A fly ash-based cementitious material performance prediction and formulation optimization system

Abstract

This invention discloses a fly ash-based cementitious material performance prediction and formulation optimization system, specifically relating to the field of cementitious material performance prediction and industrial data processing. The system includes: a data acquisition module for collecting fly ash entry times, conveying start and stop times, weighing start and stop times, stirring start and stop times, and on-site testing records, outputting a unified time-stamped record stream; an edge computing module for receiving the unified time-stamped record stream, writing it to a local cache in arrival order, and outputting a continuous record queue sorted by time stamp; and a batch determination module for reading the continuous record queue. By performing time stamp conversion, batch reconstruction, and acceptance determination on multi-source records at the edge, performance prediction is first established on the basis of verified input sources, time series, and value ranges, thereby relatively suppressing erroneous predictions from entering the field application stage under conditions of sample mismatch, data loss, or abnormal operating conditions.

Description

Technical Field

[0001] This invention relates to the field of cementitious material performance prediction and industrial data processing technology, and more specifically, to a fly ash-based cementitious material performance prediction and formulation optimization system. Background Technology

[0002] In the production and application of fly ash-based cementitious materials, existing performance prediction methods mainly focus on using raw material test results, historical test data and production process parameters to predict indicators such as strength, fluidity and setting time after molding. The common practice is to summarize the basic indicators of fly ash, metering information, stirring records and existing performance data, and then output the corresponding prediction results from the host computer or cloud model, and use this to assist in on-site judgment. However, in the scenario of frequent switching of fly ash sources, shared storage, and continuous feeding without stopping the machine in the use of municipal solid waste incineration power plants, the site is often subject to the constraints of inconsistent data collection time points, unclear batch boundaries, missing local records, and short-term fluctuations in operating conditions. Moreover, the system is required to provide usability judgments directly within the current production cycle by relying on edge computing nodes. Under these conditions, although the existing method can continuously output predicted values, the following situation will repeatedly occur: the usability of the prediction results for the same type of input is inconsistent in different shifts, and the system still gives results as normal when some data is missing or the operating conditions are abnormal. Only after reviewing the data can it be found that the raw material state or process conditions corresponding to the prediction have exceeded the coverage of the existing samples, resulting in erroneous results entering the field use process in the form of normal results. The fundamental reason is that existing technologies generally focus on improving the predictive ability itself, but lack a processing mechanism to verify on-site whether the current input conditions have a basis for acceptance before predicting the output. The technical problem this application aims to solve is: how to make a pre-prediction confidence judgment on the current input data and on-site working conditions based on edge computing during the performance prediction of fly ash-based cementitious materials, so as to avoid directly using prediction results that do not have a reliable basis when there is sample mismatch, data loss or abnormal working conditions. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a fly ash-based cementitious material performance prediction and formulation optimization system. This system performs time-series reorganization, batch reconstruction, historical alignment, and pre-prediction acceptance determination on multi-source records from the production site. Performance prediction is performed when acceptance is successful, and output is blocked when acceptance is unsuccessful, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a system for predicting the performance and optimizing the formulation of fly ash-based cementitious materials, comprising: The data acquisition module is used to collect the time of fly ash entering the silo, the time of start and stop of conveying, the time of start and stop of weighing, the time of start and stop of mixing, and on-site test records, and output a unified time-stamped record stream; The edge computing module is used to receive a unified time-stamped record stream, write it to the local buffer in the order of arrival, and output a continuous record queue sorted by time stamp. The batch determination module is used to read the continuous record queue, take the start time of a weighing as the batch start time and the corresponding discharge completion time as the batch end time, extract the warehousing records, conveying records, weighing records, mixing records and on-site inspection records between the start and end times, and output the current batch record set; The acceptance judgment module is used to read the current batch record set, determine whether the source of fly ash is unique from the start of the batch to the end of weighing, whether the weighing value increases continuously, and whether the start and stop times of stirring are between the end of weighing and the completion of discharge. Then, it screens out the control batch with the same fly ash source and the same process sequence from the historical batches, extracts the value range corresponding to each record item, determines whether all record items of the current batch fall into the corresponding value range, and outputs the acceptance result. The result control module is used to read the current batch record set and the acceptance result. When the acceptance result is passed, it extracts the fly ash source record, weighing record, mixing record and on-site test record from the current batch record set as the performance prediction input and outputs the performance prediction result. When the acceptance result is failed, it outputs the prohibition of acceptance mark and the corresponding out-of-bounds record item.

[0005] In a preferred embodiment, the execution of the acquisition module includes: Arrival trigger acquisition is performed on fly ash silo entry signal, conveying start / stop signal, weighing signal, stirring start / stop signal and on-site detection signal, the occurrence time and signal type of each signal are read, and the corresponding raw event record is generated; The occurrence times in each original event record are converted into a unified time scale under the same timing benchmark, and the signal type, unified time scale and signal value are written into the event queue in a single record manner to generate a unified time scale record stream. Based on the time sequence of two adjacent records in the event queue, the records with the same time stamp are rearranged in the order of fly ash entering the warehouse, conveying start and stop, weighing start and stop, stirring start and stop, and on-site detection, and then the record stream with the same time stamp is output.

[0006] In a preferred embodiment, the acquisition module performs the following actions when reading the signal type: For each signal acquired upon arrival, the interface number, number of fields, field order, status bit, current value, previous value, and time are read. The field difference, time difference, and value difference are calculated. The sum of the absolute values ​​of the field differences, the absolute value of the time difference, and the absolute value of the value difference are summed with the reference records of each candidate signal type. The candidate signal type with the smallest sum is taken as the initial signal type. When the sum of two or more candidate signal types is the same, read the established signal types of the records adjacent to the current signal, substitute each candidate signal type into the current position to form a three-item type sequence, and then compare each three-item type sequence with the process sequence from warehousing to conveying, conveying to weighing, weighing to stirring, and stirring to testing item by item, calculate the number of missing items and the number of reverse order, and take the candidate signal type with the smallest sum of missing items and reverse order as the correction signal type; The signal status is read again according to the corrected signal type. The cumulative value and the difference between adjacent values ​​of the symmetrical signal are calculated. The record where the cumulative value increases for the first time is defined as the start, the record where the difference between adjacent values ​​is continuously non-zero is defined as the process, and the record where the cumulative value remains unchanged and the subsequent records have the same value is defined as the end. The start and stop of the silo signal, conveying signal and stirring signal are determined by the change of the status bit from zero to one and from one to zero, respectively. The entry and result of the detection signal are determined by the presence or absence of the detection item and detection value in the field, and the subdivided signal type is output.

[0007] In a preferred embodiment, the execution of the edge computing module includes: Receive a unified timestamp record stream, write each record to the local buffer in the order of arrival, generate a corresponding arrival sequence number for each record, and output a buffer record set with the arrival sequence number; The cached record set is sorted in ascending order according to a unified time index. If the unified time index is the same, it is sorted in ascending order according to the arrival sequence number. The sorted records are then connected end to end in sequence to output a continuous record queue. When outputting the continuous record queue, the unified time index of the next record is compared with the unified time index of the previous record one by one. Records whose unified time index is less than that of the previous record are removed from their current positions and inserted before the first record whose unified time index is greater than that record. Then, the continuous record queue with continuously increasing time indexes is output.

