponding to the winding drum number. When the winding drum number is consistent with the quality inspection grade number, establish a grade mapping edge between the winding drum node and the quality inspection grade node; connect the spinneret group node, monofilament bundle node, drafting section node, winding drum node, quality inspection grade node and the corresponding filament exit receiving edge, tension transmission edge, forming receiving edge and grade mapping edge in production time sequence to generate a mass transfer potential field that runs through the spinneret group to the quality inspection node;

[0126] In the mass transfer potential field, the potential field connecting edges that are continuously connected along the spinneret group, monofilament bundle, drawing section and winding drum are combined into a continuous streamline. The closed correspondence between the mass of incoming material, effective output mass, waste yarn mass, rework mass and grade downgrade mass is combined into a conserved loop. The correspondence between fineness, strength, elongation, yarn unevenness and oil content and quality inspection grade is combined into a grade mapping surface.

[0127] For continuous streamlines, the tension potential difference is obtained by comparing the changes in monofilament tension and the tension continuity mapping results corresponding to adjacent potential field nodes. For conserved loops, the quality closure deviation between the incoming material quality and the effective output quality, waste filament quality, rework quality, and grade degradation quality is compared to obtain the quality expansion / contraction rate. For grade mapping surfaces, the grade offset gradient is obtained by comparing the deviation direction and deviation magnitude between each quality inspection index and the target grade boundary.

[0128] The tension potential difference is obtained by reading the mean tension, maximum tension, minimum tension, and standard deviation of the single filament corresponding to two adjacent potential field nodes in the continuous streamline. The absolute value of the difference between the four tension data of the preceding and following nodes is calculated. The absolute values ​​of the four differences are added together to generate the node tension difference. The tension continuity mapping result between adjacent nodes is read. When the mapping result is C0, the node tension difference remains unchanged. When the mapping result is C1, the node tension difference is multiplied by 1.3. When the mapping result is C2, the node tension difference is multiplied by 1.8 to obtain the tension potential difference between adjacent nodes. All the tension potential differences between adjacent nodes in the same continuous streamline are arranged in the order of production time to generate the tension potential difference sequence of the continuous streamline.

[0129] The quality expansion / contraction rate is obtained as follows: Read the incoming material quality, effective output quality, waste material quality, rework quality, and grade downgrade quality corresponding to the conservation loop; add the effective output quality, waste material quality, rework quality, and grade downgrade quality to generate a traceable total quality; calculate the absolute value of the difference between the incoming material quality and the traceable total quality; divide the absolute value of the difference by the incoming material quality to obtain the quality closure deviation ratio; when the quality closure deviation ratio does not exceed 2%, the quality expansion / contraction rate is marked as M0; when the quality closure deviation ratio is greater than 2% but not more than 5%, the quality expansion / contraction rate is marked as M1; when the quality closure deviation ratio is greater than 5%, the quality expansion / contraction rate is marked as M2.

[0130] The grade offset gradient is obtained by: reading the fineness, strength, elongation, evenness, oil content, and filament quantity corresponding to the winding drum in the grade mapping surface; reading the allowable ranges for fineness, strength, elongation, evenness, oil content, and filament quantity corresponding to the target grade; judging whether each quality inspection indicator falls within the corresponding allowable range; if it falls within the allowable range, it is recorded as 0; if it is below the lower limit of the allowable range or above the upper limit of the allowable range, it is recorded as 1; adding the six judgment results to obtain the grade offset item number; when the grade offset item number is 0, the grade offset gradient is marked as Q0; when the grade offset item number is 1 to 2, the grade offset gradient is marked as Q1; when the grade offset item number is 3 or more, the grade offset gradient is marked as Q2.

[0131] Tension potential difference, mass expansion / contraction rate, and grade offset gradient are compared with corresponding preset critical thresholds. Breakage critical points are marked at locations exceeding these thresholds. Based on the correlation relationships within the same spinneret group, the same spinning position, the same winding drum, and continuous process phase boundaries, the breakage critical points, associated potential field connection edges, out-of-limit indicators, phase boundary entropy increase labels, and corresponding marketable mass entropy increase units are aggregated to form a mass breakage lock-in loop. Specifically, the breakage critical points are marked at locations exceeding the corresponding preset critical thresholds.

[0132] Read the tension potential difference sequence in the continuous streamline, the mass expansion / contraction rate marker in the conserved loop, and the grade offset gradient marker in the grade mapping surface; when the tension potential difference between adjacent potential field nodes is greater than 0.90, or the tension continuity mapping result is C2, mark the connection edge between adjacent potential field nodes as the tension breakage critical point; when the mass expansion / contraction rate is marked as M2, mark the position between the incoming material node and the traceable total mass node in the corresponding conserved loop as the mass conservation breakage critical point; when the grade offset gradient is marked as Q2, mark the grade mapping edge between the corresponding winding drum node and the quality inspection grade node as the grade breakage critical point; read the batch number, spinneret group number, spinning position number, winding drum number, and sampling time period corresponding to each breakage critical point, and write the breakage critical points with the same number and sampling time period or consecutively before and after into the same critical point set; arrange the breakage critical points in the same critical point set in the order of spinneret group node, monofilament bundle node, drawing section node, winding drum node, and quality inspection grade node, and write the arranged node positions into the breakage critical point record;

[0133] A quality fracture lock-in loop is formed, specifically as follows:

[0134] Connect the fracture critical points of the same batch number and whose process continuity time interval between adjacent nodes in the fracture critical point set sequence does not exceed 10 s to generate a fracture critical chain.

