An enterprise procurement demand prediction method based on big data analysis

Abstract

The application discloses an enterprise procurement demand prediction method based on big data analysis, which comprises the following steps: collecting procurement, material taking, maintenance, planning, inventory and delivery date data and generating a procurement trigger account book; constructing double waiting clock and un-clear trigger balance based on the procurement trigger account book and historical actual consumption record and generating a renewal purchase intensity result set; generating a procurement pre-occupation slot set in combination with the delivery date, review, receipt, starting order and arrival constraints; executing sparse integer transportation and generating pre-occupation and net procurement results; performing single, batch and whole quantity auditing on the pre-occupation and net procurement results and generating a procurement suggestion result set; and performing mirror playback reconciliation according to the execution, arrival and taking results and rewriting core objects. The application adopts the procurement trigger account book and sparse integer transportation method, realizes enterprise procurement demand prediction and suggestion generation, and has the advantages of accurate prediction, strong execution and closed-loop optimization.

Description

Technical Field

[0001] This invention relates to the field of enterprise supply chain management technology, and in particular to a method for predicting enterprise procurement demand based on big data analysis. Background Technology

[0002] Existing enterprise procurement demand forecasting technologies typically rely on Enterprise Resource Planning (ERP) systems, Material Requirements Planning (MRP) systems, or supply chain collaboration platforms. They utilize data such as purchase orders, production plans, inventory ledgers, and historical requisition records, employing rule-based calculations, statistical forecasting, or machine learning models to generate material requirements, replenishment quantities, or ordering suggestions. Some solutions further integrate supplier delivery dates, procurement cycles, in-transit inventory, and safety stock to calculate lead time demand, assisting in order placement, replenishment arrangements, and supply chain coordination. These solutions have already been applied in discrete manufacturing, equipment maintenance, and multi-category inventory management scenarios.

[0003] However, existing technologies mostly rely on historical consumption sequences or static planning data, making it difficult to uniformly compile and sequentially express heterogeneous triggering information such as maintenance work orders, substitution relationships, delivery anomalies, and order cancellation reductions. This results in the difficulty of continuously depicting the demand formation process. At the same time, existing solutions typically output purchase quantities or replenishment quantities directly, lacking modeling of supporting units that address review cycles, receiving capabilities, minimum order constraints, and delivery compatibility constraints. They also lack a mirror playback update mechanism that writes back the actual purchase execution results, actual delivery results, and actual requisition results to the preceding state objects. Therefore, they are deficient in sparse demand, multi-constraint execution, and closed-loop correction.

[0004] Therefore, how to provide a method for predicting enterprise procurement demand based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a method for predicting enterprise procurement demand based on big data analysis. This invention employs a procurement trigger ledger and a sparse integer transport method to achieve enterprise procurement demand prediction and suggestion generation, possessing the advantages of accurate prediction, strong executability, and closed-loop optimization.

[0006] A method for predicting enterprise procurement demand based on big data analysis according to an embodiment of the present invention includes the following steps:

[0007] Retrieve purchase orders, material requisition records, maintenance work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery date records; perform time alignment and object mapping; and generate a purchase trigger ledger.

[0008] Based on the procurement trigger ledger and historical consumption records, a consumption waiting clock, a trigger waiting clock, and an outstanding trigger balance are constructed to generate a set of renewal intensity results.

[0009] Based on supplier delivery records, review cycles, receiving capacity, minimum order quantity constraints, and delivery compatibility constraints, a set of pre-reserved procurement slots is generated;

[0010] Based on the repurchase intensity result set, the outstanding trigger balance and the purchase pre-occupied slot set, sparse integer transport is performed to generate a slot-level pre-occupancy result set and a material-level net purchase quantity result set;

[0011] Based on the slot-level pre-occupancy result set and the material-level net purchase quantity result set, perform single-order decoding, batch decoding and whole-quantity auditing to generate a purchase suggestion result set;

[0012] Based on the procurement suggestion result set, actual procurement execution result, actual delivery result, and actual requisition result, a mirror replay reconciliation is performed to generate an updated configuration set and write back the procurement trigger ledger, consumption waiting clock, trigger waiting clock, and procurement pre-occupied slot set.

[0013] Optionally, the generation of the procurement trigger ledger specifically includes:

[0014] Extract source fields from purchase orders, material requisition records, maintenance work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery date records; perform field normalization, unit of measurement unification, quantity direction normalization, and missing data validation to form the original set of purchase source records.

[0015] Perform source time difference correction, object time difference correction and granularity consolidation on the time field in the original procurement source record set to form a unified time index result set;

[0016] Calculate object mapping scores based on material identifiers, source object identifiers, supplier identifiers, and substitution relationships in the original procurement source record set, and form an object mapping result set;

[0017] Extract procurement trigger candidate records from the object mapping result set to form a procurement trigger candidate set;

[0018] The procurement trigger candidate set is deduplicated and merged, net direction amount synthesized, and source credibility sorted to form a procurement trigger item set;

[0019] The procurement trigger entry set is organized into a ledger order based on material identifier, unified time index, and source object identifier, and ledger encapsulation is performed to generate the procurement trigger ledger.

[0020] Optionally, the generation of the renewal strength result set specifically includes:

[0021] Based on the material identifier and unified time index, the procurement trigger ledger and historical consumption records are indexed and sorted to form a clock update input set.

[0022] Update the consumed waiting clock index by index according to the clock update input set to form a consumed waiting clock result set;

[0023] Extract the accumulated values ​​of positive trigger effect strength and negative cancellation effect strength from the clock update input set, update the trigger wait clock index by index, and form a trigger wait clock result set;

[0024] Update the outstanding trigger balance according to the carry-over relationship of positive triggering, reverse offsetting and historical actual consumption, and form an outstanding trigger balance result set;

[0025] Perform dimensional unification and credibility correction on the consumption waiting clock result set, trigger waiting clock result set, outstanding trigger balance result set and source credibility aggregate value to form the input set for repurchase intensity calculation;

[0026] Based on the input set for calculating renewal strength, perform renewal strength synthesis to generate a result set of renewal strength.

[0027] Optionally, the consumption waiting clock is updated according to the following rules: when there is a historical consumption quantity at the unified time index, the corresponding consumption waiting clock is set to zero; when there is no historical consumption quantity at the unified time index, the consumption waiting clock at the previous unified time index is incremented by one as the current consumption waiting clock; when there is no historical consumption quantity at the first unified time index, the unified time granularity quantity from the observation start point to the unified time index is written into the consumption waiting clock to form a consumption waiting clock result set.

[0028] Optionally, the trigger waiting clock is updated according to the following rules: when the cumulative value of the positive trigger influence intensity minus the cumulative value of the reverse cancellation influence intensity at the current unified time index is still greater than zero, the corresponding trigger waiting clock is set to zero; when the difference is less than or equal to zero, the trigger waiting clock at the previous unified time index is incremented by one as the current trigger waiting clock; when there is no net positive trigger at the first unified time index, the number of unified time granularities from the observation start point to the unified time index is written into the trigger waiting clock to form a trigger waiting clock result set.

[0029] Optionally, the generation of the procurement pre-reserved slot set specifically includes:

[0030] Extract material identifiers and supplier identifiers from the procurement trigger ledger, and align them with supplier delivery date records, review cycles, receiving capacity, minimum order quantity constraints, and arrival compatibility constraints to form a procurement pre-reservation slot generation input set;

[0031] Based on the input set generated from the pre-reserved slots for procurement, the review release period, the delivery start period, and the expected delivery period are determined, forming a candidate set for slot boundaries.