[0008] In a preferred embodiment, the execution of the batch determination module includes: Read each weighing start record and each discharge completion record from the continuous record queue, pair each weighing start record with the first discharge completion record that follows it, and generate a candidate batch interval; For the cross record between two candidate batch intervals, calculate the absolute value of the time difference between the unified time scale of each cross record and the starting point of the two candidate batch intervals, as well as the absolute value of the time difference between the unified time scale of each cross record and the ending point of the two candidate batch intervals. Sum the two sets of absolute time difference values ​​respectively, and determine the candidate batch interval with the smallest sum of absolute time difference values ​​as the interval to which the cross record belongs, thus generating the batch interval. Based on the start and end points of each batch interval, extract the warehousing records, conveying records, weighing records, mixing records, and on-site testing records located between the corresponding start and end points from the continuous record queue, and output the current batch record set in ascending order according to a unified time scale.

[0009] In a preferred embodiment, the execution of the acceptance determination module includes: Read the fly ash source records, weighing records, mixing records, and on-site testing records from the current batch record set. Count the number of fly ash source types from the start of the batch to the end of the weighing process, and mark a source as established if the number of types is one. Calculate the forward difference between adjacent weighing records, and mark weighing as established if all forward differences are greater than or equal to zero. Then compare the order of the weighing end time, mixing start time, mixing stop time, and discharge completion time, and mark the time sequence as established if the weighing end time is not later than the mixing start time and the mixing stop time is not later than the discharge completion time. Output the current batch record set that simultaneously satisfies the criteria of source establishment, weighing establishment, and time sequence establishment as the batch to be compared.

[0010] In a preferred embodiment, the execution of the acceptance determination module further includes: For the batch to be compared, control batches with consistent fly ash sources and consistent process sequences are screened out from historical batches. For each control batch, the unified timescale is converted into a normalized timescale based on the duration from the start to the end of the batch. Then, the unified timescale of each record item in the batch to be compared is converted into the corresponding normalized timescale. In each control batch, linear interpolation of neighboring points is performed on the same record item according to the normalized timescale to obtain the set of corresponding values ​​of each control batch at each normalized timescale.

[0011] In a preferred embodiment, the execution of the acceptance determination module further includes: For each record item, first calculate the median of the set of corresponding values ​​at the same normalized time scale, then calculate the absolute difference between each corresponding value and the median, and take the median of the absolute differences as the discrete quantity. After deleting the corresponding values ​​whose absolute difference is greater than twice the discrete quantity, repeat the median calculation and deletion calculation until the number of retained values ​​is the same in the two rounds. Then, determine the minimum and maximum values ​​among the retained values ​​as the value range of the record item at the normalized time scale. The current value of each record in the batch to be compared is compared with the value range at the corresponding normalized time scale. When all current values ​​fall within the corresponding value range, the accepted result is output. When there is a current value that falls outside the corresponding value range, the accepted result and the corresponding out-of-range record are output.

[0012] In a preferred embodiment, the execution of the result control module includes: When the acceptance result is passed, the source identifier of the fly ash source record, the mixing time of the mixing record, and the test value of the on-site test record are extracted from the current batch record set, and the current prediction sequence is generated in the order of source identifier, final weighing value, mixing time, and test value.

[0013] In a preferred embodiment, the execution of the result control module further includes: For the current prediction sequence, extract historical prediction sequences with the same source identifier from the historical batches. Calculate the sum of the absolute values ​​of the position differences between the current prediction sequence and each historical prediction sequence. Take the three historical prediction sequences with the smallest sum of absolute values ​​as the nearest neighbor sequences. Then, calculate the weighted average of the performance results corresponding to the three nearest neighbor sequences according to the reciprocal of the sum of their absolute values, and output the performance prediction result. When the acceptance result is not approved, extract the record name, unified time stamp and record value corresponding to the out-of-bounds record item, generate the out-of-bounds record table in ascending order of unified time stamp, and output the prohibition of acceptance flag and the out-of-bounds record table.

[0014] The technical effects and advantages of this invention are as follows: By performing time-scale conversion, batch reconstruction, and acceptance judgment on multi-source records at the edge, performance prediction is first established on the basis of verified input sources, time series, and value ranges, thereby relatively suppressing erroneous predictions from entering the field use stage under conditions of sample mismatch, data loss, or abnormal operating conditions. By aligning the current batch record set with the historical batches using normalized time scales, and generating value ranges for each record item based on the corresponding value set, records under different batch durations and process beats can obtain a basis for comparison at the same position, thereby improving the judgment bias caused by direct misalignment comparison between batches. When the acceptance result passes, the source identifier, final weighing value, stirring time, and detection value are extracted to generate the current prediction sequence. Combined with the performance prediction results of similar historical batches, the system directly outputs a prohibition mark and out-of-bounds record items when the acceptance result fails, thereby matching the on-site result output method with the current batch status. Attached Figure Description