[0135] For the same spinneret group with a critical chain length of no less than 3 nodes, check whether the winding drum numbers corresponding to the first node and the last node are the same; if they are the same, mark the connecting edge between the first node and the last node as a closed edge to form a ring structure.

[0136] Read the phase boundary entropy increase tags corresponding to each fracture critical point in the ring structure, filter the tags with tension continuity mapping result of C2 or energy coupling strength of E2, and count their number; when the number of filtered tags accounts for more than 50% of the total number of nodes in the ring structure, the ring structure is confirmed as an effective locking structure.

[0137] In an effective locked structure, read the type of over-limit index, and write the tension fracture critical point, mass conservation fracture critical point and grade fracture critical point into the tension fracture set, the mass conservation fracture set and the grade fracture set respectively, and record the number of nodes in each fracture set.

[0138] When all three types of fracture sets are not empty and the interval between any two types of fracture set nodes in the ring structure does not exceed two connecting edges, the ring structure is determined as a mass fracture lock-in loop, and all node numbers, connecting edge numbers, over-limit indicators, phase boundary entropy increase labels, and corresponding salable mass entropy increase unit numbers in the lock-in loop are written into the lock-in loop record table.

[0139] In this embodiment, the generation of the multi-risk quantile prediction interval for the effective manufacturing cost per unit of the target batch includes:

[0140] Read the cost entropy enhancement degree, the rate of loss of marketable quality, the direction of entropy increase gradient, the weight of material flow direction, the result of tension continuity mapping and the energy coupling strength in the layered marketable quality entropy increase field, and generate an entropy increase gradient sequence in the order of spinneret forming layer, cooling and curing layer, oiling and bundling layer, stretching and shaping layer, winding and spinning layer and quality inspection and grading layer.

[0141] Read the fracture critical point, associated potential field connection edge, over-limit index, phase boundary entropy increase label, corresponding marketable quality entropy increase unit, spinneret group number, spinning position number and winding drum number in the quality fracture lock loop, and generate a fracture propagation sequence in the order of fracture initiation point, fracture transmission node, fracture closure node and quality inspection mapping node.

[0142] The entropy-increasing gradient sequence and the fracture propagation sequence are input into the improved quantile regression neural network, which includes an entropy-increasing input gate, a fracture stage gate, a quantile sharing mapping layer, a tail quantile amplification layer, and a monotonic quantile correction layer.

[0143] The entropy increase input gate reads the cost entropy increase degree, the rate of loss of marketable quality and the direction of entropy increase gradient. The fracture stage gate reads the number of fracture critical points, fracture propagation length, the coverage of the lock-in loop and the quality inspection level offset status. The reading results are input into the quantile sharing mapping layer to generate the basic quantile representation of the unit effective manufacturing cost.

[0144] Based on the fracture initiation stage, fracture propagation stage, fracture closure stage and quality inspection mapping stage of the fracture propagation sequence, dynamic quantile constraints are applied to the basic quantile representation, and risk amplification is performed on the cost representation corresponding to the 90th and 95th quantiles through the tail quantile amplification layer.

[0145] By using a monotonic quantile correction layer, the cost representations corresponding to the 10th, 50th, 80th, 90th, and 95th quantiles are sequentially corrected so that the unit effective manufacturing cost corresponding to the subsequent quantile is not lower than the unit effective manufacturing cost corresponding to the preceding quantile, thus generating a multi-risk quantile prediction interval for the unit effective manufacturing cost of the target fiber production batch.

[0146] The entropy increase input gate includes: a hierarchical sequence reader, an entropy increase field arranger, a gradient direction encoder, a loss rate classifier, and an entropy increase gating register. The hierarchical sequence reader receives the entropy increase gradient sequence from the layered saleable quality entropy increase field, ordered according to the spinneret layer, cooling and curing layer, oiling and bundling layer, drawing and shaping layer, winding and spinning layer, and quality inspection and grading layer. The entropy increase field arranger sequentially reads the cost entropy enhancement degree, saleable quality loss rate, entropy increase gradient direction, material flow direction weight, tension continuity mapping result, and energy coupling strength corresponding to each layer, and generates a six-dimensional hierarchical entropy increase vector according to the order of cost entropy enhancement degree, saleable quality loss rate, entropy increase gradient direction, material flow direction weight, tension level, and energy level. The gradient direction encoder divides the entropy increase gradient direction... The upstream diffusion, same-layer diffusion, and downstream diffusion are written to 0, 1, and 2 respectively; the tension continuity mapping results C0, C1, and C2 are written to 0, 1, and 2 respectively; and the energy coupling strengths E0, E1, and E2 are written to 0, 1, and 2 respectively. The loss rate classifier reads the marketable mass loss rate. When the loss rate does not exceed 3 kg / h, it is written to L0; when the loss rate is greater than 3 kg / h but not more than 10 kg / h, it is written to L1; and when the loss rate is greater than 10 kg / h, it is written to L2. The entropy increase gating register splices the six-dimensional hierarchical entropy increase vectors of the six process layers into a thirty-six-dimensional entropy increase input record according to the process order. The hierarchical positions with tension level 2, energy level 2, or loss rate level L2 are written to the entropy increase enhancement bit, and the entropy increase gating vector is output.

