Purchase data automatic real-time response method for market demand fluctuation

Abstract

The invention relates to the field of purchase data automatic real-time response, in particular to a purchase data automatic real-time response method for market demand fluctuation, which comprises the following steps: constructing a multi-source data acquisition module to acquire market demand, industry dynamics, inventory and historical purchase data in real time; carrying out cleaning, abnormal value elimination and normalization processing on the data; calculating a demand fluctuation coefficient based on a weighted linear regression fitting formula, and dividing fluctuation grades according to a preset threshold value; according to the fluctuation level and the comprehensive supply and demand difference, automatically matching a purchase quantity and a purchase period adjustment strategy; the fluctuation model precision and the purchasing strategy effect are evaluated through a dual verification mechanism, and model parameters and a rule base are dynamically optimized based on feedback data to form a closed-loop adaptive optimization process. Through adoption of the technical scheme, high-sensitivity perception, accurate quantification and intelligent response to market fluctuation can be realized, and timeliness and accuracy of purchase decision and system self-evolution capability are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of automated real-time response to procurement data, and more specifically to a method for automated real-time response to market demand fluctuations in procurement data. Background Technology

[0002] With the deepening development of the digital economy, enterprises are increasingly demanding higher market responsiveness in their supply chain management. Procurement, as a crucial link connecting market demand and internal operations, directly impacts inventory optimization, capital utilization, and overall competitiveness through its decision-making efficiency. In today's highly uncertain business environment, market demand is affected by multiple variables, including policy guidance, shifts in consumer preferences, seasonal factors, and unforeseen events, exhibiting high-frequency, non-linear, and unpredictable fluctuations. Traditional procurement systems generally rely on manual experience to periodically review and analyze historical sales data, making it difficult to build a dynamic response mechanism for the future. This leaves enterprises passive when facing sudden surges or drops in demand.

[0003] While automated procurement decision-making methods have gradually incorporated data analysis tools, existing solutions still have significant limitations. Current technologies mostly rely on static thresholds or simple moving average models to assess demand changes, failing to effectively integrate multi-source, heterogeneous real-time market signals (such as user search activity, competitor sales, and raw material price fluctuations), and lacking the ability to quantitatively represent the intensity of fluctuations. Furthermore, the procurement strategy generation process is often disconnected from inventory status, historical deviations, and industry characteristics; the rule base is fixed and cannot be adaptively adjusted, making it difficult to achieve coordinated optimization of procurement volume and cycle time. More critically, current systems generally lack closed-loop verification mechanisms for the effectiveness of analytical models and strategies, potentially leading to inventory buildup or supply shortages after strategy implementation, further weakening the robustness and reliability of the procurement system.

[0004] Therefore, there is an urgent need for an automated response method that can deeply integrate multi-dimensional real-time data, accurately quantify the level of demand fluctuations, and automatically generate and dynamically calibrate procurement strategies to address the fundamental deficiencies of existing procurement models in terms of timeliness, accuracy, and adaptability. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an automated real-time response method for procurement data to market demand fluctuations. By constructing a multi-source data real-time acquisition mechanism, it achieves high-frequency and comprehensive perception of market demand signals, fundamentally solving the information lag problem of traditional procurement models.