[0015] Figure 1 This is a system module diagram of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Refer to the instruction manual appendix Figure 1 The present invention provides a fly ash-based cementitious material performance prediction and formulation optimization system, comprising: The data acquisition module is used to collect the time of fly ash entering the silo, the time of start and stop of conveying, the time of start and stop of weighing, the time of start and stop of mixing, and on-site test records, and output a unified time-stamped record stream; In this embodiment, the acquisition module is responsible for accessing, identifying, converting, and organizing multi-source records from the production site. Its processing objects include fly ash inlet signals, conveyor start / stop signals, weighing signals, agitator start / stop signals, and on-site detection signals. Since these signals originate from the inlet sensor interface, conveyor control interface, weighing device interface, agitator control interface, and detection device interface or detection input interface, respectively, and the message structure, arrival time, and time field sources of different interfaces are different, the acquisition module first generates a raw event record for each arriving data, then performs time conversion, type determination, state subdivision, and sequence organization on the raw event record, and finally outputs a unified time-stamped record stream for the edge computing module to read. The raw event record includes at least the interface number, occurrence time, signal type, subdivided signal type, and signal value. The event queue is used to save the raw event records in the order of arrival, and the reference record is used to store sample records corresponding to each candidate signal type in historically verified batches. A determined signal type refers to a signal type that has already obtained a unique determination result. This implementation process includes the following: When collecting fly ash inlet signals, conveyor start / stop signals, weighing signals, agitator start / stop signals, and on-site detection signals, the acquisition module immediately latches the occurrence time of the arriving data upon receiving a data update frame, status change frame, or interrupt trigger at any interface, and simultaneously reads the data content output by the current interface. Specifically, the fly ash inlet signal corresponds to the output of the inlet sensor, the conveyor start / stop signal corresponds to the output of the conveyor equipment start / stop port, the weighing signal corresponds to the output of the weighing device, the agitator start / stop signal corresponds to the output of the agitator start / stop port, and the on-site detection signal corresponds to the output of the detection equipment or data written to the detection input terminal. After completing the data reading for each instance, the interface number and occurrence time are extracted first, then the signal type is determined. Subsequently, the interface number, occurrence time, signal type, and signal value are written into a raw event record and appended to the event queue in the order of arrival. When converting the occurrence time in each original event record, the acquisition module selects the local clock on the edge side as the timing reference. For original event records with a built-in time field, the time field is read first, and then converted to the time under the local clock according to the correspondence between the interface and the local clock. For original event records without a time field, the latched occurrence time is used directly. After the conversion is completed, the signal type, converted time and signal value in each original event record are written to the event queue as a single record, forming a record set with a unified time stamp. Each record in the unified time stamp record stream corresponds one-to-one with an original event record. When subsequent modules trace the original source, they only need to read the corresponding original event record according to the record position. When organizing the event queue, the acquisition module first reads all records in ascending order according to a unified time scale, and then checks the record groups with the same unified time scale. For any record group with the same time scale, the records within the group are rearranged in the order of fly ash entering the silo, conveying start and stop, weighing start and stop, stirring start and stop, and on-site detection. Then, all records are connected end to end in ascending order according to a unified time scale and output as a unified time scale record stream. The reason for using this order is that at the same time, the raw material entering and conveying state precedes the weighing process, the weighing process precedes the stirring state, and the detection record is located at the end of the process chain. After writing in this order, the subsequent batch determination module can directly identify the process chain along the record order when intercepting the batch interval, without having to make additional corrections to the sequential relationship within the same time. When reading signal types, the acquisition module extracts the interface number, number of fields, field order, status bit, current value, previous value, and time from the currently arriving data, and calls the reference records corresponding to each candidate signal type from the reference library. The reference records are taken from the original event records of historically verified batches. An independent record set is established for each candidate signal type, and each reference record contains at least the interface number, number of fields, field order, status bit distribution, current value, previous value, and occurrence time. Subsequently, the field difference, time difference, and value difference are calculated for each reference record under each candidate signal type and the currently arriving data. The field difference includes the difference in the number of fields and the difference in field numbers under the same order. Fields with insufficient numbers are included in the calculation as empty spaces. The time difference is the absolute value of the difference between the current time and the time in the reference record. The value difference is calculated by subtracting the previous value from the current value and the corresponding difference in the reference record. After completing the above calculations, the sum of the absolute values ​​of the field differences, the absolute value of the time difference, and the absolute value of the value difference are calculated respectively. These three are then added together to obtain the total for that candidate signal type. Finally, the candidate signal type with the smallest total is selected as the initial signal type. When two or more candidate signal types have the same sum, the acquisition module reads the established signal types from the records before and after the current record, and substitutes each parallel candidate signal type into the current position to form a three-type sequence with the preceding and following established signal types. Then, each three-type sequence is compared item by item with the process sequence from warehousing to conveying, conveying to weighing, weighing to mixing, and mixing to testing, counting the number of missing items and the number of inversions. The number of missing items refers to the number of type pairs that should appear in the process sequence but do not, and the number of inversions refers to the number of type pairs that should be earlier but are later in the sequence. For all parallel... After comparing the candidate signal types, the candidate signal type with the smallest sum of missing items and inversions is selected as the corrected signal type. If the sums are still the same, the number of times the interface number matches the reference record interface number under each candidate signal type is compared. The candidate signal type with more matches is selected as the corrected signal type, thus ensuring that the type determination result at the current position remains unique. The determined signal types include the initial signal type that has not produced a tie for the smallest sum, as well as the signal type that has been corrected. The current record only calls the previous and next determined records and does not call records that have not yet been determined. When continuing to read signal status according to the corrected signal type, the acquisition module calls the corresponding status subdivision rule according to the corrected signal type. For weighing signals, continuous records under the same interface are first extracted in ascending order according to a unified time scale. The cumulative value and adjacent value difference of each record are calculated. The record where the cumulative value first increases is determined as the start, the record where the adjacent value difference is continuously non-zero is determined as the process, and the record where the cumulative value remains unchanged and the subsequent records have the same value is determined as the end. For silo entry signals, conveying signals, and stirring signals, the binary values ​​before and after the status bit flip are read. The record corresponding to the status bit changing from zero to one is determined as the start, and the status bit... The record corresponding to the change from 1 to zero is identified as a stop; for detection signals, the filling status of detection items and detection values ​​in the fields is checked, and records containing only detection items but empty detection values ​​are identified as inputs, while records where both detection items and detection values ​​exist are identified as results; among them, detection items are used to identify the detection content, and detection values ​​are used to record the numerical result corresponding to the detection content. Only when both exist can the corresponding record participate in the performance prediction input extraction in the subsequent result control module; after completing the state subdivision, the subdivision signal type is added to the original event record, and a unified time-stamped record stream with the subdivision signal type is output; Through the above processing, the records output by the acquisition module already include the source of occurrence, arrival time, signal type, and subdivided signal type. When the edge computing module writes to the local cache based on this, it can directly form a continuous record queue according to a unified time scale and subdivided signal type. When the batch determination module extracts the batch interval based on this, it can directly locate the weighing start record, the discharge completion record, and the warehousing record, conveying record, weighing record, mixing record, and on-site detection record within the batch interval. When the acceptance judgment module and the result control module call based on this, they can also directly trace the record source, read the status changes, and extract the corresponding values. In practical applications: When a batch of fly ash enters the storage silo via the inlet sensor, the inlet interface first sends status change data to the acquisition module, then the conveying equipment control interface sends start / stop data, the weighing device continuously outputs weighing records, the stirring equipment control interface outputs start / stop records, and the detection terminal writes the detection items before sample detection and writes the detection values ​​after obtaining the detection values. The acquisition module latches the occurrence time of each of the above data entries, completes signal type determination according to the reference records, and calls the previously determined signal types to complete the correction when parallel candidates appear. Then, the weighing signal is subdivided into start, process, and end, the inlet signal, conveying signal, and stirring signal are subdivided into start and stop, and the detection signal is subdivided into input and result. Finally, a unified time-scaled record stream is output according to a unified time scale and the process sequence at the same time, for subsequent modules to read directly.