[0147] The fracture stage gate includes: a lock-in ring reader, a fracture node counter, a stage position encoder, a propagation length register, a coverage register, and a grade offset register. The lock-in ring reader reads the fracture critical points, associated potential field connection edges, out-of-limit indicators, phase boundary entropy increase tags, spinneret group numbers, spinning position numbers, and winding drum numbers in the mass fracture lock-in ring. The fracture node counter counts the number of tension fracture critical points, mass conservation fracture critical points, and grade fracture critical points, respectively, and writes them into the three-dimensional fracture counting vector in the order of tension fracture count, conservation fracture count, and grade fracture count. The stage position encoder scans the fracture propagation sequence in the order of fracture initiation point, fracture transmission node, fracture closure node, and quality inspection mapping node, marking nodes located at spinneret group or monofilament bundle positions as P0, nodes located at drafting section positions as P1, and so on. The node located at the winding drum position is marked as P2, and the node located at the quality inspection level position is marked as P3; the propagation length register calculates the number of potential field connection edges between the fracture initiation point and the quality inspection mapping node. When the number of edges does not exceed 2, it is written to R0; when the number of edges is 3 to 5, it is written to R1; and when the number of edges exceeds 5, it is written to R2; the coverage range register counts the number of spinneret groups, spinning positions, and winding drums covered by the same locking ring, and writes them into the three-dimensional coverage vector in the order of "number of spinneret groups covered - number of spinning positions covered - number of winding drums covered"; the level offset register reads the level offset gradients Q0, Q1, and Q2. When the level offset gradient is Q2, the quality inspection offset status is written to 1, and otherwise written to 0; the fracture stage gate concatenates the three-dimensional fracture count vector, stage position code, propagation length marker, three-dimensional coverage vector, and quality inspection offset status into a fracture stage vector;

[0148] The quantile-sharing mapping layer includes: an input concatenation buffer, a shared weight matrix memory, a first mapping unit, a second mapping unit, and a basic quantile output register. The input concatenation buffer receives the 36-dimensional entropy-increasing gate vector output from the entropy-increasing input gate and the fracture stage vector output from the fracture stage gate, and concatenates them into a joint prediction vector in the order of entropy-increasing gate vector first and fracture stage vector last. The shared weight matrix memory stores the first layer weights, first layer biases, second layer weights, and second layer biases obtained after training in the same batch. The first mapping unit multiplies each field in the joint prediction vector sequentially with the corresponding position of the first layer weight, adds all the products, adds the first layer bias, and obtains the first hidden record, writing 0 for any result less than 0. The second mapping unit reads the first hidden record, multiplies it in the same way with the corresponding position of the second layer weight, and accumulates the results to obtain the second hidden record. The basic quantile output register maps the second hidden record to the basic cost representations corresponding to the 10th, 50th, 80th, 90th, and 95th quantiles, and stores them in the basic quantile table in ascending order of quantile.

[0149] The tail-level amplification layer includes: a high-risk marker reader, a tail-level weight table, a risk increment calculator, and a tail-level overlay. The high-risk marker reader reads the entropy increase enhancement level, the number of tension fracture critical points, the number of mass conservation fracture critical points, the number of grade fracture critical points, the propagation length marker, and the quality inspection offset status. The tail-level weight table pre-stores the amplification factor corresponding to different risk statuses. When C2, E2, and Q2 are not present and the propagation length is R0, the 90th and 95th percentiles retain their basic cost representation unchanged. When C2 or E2 is present and the propagation length is R1, the cost representation of the 90th percentile is increased by 5%, and the cost representation of the 95th percentile is... Increase by 8%; when any two of C2, E2, and Q2 exist and the propagation length is R2, increase the cost representation of the 90th quantile by 10% and the cost representation of the 95th quantile by 15%; when the three fracture sets of tension fracture, conservation fracture, and graded fracture are all non-empty, increase the cost representation of the 90th quantile by 12% and the cost representation of the 95th quantile by 18%; after reading the corresponding amplification factor, the risk increment calculator performs incremental processing on the basic cost representations of the 90th and 95th quantiles; the tail quantile overlay writes the incrementally processed 90th and 95th quantiles into the basic quantile table to form the tail amplified quantile table;

[0150] The monotonic quantile correction layer includes: a quantile sequence reader, an adjacent quantile comparator, an incremental difference corrector, and a quantile interval outputter. The quantile sequence reader reads the enlarged quantile table at the tail in the order of the 10th quantile, 50th quantile, 80th quantile, 90th quantile, and 95th quantile. The adjacent quantile comparator compares the effective manufacturing cost per unit of two adjacent quantiles sequentially. If the latter quantile value is not lower than the former quantile value, the two quantile values ​​remain unchanged. If the latter quantile value is lower than the former quantile value, the incremental difference corrector adjusts the latter quantile value to the former quantile value plus 0.02 yuan / kg based on the former quantile value. The quantile interval outputter expands downward by 0.03 yuan / kg and upward by 0.03 yuan / kg respectively based on the corrected 10th, 50th, 80th, 90th, and 95th quantiles to generate the corresponding quantile prediction intervals.