[0032] Perform arrival compatibility screening on the candidate set of slot boundaries to form an arrival compatibility determination result set;

[0033] Based on the set of arrival compatibility determination results, the receiving capacity is pruned and the minimum order quantity constraint is mapped to form the slot bearing result set;

[0034] The slot identifiers of the valid slot boundary candidates marked in the slot bearing result set are arranged and the fields are encapsulated to form a set of procurement pre-occupied slot entries;

[0035] Create a sequential index for the set of pre-reserved slot entries and perform set encapsulation to generate a set of pre-reserved slots.

[0036] Optionally, the generation of the slot-level pre-occupancy result set and the material-level net purchase quantity result set specifically includes:

[0037] Perform index alignment and object aggregation on the repurchase intensity result set, the outstanding trigger balance result set, and the purchase pre-occupied slot set to form a sparse integer transport input set;

[0038] Establish slot-edge relationships around the sparse integer transport input set to form a transport candidate relationship set;

[0039] Perform feasibility screening and priority ranking on the set of candidate transport relationships to form a set of feasible transport relationships;

[0040] Based on the set of feasible transport relationships, integer allocation and capacity deduction are performed to form a set of transport allocation results;

[0041] The set of transportation allocation results is merged into slots and materials to form a set of slot-level pre-occupancy results and a set of material-level net purchase quantity results.

[0042] Optionally, the integer allocation starts with the outstanding trigger balance under the same material identifier and the same unified time index. Purchase pre-reserved slot entries are read one by one according to the priority order in the feasible transport relationship set. For each purchase pre-reserved slot entry, an integer pre-reserved amount not greater than the current remaining maximum capacity of the purchase pre-reserved slot entry and not greater than the current outstanding trigger balance is allocated. When the current outstanding trigger balance is less than the minimum capacity of the corresponding purchase pre-reserved slot entry, the purchase pre-reserved slot entry is skipped and the next purchase pre-reserved slot entry is read. After the integer pre-reserved amount is written, the current remaining maximum capacity of the corresponding purchase pre-reserved slot entry and the current outstanding trigger balance of the corresponding material identifier are deducted simultaneously until the current outstanding trigger balance is zero or the feasible transport relationship set is traversed completely, forming a transport allocation result set.

[0043] Optionally, the generation of the procurement suggestion result set specifically includes:

[0044] Perform field alignment and sequence aggregation on the slot-level pre-occupancy result set and the material-level net purchase quantity result set to form a purchase suggestion decoding input set;

[0045] Based on the procurement recommendations, the decoded input set is divided into single segments and continuous slots are merged to form a single candidate set.

[0046] Batch decoding is performed around the candidate set to form a batch decoding result set;

[0047] Perform integer quantity auditing on the batch decoding result set to form an integer quantity audit result set;

[0048] The audit results for whole quantities are written into the suggested order placement period, suggested delivery period, and suggested quantity to form a set of procurement suggestion items;

[0049] The procurement suggestion items are sorted in ascending order by suggested order time under the same material identifier, by supplier identifier under the same suggested order time, and by suggested delivery time under the same supplier identifier, and a procurement suggestion result set is generated.

[0050] Optionally, the generation of the updated configuration set specifically includes:

[0051] The execution of procurement recommendations results set includes collecting actual procurement execution results, actual delivery results, and actual requisition results. The execution fields are aligned and time periods are aggregated to form a mirror playback input set.

[0052] The suggested paths and execution paths are reconciled one by one according to the input set of the mirror replay, forming a set of mirror replay difference results;

[0053] Based on the set of differences in the mirror playback results, the responsible object is attributed and the updated object is located, forming an updated location result set;

[0054] Based on the updated location result set, perform parameter correction and value write-back preparation to form an updated configuration entry set;

[0055] Conflict resolution and sequential arrangement are performed on the set of updated configuration entries to form an updated configuration set;

[0056] The results are written back to the procurement trigger ledger, consumption waiting clock, trigger waiting clock, and procurement pre-occupied slot set in the order of the updated configuration set, completing the mirror replay reconciliation loop.

[0057] The beneficial effects of this invention are:

[0058] This invention constructs a procurement trigger ledger, unifying, aligning, and mapping purchase orders, material requisition records, maintenance work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery records. This organizes previously scattered, heterogeneous procurement-related data into a unified, continuously accessible input. Combined with consumption wait time clocks, trigger wait time clocks, and outstanding trigger balances, a set of repurchase intensity results is formed, simultaneously depicting actual consumption rhythm, net trigger rhythm, and the backlog of unmet demand. Compared to existing technologies that rely solely on historical consumption or static planned quantities for prediction, this invention more accurately reflects the formation process of enterprise procurement demand, especially in scenarios involving multiple intertwined factors such as maintenance parts replacement, plan changes, substitution switching, and delivery date fluctuations. It can still stably generate demand intensity representation results consistent with the actual business situation.

[0059] Building upon this foundation, the present invention further generates a set of pre-reserved procurement slots and combines this with sparse integer transport to form a set of slot-level pre-reservation results and a set of material-level net procurement quantity results. Then, through order decoding, batch decoding, and integer quantity auditing, it outputs a set of procurement suggestion results, enabling the prediction results to directly correspond to the review cycle, expected delivery time, maximum capacity, minimum capacity, and minimum order quantity constraints, significantly improving the executability and consistency of procurement suggestions. Simultaneously, the present invention uses mirror replay reconciliation to reverse-engineer the actual procurement execution results, actual delivery results, and actual requisition results, influencing the procurement trigger ledger, consumption waiting clock, trigger waiting clock, and the set of pre-reserved procurement slots. This forms a continuously correcting closed-loop update mechanism, gradually reducing the deviation between suggested quantities and execution results, improving the matching degree of delivery time, the timeliness of demand response, and the dynamic adaptability in multi-batch procurement scenarios. Attached Figure Description

[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0061] Figure 1 This is a flowchart of a method for predicting enterprise procurement demand based on big data analysis proposed in this invention;

[0062] Figure 2 This is a flowchart illustrating the generation of the repurchase intensity result set in a big data-based enterprise procurement demand forecasting method proposed in this invention.

[0063] Figure 3 This is a flowchart of sparse integer transport and result generation for an enterprise procurement demand forecasting method based on big data analysis proposed in this invention. Detailed Implementation

[0064] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0065] refer to Figures 1-3 A method for predicting enterprise procurement demand based on big data analysis includes the following steps:

[0066] Retrieve purchase orders, material requisition records, maintenance work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery date records; perform time alignment and object mapping; and generate a purchase trigger ledger.

[0067] Based on the procurement trigger ledger and historical consumption records, a consumption waiting clock, a trigger waiting clock, and an outstanding trigger balance are constructed to generate a set of renewal intensity results.

[0068] Based on supplier delivery records, review cycles, receiving capacity, minimum order quantity constraints, and delivery compatibility constraints, a set of pre-reserved procurement slots is generated;

[0069] Based on the repurchase intensity result set, the outstanding trigger balance and the purchase pre-occupied slot set, sparse integer transport is performed to generate a slot-level pre-occupancy result set and a material-level net purchase quantity result set;

[0070] Based on the slot-level pre-occupancy result set and the material-level net purchase quantity result set, perform single-order decoding, batch decoding and whole-quantity auditing to generate a purchase suggestion result set;

[0071] Based on the procurement suggestion result set, actual procurement execution result, actual delivery result, and actual requisition result, a mirror replay reconciliation is performed to generate an updated configuration set and write back the procurement trigger ledger, consumption waiting clock, trigger waiting clock, and procurement pre-occupied slot set.

[0072] In this embodiment, the generation of the procurement trigger ledger specifically includes:

[0073] Extract source fields from purchase orders, material requisition records, maintenance work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery date records; perform field normalization, unit of measurement unification, quantity direction normalization, and missing data validation to form the original set of purchase source records.

