Intelligent purchase decision-making and automatic execution method based on demand prediction

Abstract

The invention provides an intelligent purchase decision and automatic execution method based on demand prediction, and relates to the technical field of data processing, and the method comprises the steps: obtaining multi-source data, constructing a demand feature vector set, and screening key demand feature vectors through a principal component analysis method; based on the key demand feature vector, using a pre-constructed extreme learning machine demand prediction model to output a demand prediction quantity and a demand fluctuation trend; dynamically calculating an inventory safety index in combination with the inventory data, the demand predicted quantity and the demand fluctuation trend, and judging whether the inventory safety index is greater than a threshold value or not; if yes, continuing monitoring, otherwise, combining the demand prediction amount, the demand fluctuation trend, the inventory data and the supply-side historical performance data, comprehensively considering the supply and price safety factors to generate a preliminary purchase scheme, performing risk constraint verification, if yes, taking the preliminary purchase scheme as a final purchase scheme, and if not, adjusting purchase parameters for regeneration; and converting the final purchasing scheme into a standardized purchasing instruction and automatically issuing the standardized purchasing instruction to a purchasing execution system.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for intelligent procurement decision-making and automated execution based on demand forecasting. Background Technology

[0002] With the increasing demands for digital transformation and refined supply chain management, procurement management has gradually evolved from a traditional experience-driven model to a data-driven and intelligent decision-making approach. Especially in business scenarios involving multiple product categories, multiple suppliers, and high-frequency demand changes, accurately predicting procurement needs based on historical data, business fluctuations, and changes in the external environment, and then achieving scientific decision-making and automated execution accordingly, has become a key technical challenge for enterprises to reduce inventory costs, improve supply efficiency, and enhance market responsiveness.

[0003] Currently, in existing technologies, procurement decisions typically rely on Enterprise Resource Planning (ERP) systems or Supply Chain Management (SCM) systems. Their core processes are largely based on historical procurement records, inventory thresholds, or manually set safety stock rules to trigger procurement activities. Some systems introduce simple statistical analysis methods or time series models to predict future demand, and procurement personnel then manually revise the predictions based on their experience before generating a procurement plan.

[0004] However, existing demand forecasting models are generally quite simplistic, relying heavily on simple statistical methods or fixed time series analysis. They lack the ability to characterize complex influencing factors such as seasonal fluctuations, unexpected events, and adjustments to business strategies, resulting in low forecast accuracy. Furthermore, existing technologies often optimize for single materials or static rules, lacking a comprehensive balancing mechanism for procurement costs, delivery cycles, and inventory risks. This limits the correlation between forecast results and actual decisions, thus restricting the overall intelligence and adaptability of procurement decisions. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide an intelligent procurement decision-making and automated execution method based on demand forecasting. This method addresses the limitations of existing demand forecasting models, which are generally simplistic, relying heavily on simple statistical methods or fixed time series analysis. These models lack the ability to adequately characterize complex influencing factors such as seasonal fluctuations, unexpected events, and adjustments to business strategies, resulting in low forecast accuracy. Furthermore, existing technologies often optimize single materials or static rules, lacking a comprehensive balancing mechanism for procurement costs, delivery cycles, and inventory risks. This results in limited correlation between forecast results and actual decisions, thus restricting the overall intelligence and adaptability of procurement decisions.

[0006] A first aspect of this invention proposes an intelligent procurement decision-making and automated execution method based on demand forecasting, comprising: S1: Obtain multi-source data related to procurement needs; S2: Based on the multi-source data, construct a demand feature vector set, and use principal component analysis to select key demand feature vectors from the demand feature vector set; S3: Based on the key demand feature vector, predict future procurement demand using a pre-built extreme learning machine demand prediction model, and output the predicted demand amount and demand fluctuation trend within the prediction period. S4: Based on the inventory data in the multi-source data, the demand forecast, and the demand fluctuation trend, dynamically calculate the inventory safety index; S5: Determine whether the inventory safety index is greater than the preset inventory safety index; if so, return to S1 to continue monitoring after a preset time period; otherwise, proceed to S6. S6: Based on the demand forecast, the demand fluctuation trend, and the inventory data and historical supply-side fulfillment data from the multi-source data, and taking into account both supply security and price security factors, a preliminary procurement plan is generated. S7: Perform risk constraint verification on the preliminary procurement plan and determine whether the preliminary procurement plan passes the verification. If yes, retain the verified preliminary procurement plan as the final procurement plan and proceed to S8. Otherwise, adjust the procurement parameters and return to S6 to regenerate the procurement plan. S8: Convert the final procurement plan into a standardized procurement instruction, and automatically send the standardized procurement instruction to the procurement execution system through the system interface.