[0151] Read the predicted unit effective manufacturing cost values ​​of the 10th, 50th, 80th, 90th, and 95th percentiles output by the monotonic quantile correction layer and arrange them in order of quantile from low to high; extend the predicted value of the 50th percentile downward by 0.03 yuan / kg and upward by 0.03 yuan / kg to form the median quantile prediction interval;

[0152] The prediction intervals corresponding to the 80th and 90th percentile predicted values ​​are combined to form a high-risk quantile prediction interval, where the 80th percentile is used to represent the risk of higher costs and the 90th percentile is used to represent the risk of cost collapse warning. Specifically, the lower boundary of the 80th percentile prediction interval is read as the lower boundary of the high-risk quantile prediction interval, and the upper boundary of the 90th percentile prediction interval is read as the upper boundary of the high-risk quantile prediction interval. The interval is then marked as the high-risk quantile prediction interval.

[0153] The 95th percentile prediction value is extended downward by 0.03 yuan / kg and upward by 0.03 yuan / kg to form an extreme risk quantile prediction interval, which is used to represent the upper limit of the extreme unit effective manufacturing cost when the fracture lock-in loop continues to spread, the effective marketable quality rapidly declines, or the grade downgrade occurs in a concentrated manner; when the upper boundary of the extreme risk quantile prediction interval exceeds the preset extreme cost threshold, the target fiber production batch is marked as an extreme cost collapse risk state.

[0154] In this embodiment, the output cost prediction results, cost collapse level, and corresponding quality fracture lock-in loop include:

[0155] Read the multi-risk quantile prediction interval of the unit effective manufacturing cost of the target fiber production batch, and extract the median quantile prediction interval, high-risk quantile prediction interval, extreme-risk quantile prediction interval and the corresponding upper boundary of the quantile interval;

[0156] Read the preset unit manufacturing cost threshold, preset high-risk cost threshold, and preset extreme cost threshold. Compare the median quantile prediction interval, high-risk quantile prediction interval, and extreme-risk quantile prediction interval with the corresponding thresholds to generate quantile exceedance markers. Specifically, generating quantile exceedance markers involves:

[0157] The upper boundary of the median quantile prediction interval is compared with the preset unit manufacturing cost threshold. If the upper boundary is not higher than the threshold, write the marker M0. If it exceeds the threshold and the excess does not exceed 0.20 yuan / kg, write the marker M1. If the excess is greater than 0.20 yuan / kg, write the marker M2.

[0158] The upper boundary of the high-risk quantile prediction interval is read and compared with the preset high-risk cost threshold. If the upper boundary is not higher than the threshold, write the mark H0. If it exceeds the threshold and the excess does not exceed 0.35 yuan / kg, write the mark H1. If the excess is greater than 0.35 yuan / kg, write the mark H2.

[0159] The upper boundary of the extreme risk quantile prediction interval is compared with the preset extreme cost threshold. If the upper boundary is not higher than the threshold, write the mark E0. If it exceeds the threshold and the excess does not exceed 0.50 yuan / kg, write the mark E1. If the excess is greater than 0.50 yuan / kg, write the mark E2.

[0160] The three types of markers, Mx, Hx, and Ex, are concatenated into a quantile exceedance marker string in the order of median marker, high-risk marker, and extreme marker, and written into the quantile exceedance field of the current batch, where x represents the corresponding marker level.

[0161] The system reads the quantile exceedance marker, the number of quality breakage lock rings, the number of spinning positions covered by lock rings, the number of winding drums covered by lock rings, and the corresponding cost entropy enhancement. When the high-risk quantile prediction interval or the extreme-risk quantile prediction interval exceeds the corresponding threshold, the target fiber production batch is marked as a batch with unit effective manufacturing cost collapse risk. A cost collapse level is generated according to the exceedance quantile level. Specifically, the cost collapse level generation according to the exceedance quantile level is as follows:

[0162] Read the high-risk and extreme markers in the quantile over-limit marker string. When the high-risk marker is H0 and the extreme marker is E0, record the cost collapse level as G0.

[0163] When high risk is marked as H1 and extreme risk as E0, and the number of quality fracture lock-in rings does not exceed 1, the cost collapse level is recorded as G1.

[0164] When high risk is marked as H2 and extreme risk as E0, or high risk is marked as H1 and extreme risk as E1, the cost collapse level is recorded as G2;

[0165] When the extreme mark is E1, and the number of lock rings covering the spinning position is not less than 2 or the number of lock rings covering the winding drum is not less than 4, the cost collapse level is recorded as G3.

[0166] When the extreme value is marked as E2, or when the number of mass fracture lock rings is not less than 3 and the corresponding cost entropy enhancement is all at the high strength mark, the cost collapse level is marked as G4.

[0167] The cost collapse levels G0, G1, G2, G3 and G4 are written into the no collapse, mild collapse, moderate collapse, severe collapse and extremely severe collapse status fields respectively, and the corresponding quantile over-limit marker, number of locked rings, number of covered spinning positions, number of covered winding cylinders and cost entropy enhancement are written into the level judgment record.

[0168] The cost collapse level is correlated with the quality breakage lockout ring that triggers the over-limit, and the cost prediction results, unit effective manufacturing cost risk range, cost collapse level, trigger spinneret group, trigger spinning position, trigger winding drum and corresponding quality breakage lockout ring for the target fiber production batch are output.

[0169] refer to Figure 2 A fiber production cost prediction system, comprising:

[0170] The data mapping module is used to collect production and manufacturing data and establish mapping relationships for the production process;

[0171] The entropy increase unit module is used to divide the process phase boundary segments according to the production process mapping relationship and generate marketable quality entropy increase units;

[0172] The entropy increase field module is used to perform tension continuity mapping and energy consumption conversion on the saleable mass entropy increase unit to generate a layered saleable mass entropy increase field.