[0018] The edge computing module is used to receive a unified time-stamped record stream, write it to the local buffer in the order of arrival, and output a continuous record queue sorted by time stamp. In this embodiment, the edge computing module receives the unified time-stamped record stream output by the acquisition module and performs record caching, arrival sequence identification, time-stamp sorting, and sequence correction at the edge, ensuring that the continuous record queue read by the subsequent batch determination module has a clear writing order and timing relationship. The key to this process is to first preserve the original arrival order of the records, then organize the record positions according to the unified time-stamp, and then perform position correction on records that still have reverse order after sorting, so that the warehousing records, conveying records, weighing records, mixing records, and on-site inspection records in the same batch of production can be unfolded sequentially along the record queue. The local cache stores records in the order of the record units, and each record unit includes at least a unified time-stamp, arrival sequence number, signal type, subdivided signal type, and signal value. The cached record set is used to store the record results after writing to the local cache, and the continuous record queue is used to store the sorted and corrected record results. This implementation process includes the following steps: The purpose of receiving the unified time-stamped record stream and generating a cached record set is to preserve the order in which each record enters the edge side, so that the record position can still be determined even if the unified time stamps are the same in subsequent records. After the edge computing module reads the unified time-stamped record stream output by the acquisition module, it receives each record in the order of arrival and writes each record sequentially into the next free record unit in the local cache. When writing the current record, the edge computing module reads the number of records already written in the local cache, increments the number by one to get the arrival sequence number of the current record, and then writes the arrival sequence number, along with the unified time stamp, signal type, subdivision signal type, and signal value of the record into the corresponding record unit to form a cached record with an arrival sequence number. After all records are written, the edge computing module outputs the cached record set in the order of the record units in the local cache for subsequent sorting. When there are empty records, records with missing unified time stamps, or records with missing signal types in the unified time-stamped record stream, the edge computing module does not write the record into the local cache, but instead writes it into the abnormal record area and retains the original arrival position to avoid the missing field from affecting the sorting result after entering the continuous record queue. The purpose of sorting the cached record set and generating a continuous record queue is to establish a record arrangement result based primarily on a unified timestamp and secondarily on the arrival sequence number, ensuring that multiple records arriving at the same time maintain a definite order. After reading the cached record set, the edge computing module first sorts all cached records in ascending order using the unified timestamp as the first sorting key. Then, it sorts cached records with the same unified timestamp in ascending order using the arrival sequence number as the second sorting key, with records having an earlier unified timestamp placed first, and records with the same unified timestamp but a smaller arrival sequence number placed first. After sorting, the edge computing module... The block reads each cached record sequentially according to the sorting result, and connects the tail of the previous record to the head of the next record in sequence to form a continuous record queue. Each record in the continuous record queue still retains the unified timestamp, arrival sequence number, signal type, subdivision signal type and signal value from the original record unit. Subsequent batch determination modules can directly read in this order. When the cached record set is empty, the edge computing module outputs an empty queue and ends the current round of processing. When there is only one record in the cached record set, the edge computing module directly writes the record into the continuous record queue and ends the sorting process. The purpose of performing sequential correction on the continuous record queue and outputting a continuously increasing time-incrementing queue is to eliminate any remaining local inversions after sorting, ensuring that the unified time-increment of the next record is not less than that of the previous record. When outputting the continuous record queue, the edge computing module compares each record sequentially from the first record. For any subsequent record, it reads both the unified time-increment of that record and the unified time-increment of the previous record. If the unified time-increment of the subsequent record is not less than that of the previous record, it maintains the current position and continues comparing the next record. If the unified time-increment of the subsequent record is less than that of the previous record, it removes that record from the current position and starts comparing from the head of the queue again. The edge computing module searches backwards for the first record with a unified timescale greater than the unified timescale of the next record and inserts the removed record before it. After each insertion, the edge computing module returns to the record before the insertion position and continues the comparison until there are no records at the end of the queue with a unified timescale less than the unified timescale of the previous record. Then, it outputs a continuous queue of records with continuously increasing timescales for the batch determination module to read. When the unified timescale of the removed record is greater than or equal to the unified timescale of all records in the queue, the edge computing module directly inserts the removed record at the end of the queue. When the unified timescale of the removed record is less than the unified timescale of the first record in the queue, the edge computing module directly inserts the removed record at the beginning of the queue. Through the above processing, the edge computing module completes record writing, arrival order identification, double-key sorting, and local reverse order correction at the edge, ensuring that the continuous record queue simultaneously retains the original arrival relationship and the temporal relationship expanded according to a unified time scale. When the subsequent batch determination module reads this information, it can directly locate the batch interval by starting recording with weighing and completing the recording of material discharge, and sequentially extract the warehousing records, conveying records, weighing records, mixing records, and on-site inspection records within the batch interval along the queue sequence. Simultaneously, abnormal records are separated during the local caching stage and do not participate in the construction of the continuous record queue, preventing missing records from interfering with subsequent batch determination and acceptance criteria. In practical applications: when sampling... When the collection module continuously outputs multiple unified time-stamped record streams, the edge computing module first writes each record sequentially into its local cache and assigns an arrival sequence number according to the order of writing. Then, the edge computing module sorts the records by unified time stamp, and then sorts them by arrival sequence number if the unified time stamps are the same, forming an initial continuous record queue. If, due to interface transmission jitter, the unified time stamp of the later transport record is earlier than that of the previous weighing record, the edge computing module removes the transport record from its current position and inserts it before the first record with a unified time stamp greater than that transport record. After repeated comparisons and insertions, a continuous record queue is finally obtained for the batch determination module to read directly.

[0019] The batch determination module is used to read the continuous record queue, take the start time of a weighing as the batch start time and the corresponding discharge completion time as the batch end time, extract the warehousing records, conveying records, weighing records, mixing records and on-site inspection records between the start and end times, and output the current batch record set; In this embodiment, the batch determination module is used to delineate the record range corresponding to each preparation action from the continuous record queue, and organize the warehousing records, conveying records, weighing records, stirring records, and on-site testing records within this range into the current batch record set. The processing order is as follows: first, candidate batch intervals are established based on the weighing start record and the discharge completion record; then, the attribution determination is performed on the records that overlap between the candidate batch intervals; and finally, the corresponding records are extracted according to the assigned batch interval. The reason for this processing is that although the continuous record queue has been expanded according to a unified time scale, there are still cases where adjacent batches are connected one after another and records cross into two candidate batch intervals within the same production period. If the pairing and attribution are not completed first, the current batch record set read by the subsequent acceptance determination module will be mixed with records from adjacent batches. This implementation process includes the following steps: The purpose of reading each weighing start record and each discharging completion record from the continuous record queue and generating candidate batch intervals is to first determine a start and end point for each preparation action. After reading the continuous record queue, the batch determination module first scans all records in ascending order according to a unified time scale, extracting records with a subdivision signal type of "weighing start" as weighing start records and extracting records with a subdivision signal type of "discharging completion" as discharging completion records. Subsequently, taking each weighing start record as the starting point, it searches for the first discharging completion record after it and takes that discharging completion record as the corresponding end point. The time interval between the unified start time and the unified end time is recorded as a candidate batch interval. When there is no discharge completion record after a certain weighing start record, the batch determination module does not generate a candidate batch interval for that weighing start record, but instead writes the weighing start record into the unclosed record table for later matching after the discharge completion record is added. When there is no weighing start record before a certain discharge completion record, the batch determination module does not include the discharge completion record in the candidate batch interval, but instead writes the discharge completion record into the isolated end point record table to avoid the end point record being reverse-matched to the wrong start point. The purpose of performing attribution determination on the overlapping records between two candidate batch intervals is to assign records that fall into both candidate batch intervals to one of them, thus preventing identical records from entering two current batch record sets simultaneously. After all candidate batch intervals are generated, the batch determination module compares the start and end points of adjacent candidate batch intervals one by one. When the unified time scale of the start point of the later candidate batch interval is earlier than the unified time scale of the end point of the earlier candidate batch interval, it is determined that there are overlapping records between them. Subsequently, the unified time scale of each overlapping record within the overlapping range is read, and the absolute value of the time difference between this unified time scale and the unified time scale of the start point of the earlier candidate batch interval, the absolute value of the time difference between this unified time scale and the unified time scale of the end point of the earlier candidate batch interval, and the absolute value of the time difference between this unified time scale and the unified time scale of the end point of the earlier candidate batch interval are calculated respectively. The absolute value of the time difference between the starting point of a candidate batch interval and the ending point of the next candidate batch interval is calculated. Then, the absolute values ​​of the two time differences corresponding to the previous candidate batch interval are summed to obtain the time difference sum of the previous interval. The absolute values ​​of the two time differences corresponding to the next candidate batch interval are summed to obtain the time difference sum of the next interval. If the time difference sum of the previous interval is less than the time difference sum of the next interval, the cross record is assigned to the previous candidate batch interval. If the time difference sum of the next interval is less than the time difference sum of the previous interval, the cross record is assigned to the next candidate batch interval. When the two are equal, the cross record is assigned to the candidate batch interval with the earlier starting point unified time, and the assignment result is written into the cross record assignment table. Finally, non-overlapping batch intervals are obtained. The purpose of extracting the current batch record set based on the start and end points of each batch interval is to provide the subsequent acceptance and judgment module with a record set corresponding to only a single batch. After obtaining the batch interval, the batch determination module reads all records with a unified time scale between the start and end points from the continuous record queue, according to the unified time scale of the start and end points of each batch interval. Then, it filters out the warehousing records, conveying records, weighing records, mixing records, and on-site inspection records. Records belonging to the batch interval are written into the corresponding current batch record set in ascending order of the unified time scale. Records falling on the boundary of the batch interval are directly included in the batch interval. Cross records that have been written into the cross record attribution table are only included in one current batch record set according to the attribution result and are not written to other current batch record sets. When any type of record, such as warehousing records, conveying records, mixing records, or on-site inspection records, is missing in a certain batch interval, the batch determination module still outputs the current batch record set and attaches a missing record mark to the current batch record set so that the subsequent acceptance and judgment module can directly identify the missing record situation. Through the above processing, the batch determination module first establishes a correspondence between the weighing start record and the discharge completion record in the continuous recording queue, then assigns the cross-records to a single batch interval according to the time difference, and finally extracts the corresponding records according to the batch interval, so that each current batch record set corresponds to an independent preparation process record chain; when the subsequent acceptance judgment module reads the current batch record set, it can directly count the fly ash source records from the start of the batch to the end of weighing, read the change order of the weighing records, and perform judgment on the relationship between the stirring record and the discharge completion record, without having to deal with the problem of record mixing caused by the overlap of adjacent batches; in practical applications: when the first weighing start record, the first discharge completion record, the second weighing start record, and the second discharge completion record appear successively in the continuous recording queue, and the first discharge completion record and the second When there is a time overlap between weighing start records, the batch determination module first pairs the first weighing start record with the first discharge completion record that follows it to form the previous candidate batch interval, and then pairs the second weighing start record with the second discharge completion record that appears after the first discharge completion record to form the next candidate batch interval. If several mixing records fall into both candidate batch intervals at the same time, the batch determination module calculates the sum of the time differences from each mixing record to the start and end points of the candidate batch interval, and determines the candidate batch interval with the smaller sum of time differences as the interval to which the record belongs. After the assignment is completed, the batch determination module then extracts the warehousing records, conveying records, weighing records, mixing records, and on-site inspection records within the range of the previous and next candidate batch intervals to form two current batch record sets for subsequent modules to read.