[0007] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, by acquiring multi-source data related to procurement needs, constructing and filtering key demand features, and combining them with an extreme learning machine demand forecasting model to predict future procurement needs and demand fluctuation trends, it is possible to effectively characterize complex influencing factors such as seasonal fluctuations, sudden events, and business strategy adjustments, thereby improving the accuracy and stability of demand forecasting in dynamic business scenarios. At the same time, based on the demand forecasting results and inventory data, the inventory safety index is dynamically calculated, and the risks of supply interruption and price fluctuations are comprehensively considered during the generation and verification of procurement plans. This achieves closed-loop linkage between the forecasting results and the procurement decision-making and execution processes, thereby avoiding the limitations of single material or static rule optimization and significantly improving the overall intelligence level and adaptability of procurement decisions. Attached Figure Description

[0008] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0009] Figure 1 This is a flowchart illustrating an intelligent procurement decision-making and automated execution method based on demand forecasting, provided by an embodiment of the present invention. Detailed Implementation

[0010] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions 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. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0011] The following description, in conjunction with the accompanying drawings, details the intelligent procurement decision-making and automated execution method based on demand forecasting provided by the embodiments of the present invention through specific examples and application scenarios.

[0012] Reference manual attached Figure 1 The diagram illustrates a flowchart of an intelligent procurement decision-making and automated execution method based on demand forecasting, provided by an embodiment of the present invention.

[0013] This invention provides a method for intelligent procurement decision-making and automated execution based on demand forecasting, which may include the following steps: S1: Obtain multi-source data related to procurement needs.

[0014] In one possible implementation, the multi-source data includes historical sales data, order data, inventory data, historical fulfillment data from the supply side, historical purchase price data, external price data, and supplier delivery data.

[0015] Specifically, multi-source data related to procurement needs is obtained through internal enterprise information systems and external data interfaces. Historical sales and order data are obtained by calling the data interfaces of the sales management system or order management system, acquiring the sales quantity, outbound quantity, order quantity, and order generation time of each material within a historical period, granularly according to time. Inventory data is obtained through the inventory management system or warehouse management system, including current available inventory, in-transit inventory, and historical inventory change records. Inventory data can be obtained via real-time interfaces or periodic batch synchronization. Price data is obtained through the procurement management system to acquire historical purchase prices, while market prices or supplier quotations are obtained through external price data interfaces. Supplier delivery data is obtained through the supplier management system or supply chain collaboration platform. By matching historical purchase orders with actual delivery records, the average delivery cycle, on-time delivery rate, and historical interruption status of suppliers are calculated to form historical supply-side fulfillment data.

[0016] S2: Based on multi-source data, construct a set of demand feature vectors, and use principal component analysis to select key demand feature vectors from the set of demand feature vectors.

[0017] Principal component analysis (PCA) is a commonly used multivariate statistical analysis method for dimensionality reduction of high-dimensional feature data. Its core idea is to map multiple correlated original feature variables into a set of mutually orthogonal and independent comprehensive feature variables—the principal components—through linear transformation, while preserving as much of the original data's main information as possible. By ranking the principal components according to their contribution to the data variance and selecting the principal components with the highest cumulative contribution rates, the dimensionality of features can be effectively reduced, correlations and redundant information between features can be eliminated, and key features with a significant impact on the target variable can be highlighted.