[0173] The lock-in loop module is used to construct the mass transfer potential field and generate the mass break lock-in loop;

[0174] The quantile prediction module is used to extract the entropy increase gradient sequence and the fracture propagation sequence and generate multi-risk quantile prediction intervals.

[0175] The collapse determination module is used to determine the collapse risk status of the unit effective manufacturing cost based on the multi-risk quantile prediction interval and output the corresponding results.

[0176] Example 1: In a continuous fiber production cycle, the system receives a batch of production data from a spinning production line. The batch plans to input 12,000 kg of polyester melt raw material, 178 kg of masterbatch, and 92 kg of oiling agent, involving 12 spinning stations, 48 ​​spinneret groups, and 96 winding drums. The standard unit manufacturing cost of the production line is 7.85 yuan / kg, the historical average waste rate is 1.9%, and the first-grade product rate is 94.5%. In the first 60 minutes after the start of this batch, the traditional cost prediction system estimates the unit manufacturing cost based on the raw material unit price, planned power consumption, labor hours, and equipment depreciation, obtaining a unit manufacturing cost of 7.91 yuan / kg, with no obvious abnormalities detected.

[0177] The system first maps batch production data, establishing a correspondence between spinneret groups and winding drums based on batch number, spinning position number, spinneret group number, and timestamp. Spinning position 3 corresponds to spinneret groups 9 to 12, spinning position 4 corresponds to spinneret groups 13 to 16, and spinneret group 11 corresponds to winding drums 25 to 28. The system reads melt pressure, filter differential pressure, monofilament tension, draw ratio, oiling rate, winding speed, number of yarn breaks, number of splices, power consumption, and steam consumption every 30 seconds, generating 960 sets of sampling records. After mapping, the average melt pressure of spinneret group 11 is 16.9 MPa, higher than the average of 15.7 MPa for other groups in the batch. The filter differential pressure increases from 0.41 MPa to 0.68 MPa, but no shutdown record has been generated yet; therefore, the traditional method treats this as a normal fluctuation.

[0178] During the process phase delineation stage, the system divides production data into six process phase segments: spinneret output, cooling and curing, bundling and oiling, drafting and setting, winding into a bobbin, and quality inspection and grading. Taking the segment corresponding to the 11th spinneret group at the 3rd spinning position as an example, the incoming material mass at the spinneret output phase is 1002.4 kg, the effective output mass is 987.6 kg, the waste filament mass is 7.8 kg, the rework mass is 2.6 kg, the grade degradation mass is 4.4 kg, the energy consumption is 386 kWh, and the equipment occupancy time is 1.8 hours. The system writes the above data into the marketable quality entropy increase unit and marks the waste filament mass, rework mass, and grade degradation mass as unsellable loss, delayed recovery loss, and reduced marketable loss, respectively. At this time, the marketable quality loss rate of the unit is 8.2 kg / h, while the average value of the stable spinning position in the same batch is 3.1 kg / h.

[0179] In the drafting and setting phase, the draft ratio of the third spinning position was set to 3.45, but the actual recorded values ​​fluctuated between 3.37 and 3.53. The average tension of the monofilament was 22.8 cN, with a standard deviation of 0.46 cN, while the standard deviation of the tension in the stable spinning position was 0.19 cN. The system mapped the pressure anomaly at the spinneret exit phase to the tension fluctuation at the drafting and setting phase, finding that the monofilament bundle corresponding to the 11th spinneret group exhibited tension transmission instability in all four consecutive sampling windows. Subsequently, the system performed energy consumption calculations, allocating the additional 42 kWh of power consumption and 0.35 hours of equipment occupancy at the third spinning position to the 25th to 28th winding drums. The corresponding effective mass energy consumption increased from 0.078 kWh / kg to 0.104 kWh / kg, indicating a decrease in the effective mass of the first-grade product produced under the same energy consumption.

[0180] When generating the layered marketable quality entropy increase field, the system sequentially connects the spinneret forming layer, cooling and curing layer, oiling and bundling layer, drawing and shaping layer, winding and bobbin layer, and quality inspection and grading layer. The cost entropy enhancement of the path containing the 11th spinneret group increases from 0.28 in the spinneret forming layer to 0.51 in the drawing and shaping layer, and then to 0.67 in the winding and bobbin layer; the cost entropy enhancements of adjacent stable orifice groups are 0.16, 0.21, and 0.25, respectively. The results show that the loss does not occur in isolation at the winding end, but rather amplifies layer by layer along the continuous forming link starting from the spinneret group.

[0181] In the mass transfer potential field construction stage, the system uses spinneret groups, monofilament bundles, drafting sections, winding drums, and quality inspection grades as nodes to form a mass transfer potential field from the 11th spinneret group to the 25th to 28th winding drums. The system calculates the tension potential difference for the continuous streamline; the tension potential difference grade from the 11th spinneret group to the drafting section is 0.72, exceeding the preset critical value of 0.60. It calculates the mass expansion / contraction rate for the conserved loop; the closure deviation between the incoming material mass and the effective output, waste yarn, rework, and grade downgraded mass reaches 4.9%, exceeding the preset critical value of 3.0%. It calculates the grade offset gradient for the grade mapping surface; the fineness deviation increases from 0.8 dtex to 1.9 dtex, and the number of filaments increases from an average of 8 per drum to 27. Based on this, the system marks 6 fracture critical points and aggregates the 11th spinneret group, the 3rd spinning drafting section, the 25th to 28th winding drums, and the corresponding quality inspection nodes into a single mass fracture lock-in loop.