[0006] To achieve the above objectives, the present invention provides a method for automated real-time response to market demand fluctuations in procurement data, comprising: Step 1: Real-time data acquisition. A multi-source data acquisition module is built to acquire market demand data, industry dynamics data, enterprise inventory data, and historical procurement data in real time from external platforms and internal systems through compliant data acquisition methods. Market demand data includes at least product sales volume, order volume, user search popularity, and competitor sales data. Industry dynamics data includes at least policy adjustment information and raw material price fluctuation data. The data acquisition frequency is set to a high-frequency acquisition cycle based on industry characteristics and demand sensitivity, with a acquisition cycle ranging from once every 5 minutes to once every 2 hours, ensuring timely capture of market trends and guaranteeing the timeliness and completeness of the input data. Step 2: Data preprocessing. The collected multi-source heterogeneous data is cleaned and standardized to eliminate noise interference and unify data dimensions. During the cleaning process, an outlier identification algorithm based on statistical distribution characteristics is used to remove data points that deviate from the normal range, while duplicate records are removed. Missing data are filled in using time series interpolation or historical mean based on similar periods. The standardization process uses the min-max normalization algorithm to transform data of different dimensions and orders of magnitude to a unified numerical range of 0.0 to 1.0, forming a structured standardized dataset, which provides a high-quality input foundation for subsequent analysis. Step 3: Market demand fluctuation analysis. A fluctuation analysis model is constructed. Based on the preprocessed standardized dataset, a weighted linear regression fitting formula is used to quantify the degree of change in current market demand relative to historical benchmarks, outputting a continuous demand fluctuation coefficient F(x). This fitting formula is defined as follows: Where α is the current demand weighting coefficient, ranging from 0.6 to 0.8, and dynamically adjusted according to the industry's demand response speed; w i The weighting coefficients for the i-th type of market demand data are given, with the sum of all weights equal to 1. Core data such as sales volume and order volume are assigned higher weights; D i,t β is the standardized value of the i-th type of data in the current collection period t; β is the historical demand benchmark weighting coefficient, ranging from 0.2 to 0.4; D t The standardized value of historical average demand over the past m collection periods (m≥30) is used; γ is the fluctuation correction coefficient, determined by statistical regression of the deviation between the enterprise's past procurement execution results and actual demand; five preset fluctuation threshold ranges are set, namely [-∞,-0.3), [-0.3,-0.1), [-0.1,0.1], (0.1,0.3], and (0.3,+∞), corresponding to fluctuation levels of significant decrease, slight decrease, stable fluctuation, slight increase, and significant increase, respectively; the fluctuation level of the current market demand is determined by judging the range in which F(x) falls. Step 4: Procurement strategies are automatically generated, and a procurement response rule base is established. This rule base stores the mapping logic between different fluctuation levels and procurement adjustment strategies. Based on the fluctuation level output by the fluctuation analysis and the specific value of F(x), a fitting formula for the procurement quantity is used. The target purchase quantity is automatically calculated; where P is the target purchase quantity; k is the strategy adjustment coefficient, whose value is dynamically set according to the fluctuation level. During significant upward fluctuations, k > 1; during minor upward fluctuations, k is slightly greater than 1; during stable fluctuations, k = 1; and during downward fluctuations, k < 1. t The standardized value of the current cycle's comprehensive demand is a weighted composite of multiple types of market demand data; S t The current inventory level is a standardized value. The procurement adjustment strategy includes two parts: procurement quantity adjustment logic and procurement cycle adjustment logic. For significant upward fluctuations, the procurement quantity increment adjustment logic is matched, and the procurement cycle is shortened to 60% to 80% of the original cycle. For slight upward fluctuations, the moderate increment logic is matched, and the procurement cycle remains unchanged. For stable fluctuations, the regular replenishment logic is executed, and the procurement cycle remains at the baseline cycle. For slight downward fluctuations, the quantity reduction logic is matched, and the procurement cycle is extended to 1.2 to 1.5 times the original cycle. For significant downward fluctuations, the significant quantity reduction logic is matched, and the procurement cycle is extended to more than 1.5 times the original cycle, and a non-core category procurement suspension judgment process is triggered, retaining only the necessary category procurement tasks. Step 5: Model Fitting and Strategy Validation. A dual validation mechanism is set up to evaluate the effectiveness of the volatility analysis model and the generated procurement strategy. First, the model fitting is validated by constructing a validation set using n sets of known historical actual demand data and corresponding actual procurement behaviors. The data from each period in the validation set are input into the volatility analysis model, and the mean square error (MSE) between the predicted volatility coefficient Fpred(x) and the actual volatility coefficient Ftrue(x) is calculated. MSE = (1 / n) × Σ(Fpred(xi) - Ftrue(xi)) 2 If the MSE exceeds the preset tolerance threshold, the parameter optimization process is initiated, adjusting α, β, and w in the fitting formula. i And the γ parameter until the MSE meets the accuracy requirements; then perform procurement strategy verification, simulate the generated procurement strategy based on the same verification set, calculate the inventory turnover rate, stockout rate and backlog rate after simulated procurement, and compare them with the enterprise's preset target threshold; if any key indicator fails to meet the target, return to the corresponding k value, periodic adjustment range or pause logic trigger condition in the adjustment rule base to complete the strategy parameter calibration. Step 6: Real-time execution and feedback optimization. The procurement strategy, validated through dual verification, is pushed to the enterprise procurement management system via the system interface, driving the automated execution of the procurement process. Simultaneously, a closed-loop feedback mechanism is constructed to continuously collect key performance indicators (KPIs) during procurement execution, including but not limited to order completion rate, order response time, actual delivery time, and warehousing accuracy. Dynamic inventory change data is also collected synchronously to reflect the actual inventory level changes after strategy execution. This feedback data is then sent back to the fluctuation analysis model and procurement response rule base, re-invoking the verification logic from Step 5 to re-evaluate the effectiveness of the model and strategy. If performance degradation or environmental deviation is detected, a new round of parameter iteration updates is automatically triggered, dynamically adjusting the fluctuation threshold range boundaries, the fitting formula parameter set, and the strategy mapping parameters in the rule base. This enables continuous learning and adaptive evolution of the entire response mechanism, improving long-term operational stability.

[0007] Preferably, in step 1, the multi-source data acquisition module is configured with an authentication mechanism and a data encryption transmission protocol to ensure that data security standards are met when acquiring e-commerce platform sales data and search engine user search popularity data through API interfaces; the data crawler component adopts a distributed architecture deployment, supports concurrent access to multiple data sources, and a single acquisition task can complete the retrieval of all data within 90 seconds, avoiding information distortion due to delays; the acquisition frequency is set with a differentiated strategy based on product category, with fast-moving consumer goods being acquired once every 10 minutes and durable goods being acquired once every 2 hours, ensuring high-frequency response while taking into account system resource consumption.