[0018] In one possible implementation, S2 specifically includes: S201: Align historical sales data, order data, inventory data, historical fulfillment data from the supply side, external price data, and supplier delivery data from multiple sources by time dimension to obtain aligned multi-source data.

[0019] Specifically, when aligning multi-source data by time dimension, a unified time granularity and time index are first determined as the alignment benchmark. The time granularity can be daily, weekly, or monthly. Subsequently, for historical sales data, order data, inventory data, historical fulfillment data from the supply side, external price data, and supplier delivery data, the original timestamps are mapped and grouped according to the unified time granularity. By summarizing, statistically analyzing, or matching the data within the same time window, data from different sources and with different sampling frequencies are converted into data sequences with the same time index.

[0020] S202: Based on aligned multi-source data, extract basic features related to procurement needs.

[0021] Optionally, the basic characteristics include: historical demand characteristics (sales or outbound volume in the current period and several previous periods); order volume characteristics (number of orders, order growth rate); inventory status characteristics (available inventory, in-transit inventory); price characteristics (current price, price fluctuation range); and supply delivery characteristics (average delivery time, on-time delivery rate).

[0022] Specifically, historical demand characteristics are obtained by summarizing sales or outbound records; order quantity and order growth rate characteristics are obtained by statistically analyzing and calculating ratios of order records; inventory status characteristics such as available inventory and in-transit inventory are obtained by summarizing and calculating inventory status; price level and price change characteristics are obtained by statistically analyzing and calculating the difference between historical purchase prices and current prices; and supply delivery characteristics such as average delivery cycle and on-time delivery rate are obtained by statistically analyzing the time difference and completion status between historical purchase orders and actual delivery records. This forms a feature set used to characterize the basic state of procurement demand.

[0023] S203: Extract time-related temporal features based on aligned multi-source data.

[0024] Optionally, time-related features include: cyclical position features (weekly sequence number, monthly sequence number); and seasonal features (quarterly identifier, peak / off-peak season identifier).

[0025] Specifically, time-series features related to changes in procurement demand are extracted from timestamps according to a unified time index. These time-series features are obtained by discretizing and identifying time information, including periodic position features to characterize the data's position within a cycle and seasonal features to depict periodic fluctuations. The periodic position features are obtained by mapping timestamps to numerical forms such as weekly or monthly sequences, while the seasonal features are obtained by marking the quarter or peak / off-peak season interval to which the timestamp belongs.

[0026] S204: Extract demand change trend features based on historical demand features in the basic features.

[0027] Optionally, demand change trend characteristics include: month-on-month change rate, year-on-year change rate, sliding window average and sliding window standard deviation, and the slope of the demand growth trend.

[0028] Specifically, the month-on-month change rate is obtained by calculating the change in demand between adjacent periods, the year-on-year change rate is obtained by comparing the current period's demand with the corresponding historical period, and the sliding window mean and standard deviation are obtained by calculating the mean and dispersion of the demand data within the sliding time window. In addition, the slope of the demand growth trend can be obtained by fitting the demand sequence within the time window.

[0029] S205: Normalize the basic features, time-series features, and demand change trend features, and then, according to a unified time index, splice and combine the normalized basic features, time-series features, and demand change trend features to construct a demand feature vector set.

[0030] S206: Using principal component analysis, key demand feature vectors are selected from the set of demand feature vectors.

[0031] It should be noted that principal component analysis is an existing technology, and will not be elaborated upon here.

[0032] In this embodiment of the invention, by aligning multi-source data along the time dimension and extracting basic features, time-related temporal features, and demand change trend features in a hierarchical manner, the invention can comprehensively characterize the scale characteristics, temporal distribution characteristics, and changing patterns of procurement demand from multiple perspectives, avoiding information gaps caused by relying solely on single historical demand data. Based on this, the different types of features are uniformly normalized and concatenated according to time indexes to form a structurally consistent demand feature vector set. This allows multi-source heterogeneous data to serve as a unified input for subsequent modeling, improving the stability of model training and prediction processes. Furthermore, by introducing principal component analysis to reduce the dimensionality of the demand feature vector set, the correlation and redundant information between features can be eliminated while retaining the main information, highlighting key features that have a significant impact on demand changes, and providing more accurate, stable, and representative feature inputs for subsequent demand forecasting and procurement decisions.