[0074] The field standardization extracts the following data into unified fields: order placement time, planned arrival time, order quantity, and order status from purchase orders; requisition time, requisition quantity, and requisition direction from material requisition records; work order issuance time, replacement time, and maintenance object identifier from maintenance work orders; planned release time and planned demand quantity from production plans; inventory time, available inventory quantity, and frozen inventory quantity from inventory ledgers; primary and secondary material correspondence and substitution effective period from substitution relationships; and delivery start time and delivery days from supplier delivery records. All quantity fields are converted to a unified unit of measurement. Records that increase procurement pressure are standardized as positive triggers, and records that reduce procurement pressure are standardized as negative offsets, resulting in the original set of procurement source records.

[0075] Perform source time difference correction, object time difference correction and granularity consolidation on the time field in the original procurement source record set to form a unified time index result set;

[0076] The unified time index is obtained by summing the original business time with the source time difference correction and the object time difference correction, and then rounding it down according to the unified time granularity used in the procurement trigger ledger. The source time difference correction is preset according to the data source, and the object time difference correction is obtained according to the same maintenance object, the same production batch, or the same procurement document chain. The granularity rounding uses a down-rounding method to map the corrected business time to the unified time index.

[0077] Calculate object mapping scores based on material identifiers, source object identifiers, supplier identifiers, and substitution relationships in the original procurement source record set, and form an object mapping result set;

[0078] The object mapping score is obtained by multiplying the material code consistency, batch or document chain consistency, source object association consistency, unified time index overlap consistency, and substitution relationship validity by their respective preset weights and then summing them. The mapping relationship is determined by comparing the object mapping score with the preset mapping threshold. The source record with the mapping relationship is encapsulated with the corresponding material identifier, source object identifier, supplier identifier, and substitution relationship into an object mapping result set.

[0079] Among them, material code consistency means that the material code in the source record is the same as the unified material code obtained after field normalization, or the alias code, specification code, and supplier part number in the source record are the same as the unified material code after conversion by the preset mapping table; batch or document chain consistency means that at least one of the batch number, lot number, work order number, purchase order number, material requisition number, and delivery order number in the source record is the same, or the documents before and after can be traced back to the same business chain according to the preset document chain association rules; source object association consistency means that at least one of the maintenance object identifier, production batch identifier, project task identifier, and warehouse location identifier corresponding to the source record is the same, or the object identifiers are different but point to the same source object identifier according to the preset object mapping relationship; unified time index overlap consistency means that there is an overlapping index interval in the unified time index result set after the effective period of the source record is mapped to the unified time index result set after time alignment, or the difference of the unified time index of the records before and after does not exceed the number of preset tolerance indexes; the validity of the substitution relationship is defined as that the main material identifier and the substitute material identifier involved in the source record have a corresponding relationship in the substitution relationship, and the effective period of the substitution covers the unified time index;

[0080] Extract procurement trigger candidate records from the object mapping result set to form a procurement trigger candidate set;

[0081] When extracting candidate records for procurement triggers, the planned demand records pointing to future material consumption in the production plan are extracted as planned demand triggers; the material consumption records that have already occurred in the material requisition records are extracted as actual consumption triggers; the records pointing to replacement parts requirements in the maintenance work orders are extracted as maintenance replacement triggers; the records in the inventory ledger where available inventory is lower than the corresponding requisition requirements are extracted as inventory warning triggers; the records in the substitution relationship where the main substitution switch is effective are extracted as substitution switch triggers; the records in the supplier delivery date records where the delivery date is extended are extracted as delivery date abnormality triggers; and the records in the purchase order where the order is cancelled, reduced, or delayed are extracted as reverse offsetting triggers. For each candidate record for procurement triggers, the trigger type, trigger direction, effective time, duration, and impact intensity are written.

[0082] The impact intensity is obtained by multiplying the quantity value corresponding to the source record by the emergency modulation coefficient determined by the emergency level of the maintenance work order or the emergency level of the production plan, the delivery modulation coefficient determined by the supplier delivery record, and the substitution absorption modulation coefficient determined by the substitution relationship. The duration is taken from the effective time period of the corresponding source record. When the source record does not carry an effective time period, the default duration preset according to the trigger type is used.

[0083] The procurement trigger candidate set is deduplicated and merged, net direction amount synthesized, and source credibility sorted to form a procurement trigger item set;

[0084] The net trigger quantity under the same material identifier, same source object identifier, same trigger type, and same unified time index is obtained by subtracting the cumulative impact intensity of all positive triggers from the cumulative impact intensity of all negative offset triggers within the group. For groups with a non-zero net trigger quantity, purchase trigger entries are retained, and source credibility is written according to the preset source credibility order of purchase order, maintenance work order, production plan, material requisition record, inventory ledger, substitution relationship and supplier delivery date record to form a purchase trigger entry set.

[0085] The procurement trigger entry set is organized into a ledger order according to material identifier, unified time index and source object identifier, and ledger encapsulation is performed to generate the procurement trigger ledger;

[0086] The ledger encapsulation writes each procurement trigger entry with material identifier, source object identifier, supplier identifier, trigger type, trigger direction, effective time, duration, impact intensity, source credibility, and corresponding unified time index. Procurement trigger entries within the same ledger page are sorted in ascending order according to the unified time index and the correspondence between positive triggers and negative offsets is preserved, generating a procurement trigger ledger that can be directly called by subsequent construction of consumption waiting clock, trigger waiting clock, and outstanding trigger balance.

[0087] In this embodiment, the generation of the renewal strength result set specifically includes:

[0088] Based on the material identifier and unified time index, the procurement trigger ledger and historical consumption records are indexed and sorted to form a clock update input set.

[0089] The historical consumption records are purchase trigger entries in the purchase trigger ledger that originate from material requisition records and whose trigger type is consumption trigger. The entry sorting sorts forward triggers, reverse offsets, and historical consumption records under the same material identifier in ascending order according to a unified time index. At each unified time index position, the cumulative value of the positive trigger influence intensity, the cumulative value of the reverse offset influence intensity, the historical consumption quantity, and the aggregated value of the source credibility are written. The aggregated value of the source credibility is obtained by weighting the source credibility of all purchase trigger entries under the unified time index according to the influence intensity, forming a clock update input set.

[0090] Update the consumed waiting clock index by index according to the clock update input set to form a consumed waiting clock result set;

[0091] Extract the accumulated values ​​of positive trigger effect strength and negative cancellation effect strength from the clock update input set, update the trigger wait clock index by index, and form a trigger wait clock result set;

[0092] Update the outstanding trigger balance according to the carry-over relationship of positive triggering, reverse offsetting and historical actual consumption, and form an outstanding trigger balance result set;

[0093] The outstanding trigger balance of materials at the current unified time index is obtained by adding the outstanding trigger balance at the previous unified time index to the cumulative value of the positive trigger influence intensity at the current unified time index, and then subtracting the cumulative value of the negative offset influence intensity and the historical actual consumption quantity at the current unified time index. When the carry-over calculation result is negative, the outstanding trigger balance is truncated to zero. The outstanding trigger balance at the first unified time index is obtained by subtracting the cumulative value of the negative offset influence intensity and the historical actual consumption quantity from the cumulative value of the positive trigger influence intensity at that unified time index, and then performing non-negative truncation, thus forming the set of outstanding trigger balance results.

[0094] Perform dimensional unification and credibility correction on the consumption waiting clock result set, trigger waiting clock result set, outstanding trigger balance result set and source credibility aggregate value to form the input set for repurchase intensity calculation;

[0095] The dimension unification includes: dividing the consumption waiting clock at the current unified time index by the upper limit of the consumption waiting clock of the corresponding material in the preset observation window, and truncating it to one when the quotient is greater than one, to obtain the dimension-unified consumption waiting clock; dividing the trigger waiting clock at the current unified time index by the upper limit of the trigger waiting clock of the corresponding material in the preset observation window, and truncating it to one when the quotient is greater than one, to obtain the dimension-unified trigger waiting clock; dividing the uncleared trigger balance at the current unified time index by the upper limit of the uncleared trigger balance of the corresponding material in the preset observation window, and truncating it to one when the quotient is greater than one, to obtain the dimension-unified uncleared trigger balance. The source credibility correction adopts the source credibility aggregation value at the unified time index.