[0008] Preferably, in step 2, the outlier identification algorithm uses a modified interquartile range method combined with a sliding window standard deviation detection as a dual criterion. When a data point exceeds the range of Q1-3×IQR or Q3+3×IQR, and its deviation from the mean of the previous 5 periods exceeds 2 times the sliding standard deviation, it is judged as an outlier and removed. The data completion algorithm selects different strategies according to the data type. For demand data with obvious periodicity, seasonal time series interpolation is used, and for non-periodic auxiliary data, K-nearest neighbor interpolation is used to ensure that the completion result is close to the true trend. A dynamic extreme value update mechanism is introduced in the normalization process. The maximum and minimum values ​​are recalculated every 7 days based on the latest data to prevent the new data from being compressed to the extreme range due to long-term use of fixed extreme values.

[0009] Preferably, in step 3, the weight w iThe initial values ​​are set based on expert experience, but an online learning mechanism is introduced during system operation. The weight of various data on the final procurement performance is adjusted in reverse according to the results of each verification. For example, when the search popularity leads the actual sales change by more than 7 days and the correlation coefficient is higher than 0.8, its weight ratio is gradually increased. The fluctuation correction coefficient γ is obtained by linear regression of the cumulative deviation between the enterprise's planned procurement volume and actual consumption volume in the past 12 months. The initial value is set to 0.05 and dynamically corrected with quarterly updates. The upper and lower limits of the fluctuation threshold range can be manually fine-tuned by the enterprise according to its business objectives, or an adaptive mode can be enabled, in which the system automatically divides the boundaries based on the density clustering results of historical fluctuation distribution.

[0010] Preferably, in step 4, the specific value of the strategy adjustment coefficient k follows piecewise function logic: when in a period of significant upward fluctuation, k = 1 + 0.5 × (F(x) - 0.3), with an upper limit of 1.8; when in a period of slight upward fluctuation, k = 1 + 0.3 × F(x), with an upper limit of 1.3; when in a period of downward fluctuation, k = 1 - 0.4 × |F(x)|, with a lower limit of 0.5; the comprehensive demand standardized value D t Use the same weights w as in volatility analysis i Perform weighted synthesis to ensure logical consistency; standardized inventory value S t It is obtained by dividing the current physical inventory by the average demand over the last 30 periods, and then normalized and mapped to the range of 0.0 to 1.0.

[0011] Preferably, in step 5, the validation set is constructed using a rolling window approach, including historical data from the most recent 180 consecutive days, with each day serving as an independent validation sample; the preset tolerance threshold for MSE is set to 0.02, and if the validation results exceed the threshold for three consecutive times, global parameter retraining is forcibly initiated; the procurement strategy simulation uses a discrete event simulation engine to simulate the inventory inflow and outflow process over the next three procurement cycles, comprehensively evaluating the impact of the strategy on future inventory health; during the simulation, the uncertainty of supplier delivery cycles is considered, and a ±15% time disturbance factor is introduced to enhance robustness testing.

[0012] Preferably, in step 6, the procurement strategy push interface uses message queue middleware to achieve asynchronous communication, ensuring that instructions are not lost even if the downstream system is temporarily unavailable; the feedback data acquisition module is equipped with a timed task scheduler, which automatically extracts order execution data from the ERP system and inventory change logs from the WMS system every hour; the parameter iteration update process adopts a canary release mechanism, first verifying the effectiveness of the new parameter combination in a simulation environment, and then gradually applying it to the actual business flow to prevent sudden changes from causing system oscillations; the entire closed-loop feedback cycle is controlled to complete a full iteration within 24 hours, achieving weekly self-evolution capability.

[0013] Preferably, it also includes a procurement suspension judgment logic for non-core categories. This logic is activated under conditions of significant decline and volatility. First, it determines whether the average demand growth rate of the category in the past 30 days is lower than -15%. Second, it assesses whether its inventory turnover days exceed the company's set safety limit of 1.5 times. Finally, it combines the financial indicator that the gross profit margin is lower than the overall average of 70%. When all three conditions are met, a procurement freeze suggestion is generated and pushed to the management approval channel to achieve intelligent control of risky categories.

[0014] Preferably, the method is integrated into an enterprise supply chain intelligent decision-making platform, supporting complex architecture scenarios involving multiple organizations, warehouses, and suppliers; the platform has a built-in visual monitoring panel that displays the fluctuation coefficient change curve, procurement strategy execution status, inventory warning signals, and model verification reports in real time; the system has a hierarchical access control function, allowing users at different levels to view data and operation records within their respective scopes, thus meeting the enterprise's internal control compliance requirements.