[0182] During the model training phase, the system used 680 historical batches as training samples, including 412 normal batches, 168 slightly abnormal batches, and 100 cost collapse batches. Each training sample record includes entropy enhancement degree, loss rate, tension potential difference level, mass expansion / contraction rate level, level offset gradient, number of locked loops, and actual unit effective manufacturing cost. For example, training sample A has a waste rate of 1.6%, 0 locked loops, and an actual unit effective manufacturing cost of 7.88 yuan / kg; training sample B has a waste rate of 3.4%, 2 locked loops, and an actual unit effective manufacturing cost of 8.61 yuan / kg; training sample C has a waste rate of 5.1%, a rework rate of 2.7%, 5 locked loops, and an actual unit effective manufacturing cost of 9.73 yuan / kg. After 120 rounds of training, the average error at the 50th quantile of the validation set was 0.29 yuan / kg, and the average error at the 90th quantile was 0.24 yuan / kg.

[0183] In the prediction phase of this batch, the system inputs the entropy increase gradient sequence and fracture propagation sequence into the improved quantile regression neural network, outputting the effective manufacturing cost per unit as follows: 10th quantile: 7.96 yuan / kg; 50th quantile: 8.42 yuan / kg; 80th quantile: 8.91 yuan / kg; 90th quantile: 9.46 yuan / kg; and 95th quantile: 9.82 yuan / kg. Since the 90th quantile exceeds the preset high-risk threshold of 9.20 yuan / kg, the system determines that the batch has entered a state of effective manufacturing cost collapse risk before winding is completed, and outputs a collapse level of two, with the triggering object being the quality fracture lock ring corresponding to the 11th spinneret group.

[0184] The final quality inspection results for the batch showed that the weight of first-grade products was 10,568 kg, second-grade products were 806 kg, substandard products were 248 kg, rework products were 214 kg, and waste yarn was 126 kg. The converted effective marketable weight was 10,831 kg, and the actual effective unit manufacturing cost was 9.39 yuan / kg. The traditional method predicted a value of 7.91 yuan / kg with a relative error of 15.76%, and did not output the source of the anomaly. The 90th percentile prediction value of this invention was 9.46 yuan / kg with a relative error of 0.75%, and the anomaly was located at the 11th spinneret group, the 3rd spinning position, and the 25th to 28th winding drums.

[0185] In the same test set, the traditional method had an average error of 0.72 yuan / kg in the effective unit manufacturing cost of 70 batches, a batch identification rate of 62.9% for cost collapse, and an accuracy rate of 38.6% in locating the source of anomalies. The method of the present invention had an average error of 0.31 yuan / kg at the 50th percentile and an average error of 0.25 yuan / kg at the 90th percentile, a batch identification rate of 88.6% for cost collapse, and an accuracy rate of locating the source of anomalies of 84.3%.

[0186] 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 predicting fiber production and manufacturing costs, characterized in that, include: Collect production data of the target fiber production batch and establish a production process mapping relationship based on batch number, spinning station number, spinneret group number and timestamp. Based on the production process mapping relationship, the production data is divided into multiple process phase boundary segments. For each process phase boundary segment, the quality loss and resource consumption parameters are statistically analyzed to generate a marketable quality entropy increase unit. Based on the energy coupling strength and material flow direction weight of adjacent processes, tension continuity mapping and energy consumption conversion are performed on all saleable quality entropy increase units to generate a hierarchical saleable quality entropy increase field. In each unit, the cost entropy increase intensity, saleable quality loss rate and entropy increase gradient direction are recorded simultaneously. Construct a mass transfer potential field that runs through the spinneret group to the quality inspection node, calculate the tension potential difference, mass expansion and contraction rate and grade offset gradient of the continuous streamline, the conserved loop and the grade mapping surface, and mark the fracture critical point when any index exceeds the corresponding critical threshold and aggregate to form a mass fracture lock loop. Extract the entropy increase gradient sequence from the layered saleable quality entropy increase field and the fracture propagation sequence from the quality fracture closed loop, input them into the improved quantile regression neural network, and perform dynamic quantile constraints and tail quantile amplification on different fracture propagation stages to generate a multi-risk quantile prediction interval for the unit effective manufacturing cost of the target batch. Compare the prediction intervals of each risk quantile with the preset unit manufacturing cost threshold. When the prediction interval of a high-risk quantile exceeds the corresponding threshold, the target fiber production batch is determined to enter the unit effective manufacturing cost collapse risk state, and the cost prediction result, cost collapse level and corresponding quality breakage lockout are output.

2. The fiber production cost prediction method according to claim 1, characterized in that, The production data includes raw material input data, melt conveying data, spinneret forming data, cooling and solidification data, oiling treatment data, drawing and shaping data, winding into cylinder data, equipment operation data, quality inspection data, energy consumption statistics, waste filament recycling data, rework processing data, and manufacturing cost data.