[0096] Based on the input set for calculating renewal strength, perform renewal strength synthesis to generate a result set of renewal strength.

[0097] The renewal strength is obtained by multiplying the source reliability correction value by the weighted composite result of the outstanding trigger balance, the trigger waiting clock, and the consumption waiting clock. The outstanding trigger balance after unification of dimensions is assigned an outstanding trigger balance weight, the trigger waiting clock after subtracting the unification of dimensions is assigned a trigger waiting clock weight, and the consumption waiting clock after subtracting the unification of dimensions is assigned a consumption waiting clock weight. The outstanding trigger balance weight, the trigger waiting clock weight, and the consumption waiting clock weight are all preset non-negative weights, and the sum of the three is one. When the renewal strength is less than zero, it is truncated to zero; when the renewal strength is greater than one, it is truncated to one. The renewal strength corresponding to all unified time indices under the same material identifier is encapsulated into a renewal strength result set.

[0098] In this embodiment, the consumption waiting clock is updated according to the following rules: when there is a historical consumption quantity at the unified time index, the corresponding consumption waiting clock is set to zero; when there is no historical consumption quantity at the unified time index, the consumption waiting clock at the previous unified time index is incremented by one as the current consumption waiting clock; when there is no historical consumption quantity at the first unified time index, the unified time granularity quantity from the observation start point to the unified time index is written into the consumption waiting clock to form a consumption waiting clock result set.

[0099] In this embodiment, the trigger waiting clock is updated according to the following rules: when the cumulative value of the positive trigger influence intensity minus the cumulative value of the reverse cancellation influence intensity at the current unified time index is still greater than zero, the corresponding trigger waiting clock is set to zero. When the difference is less than or equal to zero, the trigger waiting clock at the previous unified time index is incremented by one as the current trigger waiting clock. When there is no net positive trigger at the first unified time index, the number of unified time granularities from the observation start point to the unified time index is written into the trigger waiting clock to form a trigger waiting clock result set.

[0100] In this embodiment, the generation of the procurement pre-reserved slot set specifically includes:

[0101] Extract material identifiers and supplier identifiers from the procurement trigger ledger, and align them with supplier delivery date records, review cycles, receiving capacity, minimum order quantity constraints, and arrival compatibility constraints to form a procurement pre-reservation slot generation input set;

[0102] The index alignment reads the corresponding page in the procurement trigger ledger according to the material identifier, supplier identifier, and unified time index, extracts the material identifier, supplier identifier, and unified time index, and aligns the delivery start time and delivery days in the supplier delivery record, the review open period and review close period in the review cycle, the receiving period and receiving limit in the receiving capacity, the corresponding minimum order quantity in the minimum order constraint, and the allowed and prohibited delivery periods in the arrival compatibility constraint to the same material identifier and the same supplier identifier, forming the procurement pre-occupied slot generation input set;

[0103] Based on the input set generated from the pre-reserved slots for procurement, the review release period, the delivery start period, and the expected delivery period are determined, forming a candidate set for slot boundaries.

[0104] The review release period is taken from the valid order interval between the review open period and the review close period. The delivery start period is taken from the unified time index position corresponding to the delivery start time in the supplier's delivery record. The expected delivery period is obtained by extending the corresponding delivery days by each unified time index within the review release period. The segmentation is performed according to the boundary between the review close period and the next review open period to obtain a slot boundary candidate set indexed by material identifier, supplier identifier and review cycle.

[0105] Perform arrival compatibility screening on the candidate set of slot boundaries to form an arrival compatibility determination result set;

[0106] The arrival compatibility screening compares the expected arrival time in the slot boundary candidate set with the allowed arrival time and prohibited arrival time in the arrival compatibility constraints. When the expected arrival time falls completely within the allowed arrival time and does not overlap with the prohibited arrival time, the corresponding slot boundary candidate is determined to be valid. When the expected arrival time overlaps with the prohibited arrival time or exceeds the allowed arrival time, the corresponding slot boundary candidate is determined to be invalid. The slot boundary candidates that are determined to be valid are written into the arrival compatibility determination result set.

[0107] Based on the set of arrival compatibility determination results, the receiving capacity is pruned and the minimum order quantity constraint is mapped to form the slot bearing result set;

[0108] The receiving capacity trimming is based on the qualified slot boundary candidate. The receiving upper limit corresponding to all receiving periods covered by the slot boundary candidate is accumulated to obtain the maximum carrying capacity of the slot boundary candidate. The minimum order quantity constraint mapping writes the corresponding minimum order quantity under the same material identifier and the same supplier identifier into the slot boundary candidate to obtain the minimum carrying capacity of the slot boundary candidate. When the maximum carrying capacity is less than the minimum carrying capacity, the slot boundary candidate is marked as invalid. When the maximum carrying capacity is greater than or equal to the minimum carrying capacity, the slot boundary candidate is marked as valid, forming a slot carrying capacity result set.

[0109] The slot identifiers of the valid slot boundary candidates marked in the slot bearing result set are arranged and the fields are encapsulated to form a set of procurement pre-occupied slot entries;

[0110] The slot identifiers are generated in the order of material identifier, supplier identifier, review period, and expected delivery time. The field encapsulation writes each marked valid slot boundary candidate into the slot identifier, material identifier, supplier identifier, review period, review release period, delivery start period, expected delivery time, maximum capacity, minimum capacity, and validity flag, forming a set of pre-occupied slot entries for procurement.

[0111] Create a sequential index for the set of pre-reserved slots for procurement and perform set encapsulation to generate a set of pre-reserved slots for procurement;

[0112] The sequential index arranges the procurement pre-reserved slot entries in ascending order of expected arrival time under the same material identifier, supplier identifier under the same expected arrival time, and slot identifier under the same supplier identifier. It retains the mapping relationship between each procurement pre-reserved slot entry and its corresponding material identifier, supplier identifier, review cycle, maximum capacity, and minimum capacity.

[0113] In this embodiment, the generation of the slot-level pre-occupancy result set and the material-level net purchase quantity result set specifically includes:

[0114] Perform index alignment and object aggregation on the repurchase intensity result set, the outstanding trigger balance result set, and the purchase pre-occupied slot set to form a sparse integer transport input set;

[0115] The index alignment and object aggregation reads the corresponding entries in the renewal strength result set, the outstanding trigger balance result set, and the procurement pre-occupied slot set according to the material identifier, supplier identifier, unified time index, review cycle, and expected delivery time. Entries with renewal strength greater than zero and outstanding trigger balance greater than zero under the same material identifier are aggregated with procurement pre-occupied slot entries whose expected delivery time is later than the corresponding unified time index into the same group of transport inputs. In each group of transport inputs, the material identifier, supplier identifier, unified time index, renewal strength, outstanding trigger balance, slot identifier, expected delivery time, maximum capacity, minimum capacity, and validity flag are written to form a sparse integer transport input set.

[0116] Establish slot-edge relationships around the sparse integer transport input set to form a transport candidate relationship set;

[0117] The slot-edge relationships include direct material matching relationships and substitution coverage relationships. Direct material matching relationships are where the material identifier in the sparse integer transportation input set is the same as the material identifier in the procurement pre-occupied slot set and the supplier identifier is the same. Substitution coverage relationships are where the material identifier in the sparse integer transportation input set and the material identifier in the procurement pre-occupied slot set have a primary-substitution correspondence in the substitution relationship and the substitution effective period covers the corresponding expected delivery period. For each material identifier, only slot-edge relationships with expected delivery periods no earlier than the corresponding unified time index and within the range of the number of pre-set covered slots after being sorted in ascending order by expected delivery period are retained to form a transportation candidate relationship set.