[0015] Compared with the closest existing technology, the present invention has the following advantages: By constructing a multi-source data real-time acquisition mechanism, high-frequency and comprehensive perception of market demand signals is achieved, fundamentally solving the information lag problem of traditional procurement models. By designing a fluctuation quantification model based on weighted regression, complex market dynamics are transformed into measurable demand fluctuation coefficients, and multiple threshold levels are combined to achieve precise classification of fluctuation levels, improving the objectivity and scientific rigor of demand change identification. By establishing a response rule base that includes adjustments to both procurement quantity and procurement cycle, and introducing a procurement quantity calculation formula that links the strategy adjustment coefficient k with the comprehensive supply-demand gap, refined matching of procurement actions with market conditions is achieved, significantly reducing the probability of over-purchasing and supply shortages. Crucially, this invention is the first to embed a dual verification mechanism and a closed-loop feedback optimization path into the procurement response process. Through quantitative evaluation of model prediction accuracy and strategy execution effectiveness, dynamic calibration of system parameters and rule iteration are driven, forming a complete autonomous cycle of "perception-analysis-decision-verification-feedback-optimization," greatly enhancing the adaptability, robustness, and long-term effectiveness of the procurement system, providing solid technical support for enterprises to cope with highly uncertain market environments. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for automated real-time response to market demand fluctuations in procurement data provided by the present invention; Figure 2 This is a schematic diagram of the core principle framework of a method for automating real-time response to market demand fluctuations in procurement data provided by the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0019] Example 1: This invention provides an automated real-time response method for procurement data in response to market demand fluctuations, such as... Figure 1 As shown, it includes: Step (1) involves real-time data acquisition. A multi-source data acquisition module is constructed to acquire market demand data, industry dynamic data, enterprise inventory data, and historical procurement data from external platforms and internal systems in real time through compliant data acquisition methods. Specifically, the multi-source data acquisition module is configured with an access authentication mechanism and a data encryption transmission protocol to ensure that data security standards are met when acquiring e-commerce platform sales data and search engine user search popularity data through API interfaces. This module is deployed in the enterprise's private cloud environment, establishing a two-way SSL / TLS encrypted channel with external data sources, and completing identity authentication according to the OAuth2.0 protocol to prevent unauthorized access and data leakage. The data collection targets cover four main categories: The first category is market demand data, including at least product sales volume, order volume, user search popularity, and competitor sales data; the second category is industry dynamic data, including at least policy adjustment information and raw material price fluctuation data; the third category is enterprise inventory data, sourced from the Warehouse Management System (WMS), containing real-time inventory levels, in-transit inventory, and safety stock thresholds for each SKU; and the fourth category is historical procurement data, stored in the Enterprise Resource Planning (ERP) system, recording the category, quantity, supplier, delivery cycle, and actual delivery deviation of all past purchase orders. The data collection frequency is set to a high-frequency collection cycle based on industry characteristics and demand sensitivity, ranging from once every 5 minutes to once every 2 hours, ensuring timely capture of market trends and guaranteeing the timeliness and completeness of input data. Furthermore, the data crawler component adopts a distributed architecture deployment, supporting concurrent access to multiple data sources. A single collection task can complete all data retrieval within 90 seconds, avoiding information distortion due to delays. For example, in the fast-moving consumer goods (FMCG) industry, where consumer behavior changes rapidly, the system automatically sets the data collection frequency to once every 10 minutes to capture instantaneous demand spikes triggered by promotional activities. In the durable goods industry, such as home appliances or industrial equipment, where demand changes relatively slowly, the collection frequency is set to once every 2 hours, balancing system resource consumption and response accuracy. All data collection tasks are centrally managed by a scheduler using a priority queue mechanism. When a data source is detected to have timed out or returned an error code, retry logic is automatically triggered. After a maximum of three attempts, the data source is marked as failed and an alarm is triggered. Simultaneously, a backup data source or historical cached data is used as a temporary replacement to ensure the main process is not interrupted.