3. The fiber production cost prediction method according to claim 1, characterized in that, The establishment of the production process mapping relationship includes: Read the batch number, production line number, spinning station number, spinneret group number, metering pump number, winding drum number, and quality inspection number corresponding to the production batch of the target fiber; Establish the process sequence relationship between spinneret group, spinning station, winding drum and quality inspection number according to the production time sequence; Establish the quality transfer relationship between raw material input node, spinning node, drawing and shaping node, winding node and quality inspection and grading node according to the material flow direction; Establish the equipment occupancy relationship among spinning equipment, drawing equipment, winding equipment, and quality inspection equipment based on equipment operating time; By associating and mapping process succession relationships, quality transfer relationships, and equipment occupancy relationships, a production process mapping relationship corresponding to the target fiber production batch is generated.

4. The fiber production cost prediction method according to claim 1, characterized in that, The unit for generating marketable quality entropy increase includes: Based on the production process mapping relationship, read the mass loss and resource occupation parameters corresponding to the target process phase boundary segment, and write the mass loss and resource occupation parameters into the phase boundary data record in the order of incoming materials, effective output, waste filament generation, rework recycling, grade downgrade, energy consumption and equipment occupation. Based on the quality inspection level, rework status and waste fiber status corresponding to the phase boundary segment of the target process, the saleable value of the output fibers in the phase boundary data record is converted. The first-grade output portion is determined as the complete saleable quality, the second-grade output portion, the substandard output portion and the reworked and recycled portion are converted into reduced saleable quality respectively, and the waste fiber portion is marked as unsaleable quality. Align the incoming material in the target process phase segment with the complete marketable quality, reduced marketable quality, and unmarketable quality to identify the loss location where the input material has not been converted into marketable quality. Generate marketable quality loss marks based on the spinneret group, spinning position, drafting section, and winding drum corresponding to the loss location. Read the tension fluctuation, stretching offset, oiling offset, winding offset and quality inspection level offset in the target process phase boundary segment, bind the marketable quality loss mark with the corresponding process offset, and generate a phase boundary entropy increase tag that records the source of loss, loss type, loss location and transmission direction; By combining phase boundary data records, sales value conversion results, sales quality loss markers, and phase boundary entropy increase labels, a sales quality entropy increase unit corresponding to the target process phase boundary segment is generated.

5. The fiber production cost prediction method according to claim 1, characterized in that, The generation of a hierarchical marketable quality entropy-increasing field synchronously records the cost entropy enhancement degree, marketable quality loss rate, and entropy increase gradient direction in each unit, including: Read the phase boundary number, process sequence, spinning position number, spinneret group number, winding drum number, effective output quality, energy consumption, equipment usage time, monofilament tension change and phase boundary entropy increase label from each available quality entropy increase unit, and establish the succession relationship between adjacent units according to the sequence of spinneret output, cooling and solidification, bundling and oiling, drawing and shaping, winding into a drum and quality inspection and grading. Based on the connection relationship between adjacent units, the effective output quality, waste wire quality, rework quality and grade downgrade quality in the upstream available quality entropy increase unit are written into the material acceptance field of the downstream available quality entropy increase unit, and the material flow direction weight between adjacent processes is generated based on the material acceptance result. Read the records of monofilament tension change, drafting offset, winding speed change and broken joint in adjacent saleable mass entropy increase units, align the tension change of the upstream unit with the tension change of the downstream unit, generate tension continuity mapping results, and write the tension instability position into the tension transmission field of the corresponding unit. Read the energy consumption, equipment usage time, equipment continuous operation status and process continuity time in adjacent available quality entropy increase units, convert the energy consumption corresponding to the same spinning position, the same spinneret group and the same winding drum to the corresponding available quality entropy increase unit, and generate the energy coupling strength between adjacent processes. Based on the material flow direction weight, tension continuity mapping results, and energy coupling strength, each saleable quality entropy increase unit is divided into a spinneret forming layer, a cooling and curing layer, an oiling and bundling layer, a drawing and shaping layer, a winding and spinning layer, and a quality inspection and grading layer. In each layer, the cost entropy increase degree, saleable quality loss rate, and entropy increase gradient direction are recorded to generate a layered saleable quality entropy increase field.

6. The fiber production cost prediction method according to claim 1, characterized in that, The marking of the fracture critical point and the aggregation to form a mass fracture lock-in loop includes: Read the available quality entropy increase units corresponding to each layer in the layered available quality entropy increase field, and extract the spinneret group number, spinning position number, single filament bundle identifier, drawing section identifier, winding drum number, quality inspection grade, material flow direction weight, tension continuity mapping result, energy coupling strength, phase boundary entropy increase label, cost entropy increase degree, and entropy increase gradient direction. Using the spinneret group, monofilament bundle, drawing section, winding drum, and quality inspection grade as nodes of the mass transfer potential field, a potential field connection edge is established based on the filament output connection relationship from the spinneret group to the monofilament bundle, the tension transfer relationship from the monofilament bundle to the drawing section, the forming connection relationship from the drawing section to the winding drum, and the grade mapping relationship from the winding drum to the quality inspection grade, thereby generating a mass transfer potential field that runs through the spinneret group to the quality inspection node. In the mass transfer potential field, the potential field connecting edges that are continuously connected along the spinneret group, monofilament bundle, drawing section and winding drum are combined into a continuous streamline. The closed correspondence between the mass of incoming material, effective output mass, waste yarn mass, rework mass and grade downgrade mass is combined into a conserved loop. The correspondence between fineness, strength, elongation, yarn unevenness and oil content and quality inspection grade is combined into a grade mapping surface. For continuous streamlines, the tension difference is obtained by comparing the changes in monofilament tension and the tension continuity mapping results corresponding to adjacent potential field nodes. For conserved loops, the quality closure deviation between the quality of incoming material and effective output quality, waste filament quality, rework quality and grade downgrade quality is obtained. For grade mapping surfaces, the deviation direction and deviation magnitude between each quality inspection index and the target grade boundary are compared to obtain the grade offset gradient. The tension potential difference, mass expansion and contraction rate, and grade offset gradient are compared with the corresponding preset critical thresholds. The fracture critical point is marked at the position that exceeds the corresponding preset critical threshold. According to the correlation relationship of the same spinneret group, the same spinning position, the same winding drum, and the continuous process phase boundary, the fracture critical point, the associated potential field connection edge, the over-limit index, the phase boundary entropy increase label, and the corresponding marketable mass entropy increase unit are aggregated to form a mass fracture lock-in loop.