[0118] Perform feasibility screening and priority ranking on the set of candidate transport relationships to form a set of feasible transport relationships;

[0119] The feasibility screening process involves determining whether each item in the procurement pre-occupied slot in the transportation candidate relationship set is an item with a valid flag, and determining whether the corresponding maximum carrying capacity is greater than zero, whether the expected delivery time meets the delivery compatibility constraint, and whether the outstanding trigger balance is greater than zero. Transportation candidate relationships that meet all the conditions are retained as feasible transportation relationships.

[0120] The priority order is arranged in the following order: direct material delivery relationship takes precedence over alternative coverage relationship; earlier expected delivery time takes precedence over later expected delivery time; and within the same expected delivery time, the material with smaller maximum capacity takes precedence over the material with larger maximum capacity, thus forming a set of feasible transportation relationships.

[0121] Based on the set of feasible transport relationships, integer allocation and capacity deduction are performed to form a set of transport allocation results;

[0122] The set of transportation allocation results is merged into slots and materials to form a set of slot-level pre-occupancy results and a set of material-level net purchase quantity results.

[0123] The slot merging process involves summing all integer pre-occupancy quantities in the transportation allocation result set according to the slot identifier to obtain the slot-level pre-occupancy quantity corresponding to each slot identifier. The slot-level pre-occupancy quantity, along with the corresponding slot identifier, material identifier, supplier identifier, expected delivery time, and review cycle, is then encapsulated into a slot-level pre-occupancy result set.

[0124] Material merging distinguishes between direct allocation relationships and substitution coverage relationships based on material identifiers. For direct allocation relationships, the integer pre-allocation amounts allocated to the material's procurement pre-allocation slot entries are summed to obtain the direct allocation pre-allocation amount. For substitution coverage relationships, the integer pre-allocation amounts allocated from the primary material to the substitution material's procurement pre-allocation slot entries are summed to obtain the substitution absorption amount. The direct allocation pre-allocation amount for the same material identifier is subtracted from the substitution absorption amount output to the substitution material, and then the substitution absorption amount input from other primary materials to this material is added to obtain the material-level net procurement quantity result set.

[0125] In this embodiment, integer allocation starts with the outstanding trigger balance under the same material identifier and the same unified time index. Purchase pre-reserved slot entries are read one by one according to the priority order in the feasible transport relationship set. For each purchase pre-reserved slot entry, an integer pre-reserved amount is allocated that is no greater than the current remaining maximum capacity of the purchase pre-reserved slot entry and no greater than the current outstanding trigger balance. When the current outstanding trigger balance is less than the minimum capacity of the corresponding purchase pre-reserved slot entry, the purchase pre-reserved slot entry is skipped and the next purchase pre-reserved slot entry is read. After the integer pre-reserved amount is written, the current remaining maximum capacity of the corresponding purchase pre-reserved slot entry and the current outstanding trigger balance of the corresponding material identifier are deducted simultaneously until the current outstanding trigger balance is zero or the feasible transport relationship set is traversed completely, forming a transport allocation result set.

[0126] In this embodiment, the generation of the procurement suggestion result set specifically includes:

[0127] Perform field alignment and sequence aggregation on the slot-level pre-occupancy result set and the material-level net purchase quantity result set to form a purchase suggestion decoding input set;

[0128] The field alignment and sequence aggregation reads the slot identifier, slot-level pre-occupancy quantity, expected delivery time, and review period from the slot-level pre-occupancy result set according to the material identifier, supplier identifier, review period, and expected delivery time. It also reads the material identifier and corresponding net purchase quantity from the material-level net purchase quantity result set. All slot-level pre-occupancy results under the same material identifier and the same supplier identifier are sorted in ascending order by expected delivery time, and the corresponding net purchase quantity is written into the sorted sequence to form the purchase suggestion decoding input set.

[0129] Based on the procurement recommendations, the decoded input set is divided into single segments and continuous slots are merged to form a single candidate set.

[0130] The order segmentation is based on the slot identifier as the initial order unit. For adjacent slot identifiers under the same material identifier and the same supplier identifier, it is determined whether their review cycle is the same, whether the expected delivery time is continuous, and whether the slot-level pre-occupancy quantity still falls within the capacity range of the corresponding slot-level pre-occupancy quantity after merging. When all conditions are met, the slot-level pre-occupancy quantities corresponding to adjacent slot identifiers are merged into the same order candidate. When the conditions are not met, they remain as independent order candidates. For slot-level pre-occupancy quantities corresponding to the same slot identifier that are greater than the upper limit of the single suggested batch, they are segmented into multiple order candidates in a way that keeps the expected delivery time and slot identifier unchanged, forming an order candidate set.

[0131] Batch decoding is performed around the candidate set to form a batch decoding result set;

[0132] The batch decoding process reads the corresponding slot-level pre-allocation quantity for each order candidate and generates suggested batches in the order of integer priority, minimum order constraint satisfaction priority, and expected delivery time priority.

[0133] Specifically, when the slot-level pre-occupied quantity corresponding to the order candidate is less than the minimum order quantity corresponding to the order candidate, the order candidate is merged with the next order candidate with the same material identifier, the same supplier identifier and the closest expected delivery time, and batch decoding is re-executed. When the slot-level pre-occupied quantity corresponding to the order candidate is greater than the upper limit of a single suggested batch, it is split into multiple suggested batches according to the upper limit of the suggested batch, and the remaining quantity that is less than the upper limit of the suggested batch is retained as the tail batch, forming a set of batch decoding results.

[0134] Perform integer quantity auditing on the batch decoding result set to form an integer quantity audit result set;

[0135] The whole quantity audit verifies whether the cumulative value of all suggested batches under the same slot identifier is equal to the slot-level pre-occupancy quantity corresponding to that slot identifier, whether the cumulative value of all suggested batches under the same material identifier is equal to the material-level net purchase quantity corresponding to that material identifier, whether each suggested batch is a positive integer, whether it meets the minimum order constraint, and whether it does not exceed the capacity limit of the corresponding slot identifier. When any verification fails, the remaining quantity is adjusted between adjacent order candidates under the same material identifier and the same supplier identifier. If the remaining quantity adjustment still fails, the corresponding order candidate is returned to re-execute the order splitting and batch decoding to form a whole quantity audit result set.

[0136] The audit results for whole quantities are written into the suggested order placement period, suggested delivery period, and suggested quantity to form a set of procurement suggestion items;

[0137] The suggested order placement period is taken from the review release period in the review cycle of the corresponding order candidate, the suggested delivery period is taken from the expected delivery period of the corresponding order candidate, and the suggested quantity is taken from the suggested batch that has passed the whole quantity audit. Each suggested batch is written with the corresponding material identifier, supplier identifier, slot identifier, review cycle, suggested order placement period, suggested delivery period and suggested quantity to form a set of procurement suggestion items.

[0138] The procurement suggestion items are sorted in ascending order by suggested order time under the same material identifier, by supplier identifier under the same suggested order time, and by suggested delivery time under the same supplier identifier, and a procurement suggestion result set is generated.

[0139] In this embodiment, the generation of the updated configuration set specifically includes:

[0140] The execution of procurement recommendations results set includes collecting actual procurement execution results, actual delivery results, and actual requisition results. The execution fields are aligned and time periods are aggregated to form a mirror playback input set.