[0033] S3: Based on key demand feature vectors, it uses a pre-built extreme learning machine demand forecasting model to predict future procurement demand and outputs the predicted demand amount and demand fluctuation trend within the forecast period.

[0034] The Extreme Learning Machine (ELM) demand forecasting model is a fast learning model based on a single-hidden-layer feedforward neural network. It randomly generates and fixes the hidden layer parameters, solving only for the output layer weights, thus significantly reducing model training complexity while maintaining prediction accuracy. In demand forecasting applications, ELM can quickly learn the nonlinear mapping relationship between features and demand quantities using multidimensional input demand feature vectors, demonstrating excellent modeling capabilities for complex, nonlinear, and high-dimensional data. Furthermore, its training process requires no iterative parameter tuning, resulting in high computational efficiency. It is suitable for rapid retraining and online prediction in scenarios where demand data is continuously updated, thereby improving the stability, real-time performance, and adaptability of demand forecasting results in dynamic business environments.

[0035] It should be noted that the Extreme Learning Machine (ELM) demand forecasting model is a pre-built and trained model. Its training phase is based on historical demand data and corresponding key demand feature vectors, used to learn the mapping relationship between key demand features and demand quantities. After training, the model parameters remain fixed and are directly used for demand forecasting calculations during the forecasting phase.

[0036] In one possible implementation, S3 specifically includes: S301: Normalize the key demand feature vectors at each time point within the prediction period to obtain normalized key demand feature vectors.

[0037] Alternatively, the normalization process can use existing max-min normalization.

[0038] S302: Based on the normalized key requirement feature vector, construct the kernel similarity vector for each prediction time: in, Indicates the predicted time The corresponding kernel similarity vector, Indicates the predicted time The normalized key requirement feature vector, Indicates the first n The key requirement feature vector corresponding to each training sample n =1,2,…, l , l Indicates the number of training samples. T This represents the matrix transpose operation. Represents the kernel function.

[0039] Optionally, the kernel function can be a Gaussian kernel.

[0040] In this embodiment of the invention, by constructing a kernel similarity vector between the prediction time and historical training samples based on the normalized key demand feature vector, and using the kernel function to map the similarity relationship in the feature space, the nonlinear similarity relationship between demand features can be effectively characterized without explicitly constructing high-dimensional features, thereby improving the model's ability to express complex demand patterns.

[0041] S303: Based on kernel similarity vectors, an extreme learning machine demand prediction model is used to predict future procurement demand, and the output model's demand prediction is: in, Indicates the predicted time The corresponding demand forecast, Indicates the predicted time The transpose of the corresponding kernel similarity vector, This represents the coefficient vector obtained from training the kernel extreme learning machine model. K This represents the kernel matrix corresponding to the training samples. The elements of the kernel matrix are calculated using the kernel function. This represents the regularization parameter (those skilled in the art can set the size of the regularization parameter according to actual needs, but this invention does not limit it). This represents the target output vector corresponding to the training phase. This represents the matrix inversion operation.

[0042] In this embodiment of the invention, by calculating the demand forecast based on kernel similarity vectors and using the analytical solution of kernel extreme learning machine, the computational overhead caused by the traditional neural network's reliance on iterative training and repeated parameter adjustment can be avoided while maintaining the model's nonlinear fitting ability, thus enabling the demand forecast model to have high computational efficiency and stability.

[0043] S304: Perform inverse normalization on the model's demand forecast to obtain the demand forecast.

[0044] S305: Calculate the rolling standard deviation of the demand forecast and use the rolling standard deviation as the trend of demand fluctuation.