[0020] In the above method, step (2), data preprocessing, involves cleaning and standardizing the collected multi-source heterogeneous data to eliminate noise interference and unify data dimensions. Specifically, during the cleaning process, an outlier identification algorithm based on statistical distribution characteristics is used to remove data points that deviate from the normal range, while duplicate records are removed, and missing data is supplemented using time series interpolation or historical averages based on similar periods. More specifically, the outlier identification algorithm employs a dual criterion of an improved interquartile range (IQR) combined with sliding window standard deviation detection: First, the first quartile (Q1) and third quartile (Q3) of each data sequence within the current sliding window (window length of 15 acquisition periods) are calculated, thus obtaining the IQR = Q3 - Q1; second, the sliding standard deviation (σ) within the same window is calculated; when a data point Di,t exceeds the range of Q1 - 3 × IQR or Q3 + 3 × IQR, and its absolute deviation from the mean of the preceding and following 5 periods exceeds twice the sliding standard deviation (i.e., |Di,t - mean(Di,t-5:t+5)|>2σ), it is identified as an outlier and removed. This dual criterion effectively distinguishes between genuine market mutations (such as black swan events) and false spikes caused by sensor malfunctions or network jitter. For missing data, the system automatically selects a completion strategy based on the data type: for core demand data such as sales volume and order volume with obvious daily, weekly, or seasonal characteristics, seasonal time series interpolation is used, filling in the missing data with the historical average of the same position in the previous 7 days; for non-periodic auxiliary data, such as policy text sentiment scores or raw material price indices, K-Nearest Neighbors Imputation is used, selecting the 5 complete samples with the smallest Euclidean distance in the most recent 7 days and filling in the missing items with the weighted average of their corresponding fields. Standardization processing uses a min-max normalization algorithm to transform data of different dimensions and orders of magnitude to a unified numerical range of 0.0 to 1.0, forming a structured standardized dataset and providing a high-quality input foundation for subsequent analysis. It is worth noting that a dynamic extreme value update mechanism is introduced in the normalization processing, recalculating the maximum and minimum values ​​every 7 days based on the latest data to prevent the long-term use of fixed extreme values ​​from compressing new data into extreme ranges. For example, if a product's sales surge due to a new product launch, the old maximum value will no longer reflect the current volume. The dynamic update mechanism ensures that new data can still be reasonably mapped to the [0.0, 1.0] range, preventing the model from failing due to input saturation. All preprocessing operations are completed in an in-memory database, with processing latency controlled within 2 seconds, ensuring that subsequent analysis modules can obtain clean and consistent data streams in real time.

[0021] In the above method, step (3), market demand fluctuation analysis, constructs a fluctuation analysis model, and based on the preprocessed standardized dataset, uses a weighted linear regression fitting formula to quantify the degree of change in current market demand relative to historical benchmarks, outputting a continuous demand fluctuation coefficient F(x). Specifically, the fitting formula is defined as follows: Wherein, α is the current demand weighting coefficient, ranging from 0.6 to 0.8, dynamically adjusted according to the industry's demand response speed; w i The weighting coefficients for the i-th type of market demand data are given, with the sum of all weights equal to 1. Core data such as sales volume and order volume are assigned higher weights; D i,t β is the standardized value of the i-th type of data in the current collection period t; β is the historical demand benchmark weighting coefficient, ranging from 0.2 to 0.4; D t The standardized value of historical average demand over the past m collection periods (m≥30) is used; γ is the fluctuation correction coefficient, determined by statistical regression of the deviation between the company's past procurement execution results and actual demand. The model runs on a dedicated analysis server, and F(x) is calculated once after each data preprocessing is completed. Weight w i The initial values ​​were set based on expert experience, but an online learning mechanism was introduced during system operation. The weights of various data points on the final procurement performance were adjusted in reverse based on each validation result. For example, when search popularity led actual sales changes by more than 7 days and the correlation coefficient was higher than 0.8, its weight was gradually increased. The fluctuation correction coefficient γ was obtained by linear regression of the cumulative deviation between the company's planned procurement volume and actual consumption over the past 12 months. The initial value was set at 0.05 and dynamically adjusted quarterly. Five preset fluctuation threshold ranges are set: [-∞, -0.3), [-0.3, -0.1), [-0.1, 0.1], (0.1, 0.3], and (0.3, +∞), corresponding to fluctuation levels of significant decrease, slight decrease, stable fluctuation, slight increase, and significant increase, respectively. The fluctuation level of current market demand is determined by identifying the range into which F(x) falls. Furthermore, the upper and lower limits of the fluctuation threshold ranges can be manually fine-tuned by enterprises according to their business objectives, or an adaptive mode can be enabled, where the system automatically defines the boundaries based on the density clustering results of historical fluctuation distributions. For example, the DBSCAN clustering algorithm can be used to perform unsupervised clustering of the F(x) sequence over the past 180 days, identifying the centers of naturally formed fluctuation clusters and redefining the threshold boundaries accordingly, making the classification more aligned with the enterprise's own business rhythm.