[0141] The field alignment and time period aggregation reads the suggested quantity, actual order quantity and actual order period from the procurement suggestion result set, the actual delivery quantity and actual delivery period from the actual procurement execution result, the actual delivery quantity and actual delivery period from the actual delivery result, and the actual requisition quantity and actual requisition period from the actual requisition result, based on the material identifier, supplier identifier, slot identifier, review cycle, suggested order period, suggested delivery period, and unified time index. The procurement suggestion result, actual procurement execution result, actual delivery result, and actual requisition result corresponding to the same slot identifier under the same material identifier and supplier identifier are aggregated into the same mirror playback unit to form a mirror playback input set.

[0142] The suggested paths and execution paths are reconciled one by one according to the input set of the mirror replay, forming a set of mirror replay difference results;

[0143] The path reconciliation compares the suggested quantity with the actual order quantity, the suggested arrival time with the actual arrival time, the suggested quantity with the actual requisition quantity, and the corresponding slot identifier, material identifier, and supplier identifier in the same mirror replay unit to see if they are consistent. When the suggested quantity is greater than the actual order quantity, it is recorded as an unexecuted difference. When the suggested quantity is less than the actual order quantity, it is recorded as an over-executed difference. When the actual arrival time is later than the suggested arrival time, it is recorded as a delayed arrival difference. When the actual requisition quantity is greater than the actual arrival quantity, it is recorded as a gap difference. When the slot identifier, material identifier, or supplier identifier is inconsistent, it is recorded as a mismatch difference. The difference type, difference quantity, and difference time corresponding to each mirror replay unit are written into the mirror replay difference result set.

[0144] Based on the set of differences in the mirror playback results, the responsible object is attributed and the updated object is located, forming an updated location result set;

[0145] The responsibility object attribution is determined according to the difference type to determine the update object. Among them, the non-executed difference and over-executed difference are located to the corresponding slot identifier in the purchase pre-occupied slot set and the corresponding order candidate in the purchase suggestion result set. The delayed arrival difference is located to the corresponding supplier delivery date record entry in the purchase trigger ledger and the corresponding expected arrival time in the purchase pre-occupied slot set. The gap difference is located to the corresponding purchase trigger entry in the purchase trigger ledger and the unified time index corresponding to the consumption waiting clock and the trigger waiting clock. The mismatch difference is located to the corresponding entries of the source object identifier, material identifier and supplier identifier in the purchase trigger ledger. The responsibility object, update object and unified time index corresponding to each difference type are written into the update positioning result set.

[0146] Based on the updated location result set, perform parameter correction and value write-back preparation to form an updated configuration entry set;

[0147] The parameter correction and value write-back preparation includes: for update objects located to the procurement trigger ledger, correcting the duration, impact intensity, and source credibility of the corresponding procurement trigger entry; for update objects located to the consumption waiting clock, resetting or extending the consumption waiting clock at the corresponding unified time index according to the actual requisition period; for update objects located to the trigger waiting clock, resetting or extending the trigger waiting clock at the corresponding unified time index according to whether the net positive trigger is established; and for update objects located to the procurement pre-occupied slot set, correcting the expected arrival period, maximum capacity, minimum capacity, and validity flag of the corresponding slot identifier.

[0148] Conflict resolution and sequential arrangement are performed on the set of updated configuration entries to form an updated configuration set;

[0149] The conflict resolution process handles situations where there are multiple update configuration entries under the same material identifier, supplier identifier, tank identifier, or unified time index. When multiple update configuration entries act on the same update object, the update configuration entries triggered by mismatch differences are retained first, followed by those triggered by delayed arrival differences, then those triggered by gap differences, and finally those triggered by unexecuted differences and over-executed differences.

[0150] The sequential arrangement is based on ascending order of the unified effective time index, ascending order of material identifiers under the same unified effective time index, and the order of update object types under the same material identifier. The update object types are arranged in the following order: purchase trigger ledger, consumption waiting clock, trigger waiting clock, and purchase pre-occupied slot set, forming the update configuration set.

[0151] The results are written back to the procurement trigger ledger, consumption waiting clock, trigger waiting clock, and procurement pre-occupied slot set in the order of the updated configuration set, completing the mirror replay reconciliation loop.

[0152] Example 1: To verify the feasibility of this invention in practice, it was applied to the spare parts and auxiliary material procurement of a high-end equipment manufacturing enterprise. This enterprise simultaneously undertakes continuous production, equipment maintenance, and multi-category material turnover tasks. Daily procurement includes both routine consumables and low-frequency materials temporarily triggered by maintenance replacements, substitutions, and delivery date fluctuations. The original approach primarily relied on historical requisition data and manual planning for procurement decisions. While this could generate a basic procurement list, it was often difficult to identify the true source of demand in a timely manner when faced with temporary maintenance work orders, inventory fluctuations, changes in supplier delivery dates, and changes in substitution relationships. This easily led to problems such as initial assessments of tight supply followed by concentrated replenishment orders, or mismatches between suggested order placement periods and actual delivery schedules, thus affecting requisition coordination and warehousing arrangements.

[0153] In practical applications, first, continuously collect purchase orders, material requisition records, repair work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery date records from the enterprise's existing business platform, and complete time alignment according to a unified time granularity. Then, through object mapping, merge records from different sources but pointing to the same material, the same repair object, or the same supply path into the same ledger page to form a purchase trigger ledger. Subsequently, jointly process the historical actual consumption records formed by material requisitions and the purchase trigger ledger, and continuously update the consumption waiting clock, trigger waiting clock, and outstanding trigger balance, so that the purchase judgment no longer simply depends on the inventory at a certain point or a single material requisition change, but can simultaneously reflect the actual consumption interval, the rhythm of net trigger accumulation, and the demand pressure that has not been consumed by purchases or material requisitions. On this basis, continue to read supplier delivery dates, review arrangements, receiving capabilities, minimum order quantity constraints, and arrival compatibility constraints, generate a set of purchase pre-occupation slots, express in advance the order placement window and arrival window for future purchasable demands, and then perform sparse integer transportation based on the result set of reorder intensity and the outstanding trigger balance, allocate scattered demands to different purchase pre-occupation slots, and further form a set of pre-occupation results at the slot level and a set of net purchase quantity results at the material level. After the allocation is completed, combine order splitting, batch decoding, and integer quantity auditing to output a set of purchase recommendation results, so that the purchase recommendations can directly correspond to the review open interval, recommended arrival interval, and actual bearing boundary.

[0154] The data verification time in this embodiment covers a continuous production and maintenance intertwined cycle of the enterprise, and the location is at the procurement and warehousing coordination link in the main plant area of the enterprise. The collected data includes multiple rounds of purchase recommendation results, actual purchase execution results, actual arrival results, and actual material requisition results. By continuously reconciling the recommended path and the execution path, unexecuted differences, over-execution differences, delayed arrival differences, gap differences, and mismatch differences can be clearly identified, and these differences are further written back to the purchase trigger ledger, consumption waiting clock, trigger waiting clock, and the set of purchase pre-occupation slots. Combining the on-site application process, it can be seen that the present invention can unify and gather the trigger information scattered in the procurement, warehousing, repair, and planning links in the past onto a continuous data chain, so that the purchase recommendations no longer stay at the static estimation level, but are synchronized with the actual material requisition rhythm, delivery date changes, and substitution relationships. During the application period, when facing repair part replacement triggers, substitution switch triggers, and delivery date exception triggers, procurement personnel can identify the direction of real demand backlog earlier, and warehousing personnel can also arrange the receiving interval in advance according to the purchase pre-occupation slots. Therefore, in terms of business, it is reflected that the recommendation generation is closer to the on-site rhythm, the arrival arrangement is more likely to fall into the acceptable interval, and the correction of the front-end state object after purchase execution is more timely, finally verifying the feasibility and continuous optimization ability of the present invention in complex enterprise procurement scenarios.