[0020] The acceptance judgment module is used to read the current batch record set, determine whether the source of fly ash is unique from the start of the batch to the end of weighing, whether the weighing value increases continuously, and whether the start and stop times of stirring are between the end of weighing and the completion of discharge. Then, it screens out the control batch with the same fly ash source and the same process sequence from the historical batches, extracts the value range corresponding to each record item, determines whether all record items of the current batch fall into the corresponding value range, and outputs the acceptance result. In this embodiment, the acceptance judgment module is used to complete the basic constraint verification within the batch, historical batch screening, timescale alignment, interval generation, and interval comparison before the current batch record set enters the result control module, thereby determining whether the current batch record set has the basis to enter the performance prediction stage. Its processing order is as follows: first, verify the fly ash source, weighing process, and stirring sequence within the current batch record set; then, select a control batch that is related to the current batch record set from the historical batches; subsequently, convert the current batch record set and the control batch to the same normalized position for corresponding value extraction; then, form the value range for each record item accordingly; finally, the current batch record set is... The values ​​are compared item by item with the corresponding value ranges, and the accepted results are output. The source identifier in the fly ash source record is formed by concatenating the warehousing equipment number, storage silo number, and conveying channel number in a fixed order. The number of fly ash source types is the total number of all source identifiers after removing duplicates from the start of the batch to the end of weighing. The process sequence is determined by the order in which the warehousing record, conveying record, weighing record, mixing record, and on-site testing record first appear within the batch range. Each record item must include at least the weighing value, mixing time, and on-site testing value. Out-of-range record items refer to records where the current value is less than the minimum value of the corresponding value range or greater than the maximum value of the corresponding value range. This implementation process includes the following steps: First, a batch verification is performed on the current batch record set. This aims to exclude record sets with mixed sources, weighing bounces, and misaligned mixing sequences, preventing such record sets from directly entering the historical comparison process. The acceptance judgment module reads the fly ash source records, weighing records, mixing records, and on-site testing records from the current batch record set. It then extracts the source identifiers from all fly ash source records from the start of the batch to the end of the weighing process, and performs deduplication and counting on the source identifiers. The number of deduplicated source identifiers is recorded as the number of fly ash source types. When the number of fly ash source types is one, the source is considered valid. Subsequently, the weighing records are read in ascending order according to a unified time scale, starting from the second weighing record, subtracting the previous record from the next weighing value. The weighing values ​​form a forward difference sequence. When all forward differences are greater than or equal to zero, the weighing is considered successful. Then, the weighing end time, stirring start time, stirring stop time, and discharge completion time are read and compared according to their time order. When the weighing end time is not later than the stirring start time and the stirring stop time is not later than the discharge completion time, the timing is considered successful. Finally, the current batch record set that simultaneously satisfies the requirements of source success, weighing success, and timing success is written into the batch table to be compared, and the batch to be compared is output. If any one of the three conditions is not met, the unacceptable result is directly output, and the name of the unacceptable item, the corresponding unified time scale, and the corresponding record value are written into the exception table for the result control module to read. Subsequently, historical batch filtering and normalized position alignment are performed. The purpose is to map the batch to be compared to historical batches from the same source and with the same process direction to the same positional scale, facilitating comparison of records at the same process position. After reading the batch to be compared, the acceptance judgment module extracts historical batches with the same source identifier from the historical batch database. Then, it reads the process sequence of each historical batch and compares it item by item with the process sequence of the batch to be compared, retaining only historical batches with completely identical process sequences as control batches. If the number of historical batches with the same source identifier is insufficient, it continues to read historical batches with the same process sequence but different source identifiers from the historical batch database. Control batches are added based on the number of identical items in the source identifier (warehouse equipment number, storage warehouse number, and conveyor channel number) from highest to lowest, until the set number of reads is reached. The set number of reads is taken from the historical comparison batch count in the system configuration table. The system configuration table is written to the edge storage area during system deployment. After the comparison batch screening is completed, the batch start time and batch end time of each comparison batch and the batch to be compared are read respectively. The normalized time is calculated by subtracting the batch start time from the record's normalized time and then dividing by the difference between the batch end time and the batch start time, so that the batch start corresponds to zero and the batch end corresponds to one. Then, the normalized time of each record item in the batch to be compared is read. In each comparison batch, the two records before and after the normalized time of the same record item are read. If the normalized time is the same as the normalized time of a record, the value of that record is directly taken. If it is between two records, the interpolation is calculated according to the distance ratio between the two records. If it is before the first record or after the last record, the value of the first record or the last record is taken respectively. Thus, the corresponding value set of each comparison batch at each normalized time is obtained and written into the comparison value table for the next step. Next, the corresponding value set is cleaned and the value range is generated. The purpose is to remove values ​​that deviate from the set from multiple batches of corresponding values ​​at the same normalization position, and then use the retained values ​​to generate the comparison boundary for each record item. After reading the comparison value table, the acceptance judgment module sorts the corresponding value set of each record item at the same normalization time scale by numerical size, and takes the value corresponding to the middle position of the sort as the median. Then, it calculates the absolute difference between each corresponding value in the corresponding value set and the median, and sorts all the absolute differences by numerical size again, taking the absolute difference corresponding to the middle position of the sort as the median of the absolute difference, and records the median of the absolute difference as the discrete quantity. Subsequently, corresponding values ​​with an absolute difference greater than twice the discrete quantity are deleted, and the median calculation, absolute difference calculation, and deletion calculation are re-executed for the retained values ​​after deletion. The process is as follows: If the number of retained values ​​in the previous round is the same as that in the next round, the iteration ends; otherwise, the next round of calculation continues. When the discrete value is zero, only the corresponding values ​​with an absolute difference greater than zero are deleted. If all absolute differences are zero, all corresponding values ​​are retained. After the iteration ends, the minimum and maximum values ​​among the retained values ​​are read. The minimum value is recorded as the lower bound of the value interval of the record item at the normalized time scale, and the maximum value is recorded as the upper bound of the value interval of the record item at the normalized time scale. The lower and upper bounds of the value intervals are written into the interval table for subsequent comparison. If the set of corresponding values ​​at a certain normalized time scale is empty, the acceptance judgment module does not generate the value interval for that position, but writes a missing marker in the interval table so that the subsequent comparison process can directly identify that there is no historical support at that position. Finally, a comparison is performed item by item between the current value and the value range. The purpose is to output a pass or fail acceptance result based on whether the current value of the batch under comparison at each normalized position falls within the historical support range. The acceptance determination module reads the current value and its normalized time scale of each record item in the batch under comparison, and then reads the lower and upper bounds of the value range for the corresponding record item at the same normalized time scale from the interval table. Subsequently, a comparison is performed item by item according to the record item name and normalized time scale. When the current value is greater than or equal to the lower bound of the value range and less than or equal to the upper bound of the value range, the record item is recorded as an item within the range; when the current value is less than the upper bound of the value range, the record item is recorded as an item within the range. When the lower bound of the range is greater than the upper bound of the range, the record is marked as an out-of-range record and simultaneously written into the out-of-range record table. The out-of-range record table includes at least the record name, unified timestamp, current value, lower bound of the range, and upper bound of the range. After all record items are compared, if there are no out-of-range record items and no missing markers, the accepted result is output. If there are out-of-range record items or missing markers, the unaccepted result and the corresponding out-of-range record table are output. When the field detection record contains multiple detection values, each detection value is treated as an independent record item and participates in the above comparison, and corresponding comparison results are generated respectively. Through the above processing, the acceptance judgment module first completes the source, weighing, and timing verification within the current batch record set. Then, it aligns the batch to be compared with historical batches that have passed to the same normalized position for numerical comparison. Finally, it generates acceptance results for passing or failing and the corresponding out-of-bounds record table. This ensures that the acceptance results read by the result control module include both the judgment result of whether to enter performance prediction and the specific out-of-bounds position and content when failing. This processing method, on the one hand, removes the current batch record set with mixed sources, weighing bounces, and misaligned stirring timing before comparison. On the other hand, it converts historical batches with different batch durations to the same process position before forming a value range, so that subsequent comparisons are based on historical records falling at the same position, rather than directly comparing the original time values ​​under different batch lengths. In practical applications: when the source identifier of a batch to be compared is the warehousing equipment number A1, storage warehouse number C3, and conveying channel... After L2 is assembled, the acceptance judgment module first filters out control batches from the historical batch database that have the same source identifier A1-C3-L2 and whose process sequence is warehousing, conveying, weighing, stirring, and testing. Then, the unified time scale of these control batches and the batch to be compared is converted to a normalized time scale between zero and one. At the positions of 0.3, 0.5, and 0.8 of the normalized time scale, the corresponding value sets of weighing value, stirring time, and test value are extracted. Subsequently, the median and discrete value of the corresponding value set at each position are calculated and the deviation value is deleted. Then, the value range is generated using the retained value. If the cumulative value of the batch to be compared at the normalized time scale of 0.5 falls within the corresponding value range, and the test value at the normalized time scale of 0.8 is greater than the upper limit of the corresponding value range, then the test value is written into the out-of-bounds record table. The acceptance judgment module outputs the acceptance result of failure and the corresponding out-of-bounds record table. The result control module stops the batch from entering the performance prediction based on this.