[0045] In this embodiment of the invention, by uniformly normalizing the key demand feature vectors within the prediction period and constructing a kernel similarity relationship between the prediction time and historical samples based on a kernel function, the invention can effectively characterize the nonlinear mapping between demand features and demand quantities within the framework of extreme learning machines, improving the fitting ability for complex, high-dimensional demand features. Simultaneously, by employing the analytical solution form of kernel extreme learning machines to calculate the demand forecast, the computational overhead of repeated iterative training is avoided, making it suitable for rapid prediction in scenarios where demand data is continuously updated. Furthermore, by calculating the rolling standard deviation of the prediction results, demand uncertainty is quantified into demand fluctuation trends, enabling the model output to simultaneously include demand forecast quantities and fluctuation information, thereby providing a more comprehensive and reliable input for subsequent inventory safety assessments and procurement decisions.

[0046] S4: Dynamically calculate the inventory safety index based on inventory data, demand forecasts, and demand fluctuation trends from multiple sources.

[0047] It should be noted that traditional static inventory control methods are usually based on fixed safety stock or fixed reorder points, which makes it difficult to reflect the impact of demand fluctuations and business environment adjustments in a timely manner, and can easily lead to stockout risks or inventory backlog problems.

[0048] Therefore, this invention innovatively designs an inventory safety index, which can identify potential stockout risks or excessive inventory risks earlier, providing a more accurate, sensitive and adaptive basis for procurement-triggered decisions, and effectively improving the security and reliability of inventory management and procurement decisions in dynamic business environments.

[0049] Among them, the inventory safety index is a quantitative indicator used to comprehensively characterize the safety level of inventory status. It compares the current available inventory with the reorder point calculated based on demand forecast results, reflecting the inventory's ability to guarantee future demand under a given service level target.

[0050] In one possible implementation, the calculation method for the inventory safety index specifically includes: Obtain inventory data at the assessment time from multi-source data and calculate available inventory.

[0051] Specifically, available inventory is calculated from current inventory status data. The specific calculation method is to count the existing inventory quantity of materials at the assessment time, and then deduct the inventory quantity that has been occupied by orders, reserved but not yet shipped, or allocated to other business needs. Optionally, inventory that is about to expire or become unavailable is also considered, thereby obtaining the inventory quantity that can be used to meet future procurement needs without affecting existing commitments.

[0052] Obtain procurement and supply-side parameters, including average lead time and target service level coefficient.

[0053] Based on the demand forecast, calculate the average demand forecast over the average lead time.

[0054] Specifically, based on the demand forecast, when calculating the average demand forecast, the assessment time is taken as the starting point. According to the average lead time of the procurement or supply side, the demand forecast for each forecast time within the lead time is selected. The selected demand forecasts are then accumulated or averaged to obtain the average demand forecast within the average lead time.

[0055] It should be noted that the demand forecast mean is used to characterize the overall demand level that is expected to be met before the material arrives.

[0056] Based on demand fluctuation trends, the cumulative demand variance over the average lead time is calculated.

[0057] Specifically, when calculating the cumulative demand fluctuation, the assessment time is taken as the starting point. Based on the average lead time of procurement or supply, the demand fluctuation trend value corresponding to each forecast time within the lead time is selected. The degree of demand fluctuation at each forecast time is squared and then summed. The summation result is then squared to obtain the cumulative demand variance within the average lead time.

[0058] It should be noted that the cumulative demand variance is used to characterize the overall level of demand uncertainty during the lead time.

[0059] Safety stock is calculated based on the target service level coefficient and the cumulative demand variance.

[0060] Specifically, by combining the service level coefficient with the cumulative demand fluctuations obtained within the average lead time, the uncertainty of demand is amplified or scaled to obtain a safety stock that matches the target service level.

[0061] It should be noted that safety stock is used to buffer normal supply in the event of demand fluctuations or supply delays.

[0062] Calculate the reorder point based on the average demand forecast and safety stock.

[0063] Specifically, the average demand forecast within the average lead time is used as the basic demand that needs to be met before replenishment arrives. A safety stock level determined based on the target service level is then added to this level to obtain the inventory threshold that triggers a new round of procurement or replenishment. When the current available inventory level drops to near this threshold, it indicates that the procurement process needs to be initiated to avoid the risk of stockouts.