[0155] Table 1 Comprehensive Performance Comparison Table of Enterprise Purchase Demand Forecasting Methods

[0156] Indicator Name ARIMA algorithm XGBoost algorithm Croston Intermittent Demand Forecasting Algorithm Method of the present invention Matching rate between procurement recommendations and actual usage 68.9% 74.6% 79.8% 85.1% Renewal strength recognition accuracy 70.4% 77.2% 81.1% 88.3% Expected delivery time hit rate 65.8% 71.5% 76.9% 82.6% Recommended order placement time hit rate 63.7% 69.8% 74.2% 80.4% Material-level net purchase quantity deviation rate 22.1% 17.6% 13.9% 10.8% Emergency order additions 31 24 19 15 Number of material shortage incidents 21 16 13 11 Recommended batch secondary adjustment rate 31.4% 24.7% 18.9% 13.8% Minimum order quantity constraint conflict number 26 18 12 9 Slot bearing capacity exceeding limit times 19 13 10 7 Average inventory turnover days 49.6 days 44.8 days 41.7 days 39.8 days

[0157] As shown in Table 1, the method of this invention outperforms the ARIMA algorithm, XGBoost algorithm, and Croston intermittent demand forecasting algorithm in all indicators, and the improvement is reasonably reasonable in terms of business operations. The ARIMA algorithm performs the worst in terms of matching rate between procurement suggestions and actual usage, hit rate of expected delivery time, and hit rate of suggested order time. This indicates that when relying solely on the continuous time series variation pattern for fitting, it is difficult to effectively handle non-stationary and non-continuous procurement demand sources such as maintenance work order triggers, substitution switching triggers, and delivery date anomaly triggers. The XGBoost algorithm improves the accuracy of repurchase intensity identification and matching rate between procurement suggestions and actual usage compared to the ARIMA algorithm, indicating that it has a stronger ability to fit nonlinear features from multiple sources. However, since it is essentially still based on feature regression and lacks a continuous state characterization of the demand formation process, its improvement in suggested order time hit rate, expected delivery time hit rate, and number of slot capacity overloads is limited.

[0158] The Croston intermittent demand forecasting algorithm significantly outperforms the ARIMA and XGBoost algorithms in terms of material-level net purchase quantity deviation rate, number of emergency order additions, and number of material shortage events, indicating its better adaptability to sparse, intermittent demand forecasting. This result is consistent with the Croston intermittent demand forecasting algorithm's technical characteristics of handling demand intervals and demand quantities separately for discontinuous demand sequences, thus demonstrating good stability in low-frequency purchase parts, spare parts, and maintenance parts scenarios. However, as shown in Table 1, the Croston intermittent demand forecasting algorithm still lags behind the method of this invention in terms of expected arrival time hit rate, suggested order time hit rate, number of minimum order quantity constraint conflicts, and number of slot capacity overruns. This indicates that although it can identify intermittent demand itself well, it cannot further handle the execution constraints formed by review cycles, supplier delivery dates, receiving capacity, arrival compatibility constraints, and substitution relationships in enterprise procurement scenarios. Therefore, its output results remain more at the demand quantity forecasting level and are difficult to directly transform into suggested results that are highly matched with the procurement execution process.

[0159] The reason why the method of this invention achieves better results in the matching rate between procurement recommendations and actual requisition, the accuracy of repurchase intensity identification, and the deviation rate of net procurement quantity at the material level is that it does not directly fit historical demand sequences. Instead, it first generates a procurement trigger ledger, and then constructs a consumption waiting clock, a trigger waiting clock, and an outstanding trigger balance. This allows it to simultaneously express the material consumption rhythm, the procurement trigger rhythm, and the backlog of undigested demand. Compared with the ARIMA algorithm, this method does not rely on the stationarity of a single time series; compared with the XGBoost algorithm, this method is not a static feature stacking, but transforms the demand formation process into a continuously updatable state chain; compared with the Croston intermittent demand forecasting algorithm, this method not only focuses on the quantity and interval of intermittent demand, but also further incorporates different trigger sources and their continuous impact into a unified modeling framework, thus exhibiting higher consistency in repurchase intensity identification.

[0160] At the execution matching level, the method of this invention integrates supplier delivery records, review cycles, receiving capacity, minimum order quantity constraints, and arrival compatibility constraints into a unified procurement pre-reserved slot set. Then, through sparse integer transport, it generates a slot-level pre-reservation result set and a material-level net procurement quantity result set. Therefore, the expected arrival time hit rate reaches 82.6%, and the suggested order time hit rate reaches 80.4%. Simultaneously, it controls the number of minimum order quantity constraint conflicts to 9 times and the number of slot capacity overruns to 7 times. In contrast, while the ARIMA and XGBoost algorithms can provide quantity predictions, they lack an explicit future slot capacity structure, leading to potential misalignments between order suggestions and receiving capacity. The Croston intermittent demand forecasting algorithm, although more sensitive to intermittent demand, lacks sufficient coupling processing between the review window and the arrival window, thus its accuracy at the execution level is still inferior to the method of this invention. Therefore, the improvement of this invention is not only reflected in the forecasting end but also in the efficiency and consistency of the conversion from forecast results to executable procurement suggestions.

[0161] From a business perspective, the method of this invention reduced the number of emergency order additions to 15, the number of material shortage events to 11, and the average inventory turnover days to 39.8 days. This indicates that it did not reduce material shortages by simply increasing the purchase volume, but rather improved inventory utilization efficiency while enhancing the supply-demand matching. The ARIMA and XGBoost algorithms showed higher numbers of emergency order additions and material shortage events, reflecting the passive replenishment problem caused by the lag in demand identification. Although the Croston intermittent demand forecasting algorithm has significantly improved this problem, it still lacks further constraints on arrival windows, capacity limits, and substitution coverage. Therefore, there is still room for optimization in terms of inventory turnover and execution consistency. This invention continuously writes the actual purchase execution results, actual arrival results, and actual requisition results back to the purchase trigger ledger, consumption waiting clock, trigger waiting clock, and purchase pre-occupied slot set through mirror replay reconciliation. This allows subsequent forecasts and suggestions to continuously correct the preceding state objects, thereby maintaining better adaptability and stability in continuous operation.

[0162] As shown in Table 1, compared with the ARIMA algorithm, XGBoost algorithm, and Croston intermittent demand forecasting algorithm, the method of this invention not only performs better in terms of procurement demand forecasting accuracy, but also forms a more complete technical chain in terms of matching suggested order placement time, matching expected delivery time, satisfying minimum order quantity constraints, slot capacity control, and post-execution closed-loop correction. Therefore, it can simultaneously improve forecasting accuracy, procurement execution, and continuous optimization capabilities, and ultimately demonstrate more prominent comprehensive advantages in reducing emergency orders, reducing material shortage risks, and improving inventory turnover efficiency.

[0163] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting enterprise procurement demand based on big data analysis, characterized in that, Includes the following steps: Retrieve purchase orders, material requisition records, maintenance work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery date records; perform time alignment and object mapping; and generate a purchase trigger ledger. Based on the procurement trigger ledger and historical consumption records, a consumption waiting clock, a trigger waiting clock, and an outstanding trigger balance are constructed to generate a set of renewal intensity results. Based on supplier delivery records, review cycles, receiving capacity, minimum order quantity constraints, and delivery compatibility constraints, a set of pre-reserved procurement slots is generated; Based on the repurchase intensity result set, the outstanding trigger balance and the purchase pre-occupied slot set, sparse integer transport is performed to generate a slot-level pre-occupancy result set and a material-level net purchase quantity result set; Based on the slot-level pre-occupancy result set and the material-level net purchase quantity result set, perform single-order decoding, batch decoding and whole-quantity auditing to generate a purchase suggestion result set; Based on the procurement suggestion result set, actual procurement execution result, actual delivery result, and actual requisition result, a mirror replay reconciliation is performed to generate an updated configuration set and write back the procurement trigger ledger, consumption waiting clock, trigger waiting clock, and procurement pre-occupied slot set.