[0021] The result control module is used to read the current batch record set and the acceptance result. When the acceptance result is passed, it extracts the fly ash source record, weighing record, mixing record and on-site test record from the current batch record set as performance prediction input and outputs the performance prediction result. When the acceptance result is failed, it outputs the prohibition of acceptance mark and the corresponding out-of-bounds record item. In this embodiment, the result control module is used to convert the current batch record set into the result format required for subsequent output after the acceptance judgment module provides the acceptance result. When the acceptance result is "pass," the result control module extracts the record items constituting the performance prediction input from the current batch record set, and further selects historical prediction sequences that are close to the current prediction sequence from the historical pass batches to calculate the performance prediction result. When the acceptance result is "fail," the result control module no longer performs performance prediction, but generates an out-of-bounds record table based on the out-of-bounds record items and outputs a prohibition on acceptance flag. This process is preceded by the acceptance result and out-of-bounds record items from the acceptance judgment module, and followed by the output process of the performance prediction result or the prohibition on acceptance flag. Its core is to guide the pass and fail states to different result paths, so that the subsequent caller can directly read the corresponding results. This implementation process includes the following steps: The purpose of generating the current prediction sequence when the acceptance result is "pass" is to compress the current batch record set into a prediction input with a fixed field order, so as to compare it item by item with the historical prediction sequence. After reading the current batch record set, the result control module first extracts the source identifier from the fly ash source record, where the source identifier is taken from the fixed-order splicing value of the silo equipment number, storage silo number and conveying channel number in the fly ash source record; then it extracts the final weighing value from the weighing record, where the final weighing value is taken from the weighing value of the latest weighing record with the unified time scale in the current batch record set; and then it extracts the stirring time from the stirring record, where the stirring time is calculated by subtracting the stirring start time from the stirring stop time. Next, the test values ​​are extracted from the on-site test records. The test values ​​are taken from the on-site test records in the current batch record set where both the test items and test values ​​exist. When the on-site test record contains multiple test values, they are extracted and concatenated in the order of the test items written in the system configuration table. After the above extraction is completed, the result control module generates the current prediction sequence in the order of source identifier, final weighing value, stirring time, and test value, and writes the current prediction sequence into the prediction input table for subsequent nearest neighbor calculations. When any of the above record items are missing in the current batch record set, the result control module does not generate the current prediction sequence, but outputs a prohibition flag and writes the name of the missing record item into the exception output table. The purpose of calculating the performance prediction result for the current prediction sequence is to generate the performance prediction result for the current batch by utilizing the performance results of historical prediction sequences that are close to the current prediction sequence in the historical batches. After reading the current prediction sequence, the result control module first extracts historical prediction sequences with the same source identifier from the historical batches. Each historical prediction sequence is stored in a one-to-one correspondence with a historical performance result, which is taken from the on-site inspection record or subsequent supplementary inspection record in the corresponding historical batch. If there are at least three historical prediction sequences with the same source identifier, the difference calculation is directly performed. If there are fewer than three, historical prediction sequences with the same process sequence are extracted from the historical batches, and the number of identical items in the source identifier (warehouse equipment number, storage number, and conveyor channel number) is increased to three. If the total number of historical batches is less than three, all extracted historical prediction sequences are used in subsequent calculations. Subsequently, the result control module... The module calculates the difference between the current predicted sequence and each historical predicted sequence item by item according to the corresponding position. The source identifier position is calculated based on whether the codes are the same. If the codes are the same, it is recorded as zero, and if the codes are different, it is recorded as one. The final weighing value, stirring time, and detection value position are calculated by subtracting the absolute value of the corresponding position value from the current position value. Then, the absolute values ​​of each position are summed to obtain the sum of the absolute values ​​corresponding to the historical predicted sequence. After completing the calculation of all historical predicted sequences, the three historical predicted sequences with the smallest sum of absolute values ​​are taken as the nearest neighbor sequences. If the minimum sum of absolute values ​​is zero, the historical performance result corresponding to the nearest neighbor sequence is directly output as the performance prediction result. If the sum of the absolute values ​​of the three nearest neighbor sequences is not zero, the reciprocal of the sum of the absolute values ​​of each is taken as the weight numerator, and the sum of the weight numerators is used as the weight denominator. A weighted average is then performed on the performance results corresponding to the three nearest neighbor sequences, and the performance prediction result is output. The performance prediction result is written to the result output table for the calling end to read. The purpose of generating a prohibition flag and an out-of-bounds record table when the acceptance result is "not approved" is to retain the reason for the failure and the corresponding position, so that the calling end can directly read the blocking information. After the result control module reads the failure acceptance result and the corresponding out-of-bounds record item output by the acceptance judgment module, it first extracts the record name, unified time stamp and record value from each out-of-bounds record item, then arranges all out-of-bounds record items in ascending order of unified time stamp, and writes each arranged out-of-bounds record item into the out-of-bounds record table. The out-of-bounds record table includes at least the record name, unified time stamp and record value. If the acceptance judgment module outputs both the lower limit and the upper limit of the value range, the result control module writes both into the out-of-bounds record table. After completing the writing of the out-of-bounds record table, the result control module outputs the prohibition flag and the out-of-bounds record table for subsequent calling ends to read. When the acceptance judgment module outputs a failure acceptance result but does not return an out-of-bounds record item, the result control module writes an empty table into the out-of-bounds record table position and outputs the prohibition flag simultaneously to ensure that the output structure is fixed in the failure state. Through the above processing, when the acceptance result is "pass," the result control module converts the current batch record set into the current prediction sequence and outputs the performance prediction result through nearest neighbor calculation of the historical prediction sequence. When the acceptance result is "fail," the result control module stops performance prediction and outputs a prohibition flag and an out-of-bounds record table. This, on the one hand, fixes the source identifier, final weighing value, stirring time, and detection value into a unified input structure, allowing for item-by-item comparison between the historical and current prediction sequences; on the other hand, it directly organizes the abnormal content in the "fail" state into an out-of-bounds record table, so that subsequent calling terminals do not need to parse the current batch record set again to obtain the blocking reason and corresponding location. In practical applications: when the acceptance result corresponding to a certain current batch record set is "pass," the result control module first extracts the source identifier A1-C from the fly ash source record. 3-L2 extracts the final weighing value from the weighing record, calculates the mixing time from the mixing record, and extracts the detection value from the on-site detection record to generate the current prediction sequence. Subsequently, it extracts historical prediction sequences with the same source identifier A1-C3-L2 from historical batches that have passed the test. It calculates the sum of the absolute values ​​of the current prediction sequence and each historical prediction sequence, selects the three historical prediction sequences with the smallest sum of absolute values, and uses the corresponding historical performance results to calculate the performance prediction result of the current batch. When the acceptance result corresponding to another current batch record set is "not passed", the result control module no longer generates the current prediction sequence, but directly reads the out-of-bounds record item output by the acceptance judgment module, writes the record name, unified time scale, and record value in the out-of-bounds record item into the out-of-bounds record table in ascending order of the unified time scale, and outputs the prohibition of acceptance flag and the out-of-bounds record table.