[0064] The reorder point is not a fixed constant, but a dynamic inventory warning line determined by demand forecasts and safety stock.

[0065] It should be noted that those skilled in the art can set the inventory threshold according to actual needs, and this invention does not impose any limitations on it.

[0066] By combining available inventory and reorder point, the inventory safety index is dynamically calculated: in, express t Inventory safety index at any given time express t Available inventory at any given time Indicates a reorder point.

[0067] In this embodiment of the invention, by incorporating demand forecasts and demand fluctuation trends into the inventory assessment process, and combining average lead time and target service level, the invention dynamically calculates the average demand forecast, safety stock, and reorder point. This invention constructs an inventory safety index that can be updated in real time according to changes in demand and business environment, providing more accurate, sensitive, and adaptive support for procurement-triggered decisions, and effectively improving the security, reliability, and overall responsiveness of inventory management and procurement decisions in dynamic business environments.

[0068] S5: Determine if the inventory safety index is greater than the preset inventory safety index. If yes, return to S1 to continue monitoring after the preset time. Otherwise, proceed to S6.

[0069] Optionally, the preset inventory safety index is 1.

[0070] It should be noted that those skilled in the art can set the preset time and the preset inventory safety index according to actual needs, and this invention does not limit this.

[0071] Specifically, the inventory safety index obtained at the assessment time is compared with the value 1: when the inventory safety index is greater than 1, it indicates that the current available inventory is higher than the reorder point, and the inventory has sufficient guarantee for future demand. At this time, no procurement operation is triggered. Instead, after a preset time interval, the multi-source data acquisition step is returned to continuously monitor the inventory status. When the inventory safety index is less than or equal to 1, it indicates that the current available inventory is lower than or close to the dynamic reorder point, and the inventory's ability to guarantee future demand is insufficient, with potential stockout risks.

[0072] In this embodiment of the invention, by comparing the inventory safety index with a preset inventory safety index and using the comparison result as a branch condition for whether to trigger a procurement decision, the invention constructs an adaptive procurement triggering mechanism based on the degree of inventory risk. This transforms the procurement decision from a traditional discrete triggering method based on a fixed inventory threshold to a dynamic decision-making method based on continuous quantitative indicators. This enables precise matching between inventory safety and procurement timing, significantly improving the accuracy, sensitivity, and overall resource utilization efficiency of procurement decisions.

[0073] It should be noted that, based on key demand characteristics, extreme learning machines are used to predict future demand and demand fluctuation trends, resulting in accurate demand forecasts. Furthermore, based on these accurate forecasts, the demand forecasts and fluctuation trends are used to dynamically assess inventory status, constructing an inventory safety index to determine whether inventory is within a safe range. This organically combines demand forecasting, inventory risk assessment, and procurement trigger decisions, forming a closed-loop mechanism comprised of forecast-driven processes, risk perception, and branch decision-making. This effectively avoids the disconnect between forecast results and actual procurement decisions found in traditional methods, significantly improving the accuracy, adaptability, and overall reliability of procurement decisions in dynamic business environments.

[0074] S6: Based on demand forecasts, demand fluctuation trends, and inventory data from multiple sources, as well as historical supply-side fulfillment data, and taking into account both supply security and price security factors, a preliminary procurement plan is generated.

[0075] It should be noted that although the aforementioned steps have yielded demand forecasts, demand fluctuation trends, and inventory status data, generating a procurement plan based solely on this information would still primarily focus on demand-side and inventory-side factors, making it difficult to fully reflect the potential risks arising from unstable supply fulfillment capabilities or market price fluctuations.

[0076] Therefore, this invention innovatively introduces supply security coefficient and price security coefficient to construct supply security factors and price security factors, and uses them as constraints in the procurement plan generation process. This enables the preliminary procurement plan to meet demand forecasting and inventory constraints while taking into account supply continuity and price stability, effectively improving the robustness and feasibility of the procurement plan in complex supply environments.