[0022] In the above method, step (4) involves automatically generating a procurement strategy and establishing a procurement response rule base. This rule base stores the mapping logic between different fluctuation levels and procurement adjustment strategies. Specifically, based on the fluctuation level output by the fluctuation analysis and the specific value of F(x), combined with the procurement quantity fitting formula P = k × (F(x) + 1) × (D... t - S t The target purchase quantity is automatically calculated. Where P is the target purchase quantity; k is the strategy adjustment coefficient, whose value is dynamically set according to the fluctuation level; D... t The standardized value of the current cycle's comprehensive demand is a weighted composite of multiple types of market demand data; S t This is the standardized value of the current inventory level. The specific value of the strategy adjustment coefficient k follows the logic of a piecewise function: when in a period of significant upward fluctuation, k = 1 + 0.5 × (F(x) - 0.3), with an upper limit of 1.8; when in a period of slight upward fluctuation, k = 1 + 0.3 × F(x), with an upper limit of 1.3; when in a period of downward fluctuation, k = 1 - 0.4 × |F(x)|, with a lower limit of 0.5. The standardized value of comprehensive demand D... t Use the same weights w as in volatility analysis i Perform weighted synthesis to ensure logical consistency throughout. Standardized inventory value S t The current physical inventory level is obtained by dividing it by the average demand over the last 30 periods, and then normalized to a range of 0.0 to 1.0. The procurement adjustment strategy comprises two parts: a procurement quantity adjustment logic and a procurement cycle adjustment logic. For significant price increases, a procurement quantity increment adjustment logic is applied, shortening the procurement cycle to 60% to 80% of the original cycle. For minor price increases, a moderate increment logic is applied, keeping the procurement cycle unchanged. For stable price fluctuations, the regular replenishment logic is followed, maintaining the benchmark procurement cycle. For minor price decreases, a reduction logic is applied, extending the procurement cycle to 1.2 to 1.5 times the original cycle. For significant price decreases, a substantial reduction logic is applied, extending the procurement cycle to more than 1.5 times the original cycle and triggering a non-core category procurement suspension decision process, retaining only essential category procurement tasks. The logic for suspending procurement of non-core product categories is activated under conditions of significant price fluctuations. First, it checks whether the average demand growth rate for this category over the past 30 days is below -15%. Second, it assesses whether the inventory turnover days exceed the company's set safety limit of 1.5 times. Finally, it considers the financial indicator that the gross profit margin is below the overall average of 70%. If all three conditions are met simultaneously, a procurement freeze recommendation is generated and pushed to the management approval channel, achieving intelligent control of risky product categories. All policy rules are stored in a rule base in JSON format, supporting hot updates and taking effect without requiring a service restart.

[0023] In the above method, step (5), fitting model and strategy verification, sets up a dual verification mechanism to evaluate the effectiveness of the volatility analysis model and the generated procurement strategy. Specifically, firstly, fitting model verification is performed. A verification set is constructed using n sets of known historical actual demand data and corresponding actual procurement behaviors. The data of each period in the verification set are input into the volatility analysis model, and the mean square error (MSE) between the predicted volatility coefficient Fpred(x) and the actual volatility coefficient Ftrue(x) output by the model is calculated. The calculation formula is as follows: The default tolerance threshold for MSE is set to 0.02. If the validation results exceed this threshold three consecutive times, global parameter retraining is forcibly initiated. The validation set is constructed using a rolling window approach, including historical data from the most recent 180 consecutive days, with each day serving as an independent validation sample to ensure the model is always tested in the latest business environment. Next, the procurement strategy is validated. Based on the same validation set, the generated procurement strategy is simulated, and the simulated inventory turnover rate, stockout rate, and backlog rate are calculated and compared with the company's default target thresholds. The procurement strategy simulation uses a discrete event simulation engine to simulate the inventory inflow and outflow process over the next three procurement cycles, comprehensively evaluating the strategy's impact on future inventory health. During the simulation, the uncertainty of supplier delivery cycles is considered, and a ±15% time disturbance factor is introduced to enhance robustness testing. If any key indicator fails to meet the target, the corresponding k-value, cycle adjustment range, or pause logic trigger condition in the adjustment rule base is returned to complete the strategy parameter calibration. For example, if the simulation results show a stockout rate exceeding 5%, the system will automatically lower the upper limit of the k-value or relax the procurement cycle shortening ratio until the indicator returns to a safe range.

[0024] In the above method, step (6) involves real-time execution and feedback optimization. The procurement strategy, which has been verified and confirmed to be effective through dual validation, is pushed to the enterprise procurement management system through the system interface to drive the automated execution of the procurement process. Specifically, the procurement strategy push interface uses message queue middleware to achieve asynchronous communication, ensuring that instructions are not lost even if the downstream system is temporarily unavailable. The feedback data acquisition module is equipped with a timed task scheduler, which automatically extracts order execution data from the ERP system and inventory change logs from the WMS system every hour, and continuously collects key performance indicators during the procurement execution process, including but not limited to purchase order completion rate, order response time, actual delivery time, and warehousing accuracy; it also synchronously collects dynamic inventory change data to reflect the actual inventory level changes after the strategy is executed. The above feedback data is sent back to the fluctuation analysis model and the procurement response rule base, and the verification logic in step (5) is called again to re-evaluate the effectiveness of the model and strategy. If performance degradation or environmental deviation is detected (such as continuous increase in MSE or inventory backlog rate exceeding the threshold), a new round of parameter iteration update is automatically triggered to dynamically adjust the fluctuation threshold interval boundary, the fitting formula parameter set, and the strategy mapping parameters in the rule base. The parameter iteration and update process employs a canary release mechanism, first verifying the effectiveness of new parameter combinations in a simulation environment before gradually applying them to actual business flows, preventing sudden changes from causing system oscillations. The entire closed-loop feedback cycle is controlled to complete one full iteration within 24 hours, achieving weekly self-evolution capabilities. This method is integrated into an enterprise supply chain intelligent decision-making platform, supporting complex architecture scenarios involving multiple organizations, warehouses, and suppliers. The platform has a built-in visual monitoring panel that displays real-time fluctuation coefficient change curves, procurement strategy execution status, inventory warning signals, and model validation reports. The system features hierarchical access control, allowing users at different levels to view corresponding data and operation records, meeting enterprise internal control compliance requirements.