[0022] In addition, as a further embodiment of this solution, a chloride ion content acquisition and writing process is also included. Specifically, the acquisition module, based on acquiring fly ash entry signals, conveying start / stop signals, weighing signals, stirring start / stop signals, and on-site detection signals, simultaneously acquires the chloride ion content record corresponding to the current batch, and writes the chloride ion content record as a detection record in the on-site detection record into the original event record; during the event queue sorting, unified timescale conversion, and continuous record queue generation process, the chloride ion content record is written and sorted together with other on-site detection records according to a unified timescale; when the batch determination module truncates the current batch record set, the chloride ion content records located within the batch interval are also truncated; in the acceptance judgment module... When the block execution history is processed through batch screening, normalized timescale conversion, corresponding value set generation, value range calculation, and current value range comparison, chloride ion content is included as a record item in the on-site detection values ​​for comparison. When the chloride ion content exceeds the corresponding value range, the chloride ion content record is written into the out-of-bounds record item. When the acceptance result is passed, the result control module writes the detection value corresponding to the chloride ion content, along with the source identifier, final weighing value, stirring time, and other detection values, into the current prediction sequence. When the acceptance result is failed, the result control module writes the out-of-bounds record corresponding to the chloride ion content into the out-of-bounds record table according to a unified timescale. This ensures that the chloride ion content is integrated throughout the collection, batch determination, acceptance judgment, and result control processes. By incorporating chloride ion content into the same record generation, batch extraction, historical alignment, and interval comparison process, the acceptance judgment of the current batch no longer relies solely on source, weighing, stirring, and routine testing records. Instead, it further combines chloride ion content changes related to material properties and usage conditions for joint judgment. In this way, when abnormal changes in chloride ion content occur due to changes in fly ash source, shared storage, or local raw material composition fluctuations, the system can directly identify the corresponding anomalies during the acceptance judgment stage and block such batches from entering the normal prediction path before performance prediction. This makes the comparison basis between the current batch record set and historical batches more closely reflect the actual raw material state, reducing acceptance bias caused by chloride ion content deviations not being included in the judgment.

[0023] Working Principle: This solution first receives fly ash silo entry, conveying start / stop, weighing, stirring start / stop, and on-site detection data from the acquisition module. It then converts data from different interfaces to the same time base, forming a unified time-stamped record stream. Subsequently, the edge computing module caches, sorts, and corrects the data according to the arrival order and the unified time scale, resulting in a continuous record queue. Based on this, the batch determination module defines the batch interval corresponding to each preparation process based on the weighing start record and the discharge completion record, extracting various records within that batch to form the current batch record set. The acceptance judgment module first checks whether the batch has mixed sources, abnormal weighing, or misaligned timing. Then, it aligns and compares the batch with historically passed batches at the same process position to determine if each record item in the current batch falls within the historical support range. If the comparison passes, the result control module extracts the source identifier, final weighing value, stirring time, and detection value from the current batch, and provides performance prediction results based on similar historical batches. If the comparison fails, it directly outputs a prohibition on acceptance flag and out-of-bounds record items, thus preventing abnormal batches from entering subsequent predictions. For example, in production sites where fly ash sources frequently change, storage silos are shared, and continuous feeding occurs without interruption, multiple records such as silo entry, conveying, and weighing may appear simultaneously within the same time period, and the time sequence of data uploads from different devices may not be entirely consistent. This solution first organizes these scattered data into continuous records spread out over time, then identifies which record belongs to the same preparation process, and then determines whether this batch of fly ash always came from the same source before the weighing process ended, whether the weighing process continued to increase, and whether the stirring occurred within the correct time period. If these basic relationships are established, the batch is then compared item by item with historical normal batches at the same process position. Only after confirming that the current record has not exceeded the historical range is performance prediction allowed. If a record is found to have exceeded the historical range, the system will not continue prediction, but will directly indicate which record is abnormal and at what time, making it easier for on-site personnel to determine whether the batch needs to be reviewed or rejected.

[0024] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for predicting the performance and optimizing the formulation of fly ash-based cementitious materials, characterized in that, include: The data acquisition module is used to collect the time of fly ash entering the silo, the time of start and stop of conveying, the time of start and stop of weighing, the time of start and stop of mixing, and on-site test records, and output a unified time-stamped record stream; The edge computing module is used to receive a unified time-stamped record stream, write it to the local buffer in the order of arrival, and output a continuous record queue sorted by time stamp. The batch determination module is used to read the continuous record queue, take the start time of a weighing as the batch start time and the corresponding discharge completion time as the batch end time, extract the warehousing records, conveying records, weighing records, mixing records and on-site inspection records between the start and end times, and output the current batch record set; The acceptance judgment module is used to read the current batch record set, determine whether the source of fly ash is unique from the start of the batch to the end of weighing, whether the weighing value increases continuously, and whether the start and stop times of stirring are between the end of weighing and the completion of discharge. Then, it screens out the control batch with the same fly ash source and the same process sequence from the historical batches, extracts the value range corresponding to each record item, determines whether all record items of the current batch fall into the corresponding value range, and outputs the acceptance result. The result control module is used to read the current batch record set and the acceptance result. When the acceptance result is passed, it extracts the fly ash source record, weighing record, mixing record and on-site test record from the current batch record set as the performance prediction input and outputs the performance prediction result. When the acceptance result is failed, it outputs the prohibition of acceptance mark and the corresponding out-of-bounds record item.

2. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 1, characterized in that: The execution of the acquisition module includes: Arrival trigger acquisition is performed on fly ash silo entry signal, conveying start / stop signal, weighing signal, stirring start / stop signal and on-site detection signal, the occurrence time and signal type of each signal are read, and the corresponding raw event record is generated; The occurrence times in each original event record are converted into a unified time scale under the same timing benchmark, and the signal type, unified time scale and signal value are written into the event queue in a single record manner to generate a unified time scale record stream. Based on the time sequence of two adjacent records in the event queue, the records with the same time stamp are rearranged in the order of fly ash entering the warehouse, conveying start and stop, weighing start and stop, stirring start and stop, and on-site detection, and then the record stream with the same time stamp is output.

3. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 2, characterized in that: The acquisition module performs the following actions when reading the signal type: For each signal acquired upon arrival, the interface number, number of fields, field order, status bit, current value, previous value, and time are read. The field difference, time difference, and value difference are calculated. The sum of the absolute values ​​of the field differences, the absolute value of the time difference, and the absolute value of the value difference are summed with the reference records of each candidate signal type. The candidate signal type with the smallest sum is taken as the initial signal type. When the sum of two or more candidate signal types is the same, read the established signal types of the records adjacent to the current signal, substitute each candidate signal type into the current position to form a three-item type sequence, and then compare each three-item type sequence with the process sequence from warehousing to conveying, conveying to weighing, weighing to stirring, and stirring to testing item by item, calculate the number of missing items and the number of reverse order, and take the candidate signal type with the smallest sum of missing items and reverse order as the correction signal type; The signal status is read again according to the corrected signal type. The cumulative value and the difference between adjacent values ​​of the symmetrical signal are calculated. The record where the cumulative value increases for the first time is defined as the start, the record where the difference between adjacent values ​​is continuously non-zero is defined as the process, and the record where the cumulative value remains unchanged and the subsequent records have the same value is defined as the end. The start and stop of the silo signal, conveying signal and stirring signal are determined by the change of the status bit from zero to one and from one to zero, respectively. The entry and result of the detection signal are determined by the presence or absence of the detection item and detection value in the field, and the subdivided signal type is output.

4. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 2 or 3, characterized in that: The execution of the edge computing module includes: Receive a unified timestamp record stream, write each record to the local buffer in the order of arrival, generate a corresponding arrival sequence number for each record, and output a buffer record set with the arrival sequence number; The cached record set is sorted in ascending order according to a unified time index. If the unified time index is the same, it is sorted in ascending order according to the arrival sequence number. The sorted records are then connected end to end in sequence to output a continuous record queue. When outputting the continuous record queue, the unified time index of the next record is compared with the unified time index of the previous record one by one. Records whose unified time index is less than that of the previous record are removed from their current positions and inserted before the first record whose unified time index is greater than that record. Then, the continuous record queue with continuously increasing time indexes is output.

5. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 4, characterized in that: The execution of the batch determination module includes: Read each weighing start record and each discharge completion record from the continuous record queue, pair each weighing start record with the first discharge completion record that follows it, and generate a candidate batch interval; For the cross record between two candidate batch intervals, calculate the absolute value of the time difference between the unified time scale of each cross record and the starting point of the two candidate batch intervals, as well as the absolute value of the time difference between the unified time scale of each cross record and the ending point of the two candidate batch intervals. Sum the two sets of absolute time difference values ​​respectively, and determine the candidate batch interval with the smallest sum of absolute time difference values ​​as the interval to which the cross record belongs, thus generating the batch interval. Based on the start and end points of each batch interval, extract the warehousing records, conveying records, weighing records, mixing records, and on-site testing records located between the corresponding start and end points from the continuous record queue, and output the current batch record set in ascending order according to a unified time scale.

6. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 5, characterized in that: The execution of the acceptance determination module includes: Read the fly ash source records, weighing records, mixing records, and on-site testing records from the current batch record set. Count the number of fly ash source types from the start of the batch to the end of the weighing process, and mark a source as established if the number of types is one. Calculate the forward difference between adjacent weighing records, and mark weighing as established if all forward differences are greater than or equal to zero. Then compare the order of the weighing end time, mixing start time, mixing stop time, and discharge completion time, and mark the time sequence as established if the weighing end time is not later than the mixing start time and the mixing stop time is not later than the discharge completion time. Output the current batch record set that simultaneously satisfies the criteria of source establishment, weighing establishment, and time sequence establishment as the batch to be compared.

7. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 6, characterized in that: The execution of the acceptance determination module also includes: For the batch to be compared, control batches with consistent fly ash sources and consistent process sequences are screened out from historical batches. For each control batch, the unified timescale is converted into a normalized timescale based on the duration from the start to the end of the batch. Then, the unified timescale of each record item in the batch to be compared is converted into the corresponding normalized timescale. In each control batch, linear interpolation of neighboring points is performed on the same record item according to the normalized timescale to obtain the set of corresponding values ​​of each control batch at each normalized timescale.

8. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 7, characterized in that: The execution of the acceptance determination module also includes: For each record item, first calculate the median of the set of corresponding values ​​at the same normalized time scale, then calculate the absolute difference between each corresponding value and the median, and take the median of the absolute differences as the discrete quantity. After deleting the corresponding values ​​whose absolute difference is greater than twice the discrete quantity, repeat the median calculation and deletion calculation until the number of retained values ​​is the same in the two rounds. Then, determine the minimum and maximum values ​​among the retained values ​​as the value range of the record item at the normalized time scale. The current value of each record in the batch to be compared is compared with the value range at the corresponding normalized time scale. When all current values ​​fall within the corresponding value range, the accepted result is output. When there is a current value that falls outside the corresponding value range, the accepted result and the corresponding out-of-range record are output.

9. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 8, characterized in that: The execution of the result control module includes: When the acceptance result is passed, the source identifier of the fly ash source record, the mixing time of the mixing record, and the test value of the on-site test record are extracted from the current batch record set, and the current prediction sequence is generated in the order of source identifier, final weighing value, mixing time, and test value.

10. The fly ash-based cementitious material performance prediction and formulation optimization system according to claim 9, characterized in that: The execution of the result control module also includes: For the current prediction sequence, extract historical prediction sequences with the same source identifier from the historical batches. Calculate the sum of the absolute values ​​of the position differences between the current prediction sequence and each historical prediction sequence. Take the three historical prediction sequences with the smallest sum of absolute values ​​as the nearest neighbor sequences. Then, calculate the weighted average of the performance results corresponding to the three nearest neighbor sequences according to the reciprocal of the sum of their absolute values, and output the performance prediction result. When the acceptance result is not approved, extract the record name, unified time stamp and record value corresponding to the out-of-bounds record item, generate the out-of-bounds record table in ascending order of unified time stamp, and output the prohibition of acceptance flag and the out-of-bounds record table.

Images