7. The fiber production cost prediction method according to claim 1, characterized in that, The multi-risk quantile prediction interval for generating the effective manufacturing cost per unit of the target batch includes: Read the cost entropy enhancement degree, the rate of loss of marketable quality, the direction of entropy increase gradient, the weight of material flow direction, the result of tension continuity mapping and the energy coupling strength in the layered marketable quality entropy increase field, and generate an entropy increase gradient sequence in the order of spinneret forming layer, cooling and curing layer, oiling and bundling layer, stretching and shaping layer, winding and spinning layer and quality inspection and grading layer. Read the fracture critical point, associated potential field connection edge, over-limit index, phase boundary entropy increase label, corresponding marketable quality entropy increase unit, spinneret group number, spinning position number and winding drum number in the quality fracture lock loop, and generate a fracture propagation sequence in the order of fracture initiation point, fracture transmission node, fracture closure node and quality inspection mapping node. The entropy-increasing gradient sequence and the fracture propagation sequence are input into the improved quantile regression neural network, which includes an entropy-increasing input gate, a fracture stage gate, a quantile sharing mapping layer, a tail quantile amplification layer, and a monotonic quantile correction layer. The entropy increase input gate reads the cost entropy increase degree, the rate of loss of marketable quality and the direction of entropy increase gradient. The fracture stage gate reads the number of fracture critical points, fracture propagation length, the coverage of the lock-in loop and the quality inspection level offset status. The reading results are input into the quantile sharing mapping layer to generate the basic quantile representation of the unit effective manufacturing cost. Based on the fracture initiation stage, fracture propagation stage, fracture closure stage and quality inspection mapping stage of the fracture propagation sequence, dynamic quantile constraints are applied to the basic quantile representation, and risk amplification is performed on the cost representation corresponding to the 90th and 95th quantiles through the tail quantile amplification layer. By sequentially correcting the cost representations corresponding to the 10th, 50th, 80th, 90th, and 95th quantiles through a monotonic quantile correction layer, the unit effective manufacturing cost corresponding to the subsequent quantile is not lower than the unit effective manufacturing cost corresponding to the preceding quantile, thereby generating a multi-risk quantile prediction interval for the unit effective manufacturing cost of the target fiber production batch.

8. The fiber production cost prediction method according to claim 1, characterized in that, The output cost prediction results, cost collapse level, and corresponding quality fracture lock-in loop include: Read the multi-risk quantile prediction interval of the unit effective manufacturing cost of the target fiber production batch, and extract the median quantile prediction interval, high-risk quantile prediction interval, extreme-risk quantile prediction interval and the corresponding upper boundary of the quantile interval; Read the preset unit manufacturing cost threshold, preset high-risk cost threshold and preset extreme cost threshold, compare the median quantile prediction interval, high-risk quantile prediction interval and extreme-risk quantile prediction interval with the corresponding threshold respectively, and generate quantile exceedance markers; Read the quantile over-limit marker, the number of quality breakage lock rings, the number of spinning positions covered by lock rings, the number of winding drums covered by lock rings, and the corresponding cost entropy enhancement. When the high-risk quantile prediction interval or the extreme-risk quantile prediction interval exceeds the corresponding threshold, mark the target fiber production batch as a batch with unit effective manufacturing cost collapse risk, and generate cost collapse level according to the over-limit quantile level. The cost collapse level is correlated with the quality breakage lockout ring that triggers the over-limit, and the cost prediction results, unit effective manufacturing cost risk range, cost collapse level, trigger spinneret group, trigger spinning position, trigger winding drum and corresponding quality breakage lockout ring for the target fiber production batch are output.

9. A fiber production cost prediction system, implementing the fiber production cost prediction method according to any one of claims 1 to 8, characterized in that, include: The data mapping module is used to collect production and manufacturing data and establish mapping relationships for the production process; The entropy increase unit module is used to divide the process phase boundary segments according to the production process mapping relationship and generate marketable quality entropy increase units; The entropy increase field module is used to perform tension continuity mapping and energy consumption conversion on the saleable mass entropy increase unit to generate a layered saleable mass entropy increase field. The lock-in loop module is used to construct the mass transfer potential field and generate the mass break lock-in loop; The quantile prediction module is used to extract the entropy increase gradient sequence and the fracture propagation sequence and generate multi-risk quantile prediction intervals. The collapse determination module is used to determine the collapse risk status of the unit effective manufacturing cost based on the multi-risk quantile prediction interval and output the corresponding results.

Images