[0025] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0026] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0027] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0028] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0029] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for automated real-time response to procurement data fluctuations, characterized in that, include: Step 1: Real-time data acquisition. A multi-source data acquisition module is built to acquire market demand data, industry dynamics data, enterprise inventory data, and historical procurement data in real time from external platforms and internal systems through compliant data acquisition methods. Market demand data includes at least product sales volume, order volume, user search popularity, and competitor sales data; industry dynamics data includes at least policy adjustment information and raw material price fluctuation data. The data acquisition frequency is set to perform data scraping once every 5 minutes to every 2 hours. Step 2: Data preprocessing. The collected multi-source heterogeneous data is cleaned and standardized. During the cleaning process, an outlier identification algorithm based on statistical distribution characteristics is used to remove data points that deviate from the normal range and duplicate records are removed. Time series interpolation or historical mean based on similar periods is used to fill in missing data. The standardization process uses the min-max normalization algorithm to transform data of different dimensions to a unified numerical range of 0.0 to 1.0, forming a structured standardized dataset. Step 3: Market demand fluctuation analysis, constructing a fluctuation analysis model, and using a weighted linear regression fitting formula based on the preprocessed standardized dataset. This quantifies the degree of change in current market demand relative to a historical benchmark, outputting a continuous demand volatility coefficient F(x); where α is the current demand weighting coefficient, ranging from 0.6 to 0.8; w i D represents the weighting coefficients for the i-th type of market demand data, where the sum of all weights equals 1; i,t β is the standardized value of the i-th type of data in the current collection period t; β is the historical demand benchmark weighting coefficient, ranging from 0.2 to 0.4; D t The standardized value of the historical average demand over the past m collection periods (m≥30) is used; γ is the fluctuation correction coefficient; five preset fluctuation threshold intervals are set: [-∞,-0.3), [-0.3,-0.1), [-0.1,0.1], (0.1,0.3], and (0.3,+∞), corresponding to fluctuation levels of large decrease, small decrease, stable, small increase, and large increase, respectively; the current fluctuation level is determined by judging the interval into which F(x) falls. Step 4: Automatic generation of procurement strategies, establishment of a procurement response rule base, and fitting of the procurement quantity formula based on the fluctuation level and the specific value of F(x). The target purchase quantity is automatically calculated; where P is the target purchase quantity; k is the strategy adjustment coefficient, whose value is dynamically set according to the fluctuation level; D t S is the standardized value of the comprehensive demand for the current cycle; t The current inventory level is a standardized value. The procurement adjustment strategy includes procurement quantity adjustment logic and procurement cycle adjustment logic. For significant upward fluctuations, the procurement quantity increment adjustment logic is matched and the procurement cycle is shortened to 60% to 80% of the original cycle. For slight upward fluctuations, the moderate increment logic is matched and the procurement cycle remains unchanged. For stable fluctuations, the regular replenishment logic is executed. For slight downward fluctuations, the quantity reduction logic is matched and the procurement cycle is extended to 1.2 to 1.5 times the original cycle. For significant downward fluctuations, the significant quantity reduction logic is matched and the procurement cycle is extended to more than 1.5 times the original cycle, and the non-core category procurement suspension judgment process is triggered. Step 5: Model Fitting and Strategy Validation, with a dual validation mechanism; First, the fitted model is validated by constructing a validation set using n sets of historical known actual demand data and corresponding actual purchasing behavior, and calculating the mean square error (MSE) between the predicted volatility coefficient Fpred(x) and the actual volatility coefficient Ftrue(x) output by the model; if the MSE exceeds the preset tolerance threshold, the parameter optimization process is initiated; Second, the purchasing strategy is validated by simulating the generated purchasing strategy based on the same validation set, calculating the inventory turnover rate, stockout rate, and backlog rate after the simulated purchasing, and comparing them with the company's preset target thresholds; if any key If the indicators are not met, the system returns to adjust the parameters in the rule base; Step 6: Real-time execution and feedback optimization, the procurement strategy that has been double-verified and confirmed to be effective is pushed to the enterprise procurement management system through the system interface; at the same time, a closed-loop feedback mechanism is built to continuously collect key performance indicators and inventory dynamic change data during the procurement execution process; the feedback data is sent back to the fluctuation analysis model and the procurement response rule base, and the verification logic in step 5 is called again; if performance degradation or environmental deviation is detected, a new round of parameter iteration update is automatically triggered, dynamically adjusting the fluctuation threshold range boundary, the fitting formula parameter set, and the strategy mapping parameters in the rule base.

2. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: In step 1, the multi-source data acquisition module is configured with an authorization authentication mechanism and a data encryption transmission protocol; the data crawler component adopts a distributed architecture deployment, supports concurrent access to multiple data sources, and completes the retrieval of all data in a single acquisition task within 90 seconds; the acquisition frequency is set with a differentiated strategy based on the product category, with fast-moving consumer goods set to be acquired once every 10 minutes and durable goods set to be acquired once every 2 hours.

3. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: In step 2, the outlier identification algorithm uses a modified interquartile range method combined with a sliding window standard deviation detection as a dual criterion. When a data point exceeds the range of Q1 - 3×IQR or Q3 + 3×IQR, and its deviation from the mean of the previous 5 periods exceeds 2 times the sliding standard deviation, it is judged as an outlier and removed. The data completion algorithm selects different strategies according to the data type. For demand data with obvious periodicity, seasonal time series interpolation is used, and for non-periodic auxiliary data, K-nearest neighbor interpolation is used. In the normalization process, a dynamic extreme value update mechanism is introduced, and the maximum and minimum values ​​are recalculated every 7 days based on the latest data.

4. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: In step 3, the weight w i The initial values ​​are set based on expert experience, and an online learning mechanism is introduced during system operation to adjust the weights of various data on the final procurement performance in reverse according to the results of each verification. fluctuation The correction coefficient γ is obtained by linear regression of the cumulative deviation between the enterprise's planned purchase volume and actual consumption over the past 12 months. The initial value is set to 0.05 and is dynamically corrected with quarterly updates. The upper and lower limits of the fluctuation threshold range can be manually fine-tuned by enterprises, or the adaptive mode can be enabled so that the system can automatically divide the boundaries based on the density clustering results of historical fluctuation distribution.

5. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: In step 4, the specific value of the strategy adjustment coefficient k follows piecewise function logic: when in a period of significant upward fluctuation, k = 1 + 0.5 × (F(x) - 0.3), with an upper limit of 1.8; when in a period of slight upward fluctuation, k = 1 + 0.3 × F(x), with an upper limit of 1.3; when in a period of downward fluctuation, k = 1 - 0.4 × |F(x)|, with a lower limit of 0.5; the comprehensive demand standardized value D t Use the same weights w as in volatility analysis i Perform weighted synthesis; standardized inventory value S t It is obtained by dividing the current physical inventory by the average demand over the last 30 periods, and then normalized and mapped to the range of 0.0 to 1.

0.

6. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: In step 5, the validation set is constructed using a rolling window approach, which includes historical data from the most recent 180 consecutive days, with each day serving as an independent validation sample. The preset tolerance threshold for MSE is set to 0.

02. If the validation results exceed the threshold for three consecutive times, global parameter retraining will be forcibly initiated. The procurement strategy simulation uses a discrete event simulation engine to simulate the inventory inflow and outflow process over the next three procurement cycles, and introduces a ±15% time disturbance factor to enhance robustness testing.

7. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: In step 6, the procurement strategy push interface uses message queue middleware to achieve asynchronous communication; the feedback data acquisition module is configured with a timed task scheduler, which automatically extracts order execution data from the ERP system and inventory change logs from the WMS system every hour; the parameter iteration update process adopts a canary release mechanism, first verifying the effectiveness of the new parameter combination in a simulation environment, and then gradually applying it to the actual business flow. The entire closed-loop feedback cycle is controlled to complete one full iteration within 24 hours.

8. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: The non-core category procurement suspension decision process is activated under conditions of significant price fluctuations. First, it is determined whether the average demand growth rate of the category in the past 30 days is lower than -15%. Second, it is assessed whether its inventory turnover days exceed the company's set safety limit of 1.5 times. Finally, combined with the financial indicator that the gross profit margin is lower than the overall average of 70%, a procurement freeze suggestion is generated when all three conditions are met and pushed to the management approval channel.

9. The method for automated real-time response to market demand fluctuations in procurement data according to claim 1, characterized in that: The method is integrated into the enterprise supply chain intelligent decision-making platform, supporting complex architecture scenarios with multiple organizations, multiple warehouses, and multiple suppliers; The platform has a built-in visual monitoring panel that displays the fluctuation coefficient change curve, procurement strategy execution status, inventory warning signals, and model validation reports in real time. The system has a hierarchical access control function, allowing users at different levels to view data and operation records within their respective ranges.

10. A method for automated real-time response to market demand fluctuations in procurement data according to claim 3, characterized in that: The sliding window length is 15 collection periods; the seasonal time series interpolation method fills the missing data using the historical average of the same period position in the previous 7 days; the K-nearest neighbor interpolation method selects the 5 complete samples with the smallest Euclidean distance in the most recent 7 days and fills the missing items with the weighted average of their corresponding fields.

Images