2. The enterprise procurement demand forecasting method based on big data analysis according to claim 1, characterized in that, The generation of the procurement-triggered ledger specifically includes: Extract source fields from purchase orders, material requisition records, maintenance work orders, production plans, inventory ledgers, substitution relationships, and supplier delivery date records; perform field normalization, unit of measurement unification, quantity direction normalization, and missing data validation to form the original set of purchase source records. Perform source time difference correction, object time difference correction and granularity consolidation on the time field in the original procurement source record set to form a unified time index result set; Calculate object mapping scores based on material identifiers, source object identifiers, supplier identifiers, and substitution relationships in the original procurement source record set, and form an object mapping result set; Extract procurement trigger candidate records from the object mapping result set to form a procurement trigger candidate set; The procurement trigger candidate set is deduplicated and merged, net direction amount synthesized, and source credibility sorted to form a procurement trigger item set; The procurement trigger entry set is organized into a ledger order based on material identifier, unified time index, and source object identifier, and ledger encapsulation is performed to generate the procurement trigger ledger.

3. The enterprise procurement demand forecasting method based on big data analysis according to claim 1, characterized in that, The generation of the renewal strength result set specifically includes: Based on the material identifier and unified time index, the procurement trigger ledger and historical consumption records are indexed and sorted to form a clock update input set. Update the consumed waiting clock index by index according to the clock update input set to form a consumed waiting clock result set; Extract the accumulated values ​​of positive trigger effect strength and negative cancellation effect strength from the clock update input set, update the trigger wait clock index by index, and form a trigger wait clock result set; Update the outstanding trigger balance according to the carry-over relationship of positive triggering, reverse offsetting and historical actual consumption, and form an outstanding trigger balance result set; Perform dimensional unification and credibility correction on the consumption waiting clock result set, trigger waiting clock result set, outstanding trigger balance result set and source credibility aggregate value to form the input set for repurchase intensity calculation; Based on the input set for calculating renewal strength, perform renewal strength synthesis to generate a result set of renewal strength.

4. The enterprise procurement demand forecasting method based on big data analysis according to claim 3, characterized in that, The consumption waiting clock is updated according to the following rules: when there is a historical consumption quantity at the unified time index, the corresponding consumption waiting clock is set to zero; when there is no historical consumption quantity at the unified time index, the consumption waiting clock at the previous unified time index is incremented by one as the current consumption waiting clock; when there is no historical consumption quantity at the first unified time index, the unified time granularity quantity from the observation start point to the unified time index is written into the consumption waiting clock to form a consumption waiting clock result set.

5. The enterprise procurement demand forecasting method based on big data analysis according to claim 3, characterized in that, The trigger waiting clock is updated according to the following rules: when the cumulative value of the positive trigger influence intensity minus the cumulative value of the reverse cancellation influence intensity at the current unified time index is still greater than zero, the corresponding trigger waiting clock is set to zero. When the difference is less than or equal to zero, the trigger waiting clock at the previous unified time index is incremented by one as the current trigger waiting clock. When there is no net positive trigger at the first unified time index, the number of unified time granularities from the observation start point to the unified time index is written into the trigger waiting clock to form a trigger waiting clock result set.

6. The enterprise procurement demand forecasting method based on big data analysis according to claim 1, characterized in that, The generation of the procurement pre-reserved slot set specifically includes: Extract material identifiers and supplier identifiers from the procurement trigger ledger, and align them with supplier delivery date records, review cycles, receiving capacity, minimum order quantity constraints, and arrival compatibility constraints to form a procurement pre-reservation slot generation input set; Based on the input set generated from the pre-reserved slots for procurement, the review release period, the delivery start period, and the expected delivery period are determined, forming a candidate set for slot boundaries. Perform arrival compatibility screening on the candidate set of slot boundaries to form an arrival compatibility determination result set; Based on the set of arrival compatibility determination results, the receiving capacity is pruned and the minimum order quantity constraint is mapped to form the slot bearing result set; The slot identifiers of the valid slot boundary candidates marked in the slot bearing result set are arranged and the fields are encapsulated to form a set of procurement pre-occupied slot entries; Create a sequential index for the set of pre-reserved slot entries and perform set encapsulation to generate a set of pre-reserved slots.

7. The enterprise procurement demand forecasting method based on big data analysis according to claim 1, characterized in that, The generation of the slot-level pre-occupancy result set and the material-level net purchase quantity result set specifically includes: Perform index alignment and object aggregation on the repurchase intensity result set, the outstanding trigger balance result set, and the purchase pre-occupied slot set to form a sparse integer transport input set; Establish slot-edge relationships around the sparse integer transport input set to form a transport candidate relationship set; Perform feasibility screening and priority ranking on the set of candidate transport relationships to form a set of feasible transport relationships; Based on the set of feasible transport relationships, integer allocation and capacity deduction are performed to form a set of transport allocation results; The set of transportation allocation results is merged into slots and materials to form a set of slot-level pre-occupancy results and a set of material-level net purchase quantity results.

8. The enterprise procurement demand forecasting method based on big data analysis according to claim 7, characterized in that, The integer allocation starts with the outstanding trigger balance under the same material identifier and the same unified time index. Purchase pre-reserved slot entries are read one by one according to the priority order in the feasible transport relationship set. For each purchase pre-reserved slot entry, an integer pre-reserved amount is allocated that is no greater than the current remaining maximum capacity of the purchase pre-reserved slot entry and no greater than the current outstanding trigger balance. When the current outstanding trigger balance is less than the minimum capacity of the corresponding purchase pre-reserved slot entry, the purchase pre-reserved slot entry is skipped and the next purchase pre-reserved slot entry is read. After the integer pre-reserved amount is written, the current remaining maximum capacity of the corresponding purchase pre-reserved slot entry and the current outstanding trigger balance of the corresponding material identifier are deducted simultaneously until the current outstanding trigger balance is zero or the feasible transport relationship set is traversed completely, forming a transport allocation result set.

9. The enterprise procurement demand forecasting method based on big data analysis according to claim 1, characterized in that, The generation of the procurement suggestion result set specifically includes: Perform field alignment and sequence aggregation on the slot-level pre-occupancy result set and the material-level net purchase quantity result set to form a purchase suggestion decoding input set; Based on the procurement recommendations, the decoded input set is divided into single segments and continuous slots are merged to form a single candidate set. Batch decoding is performed around the candidate set to form a batch decoding result set; Perform integer quantity auditing on the batch decoding result set to form an integer quantity audit result set; The audit results for whole quantities are written into the suggested order placement period, suggested delivery period, and suggested quantity to form a set of procurement suggestion items; The procurement suggestion items are sorted in ascending order by suggested order time under the same material identifier, by supplier identifier under the same suggested order time, and by suggested delivery time under the same supplier identifier, and a procurement suggestion result set is generated.

10. The enterprise procurement demand forecasting method based on big data analysis according to claim 1, characterized in that, The generation of the updated configuration set specifically includes: The execution of procurement recommendations results set includes collecting actual procurement execution results, actual delivery results, and actual requisition results. The execution fields are aligned and time periods are aggregated to form a mirror playback input set. The suggested paths and execution paths are reconciled one by one according to the input set of the mirror replay, forming a set of mirror replay difference results; Based on the set of differences in the mirror playback results, the responsible object is attributed and the updated object is located, forming an updated location result set; Based on the updated location result set, perform parameter correction and value write-back preparation to form an updated configuration entry set; Conflict resolution and sequential arrangement are performed on the set of updated configuration entries to form an updated configuration set; The results are written back to the procurement trigger ledger, consumption waiting clock, trigger waiting clock, and procurement pre-occupied slot set in the order of the updated configuration set, completing the mirror replay reconciliation loop.

Images