[0077] In one possible implementation, S6 specifically includes: S601: Obtain historical supply-side performance data, historical procurement price data, and external price data from multiple sources. Among them, historical supply-side performance data includes on-time delivery rate, historical number of interruptions, and average interruption duration.

[0078] S602: Calculate the supply safety factor based on on-time delivery rate, historical number of interruptions, and average interruption duration: in, Indicates the supply safety factor. This represents the supply risk adjustment coefficient, with a value in the range (0.1). This indicates a supply disruption risk index.

[0079] It should be noted that while the supply security factor reflects the degree of risk to supply continuity, it retains the flexibility to be adjusted according to the importance of different materials and risk preferences. This makes the generated procurement plan more robust and executable in the face of supply disruptions or delivery fluctuations, and reduces the risk of stockouts or delays caused by supply instability during the procurement process.

[0080] In one possible implementation, S602 specifically includes: S6021: Normalize the historical performance data of the supply side, and calculate the supply disruption risk index based on the normalized historical performance data of the supply side.

[0081] Specifically, historical supply-side performance data is normalized to map performance indicators with different dimensions and value ranges to a unified numerical range. This historical performance data includes on-time delivery rate, historical interruption count, and average interruption duration. Subsequently, based on the varying degrees of impact of each performance indicator on supply continuity, corresponding weights are assigned to each normalized performance indicator. The normalized values ​​of each indicator are then weighted and summed with their corresponding weights to obtain a supply interruption risk index that characterizes the degree of supply interruption risk. The normalized performance indicators are confined to the (0,1) range, and each weight is non-negative and the sum of the weights is 1. Therefore, the supply interruption risk index is also confined to the (0,1) range.

[0082] It should be noted that the supply disruption risk index is (0,1). The closer it is to 1, the higher the risk; the closer it is to 0, the more stable the supply.

[0083] S6022: Based on the supply disruption risk index, the supply security coefficient is calculated through a mapping function.

[0084] Among them, the supply security factor is an adjustment parameter used to quantify the impact of supply-side performance stability on procurement decisions. By analyzing the supplier's historical performance data, factors such as on-time delivery rate, historical interruption situation and interruption duration are comprehensively mapped into a calculable risk level, and further transformed into a coefficient used to correct procurement parameters.

[0085] It should be noted that the significance of this step is to transform "supply risk" into a decision factor of "whether or not to increase the procurement volume".

[0086] S603: Calculate the price safety factor based on historical procurement price data and external price data. in, Indicates the price safety factor. This represents the price risk adjustment coefficient, with values ​​ranging from 0.1. This indicates a price volatility risk index.

[0087] It should be noted that by constructing a price fluctuation risk index based on historical procurement price data and external market price data, and further introducing a price risk adjustment coefficient to calculate a price safety coefficient, the impact of price fluctuations on procurement costs and procurement timing can be transformed from empirical judgment into quantifiable and adjustable parameters. This allows price risk to directly participate in the process of generating procurement plans and correcting procurement quantities, reducing the risk of procurement costs deviating from expectations due to abnormal price fluctuations.

[0088] In one possible implementation, S603 specifically includes: S6031: Calculate the intensity of price fluctuations based on historical procurement price data and external price data.

[0089] Specifically, based on historical procurement price data and corresponding external market price data, multiple price samples within a preset time window are selected and statistically processed. Subsequently, the deviation of each price sample from its average price is calculated, and the deviation values ​​of each price are squared and summed. The summation results are then averaged and squared to obtain the price fluctuation intensity.

[0090] It should be noted that those skilled in the art can set the size of the preset time window according to actual needs, and this invention does not limit this.

[0091] S6032: Calculate the price volatility risk index based on the intensity of price volatility.

[0092] Specifically, the price volatility risk index is obtained by dividing the intensity of price volatility by the average price within the corresponding time window.

[0093] Specifically, the price volatility risk index is limited to a preset range through a mapping function to ensure that the range of the price safety coefficient is controllable.

[0094] S6033: Based on the price volatility risk index, calculate the price safety coefficient through a mapping function.

[0095] Among them, the price safety coefficient is an adjustment parameter used to quantify the impact of market price fluctuation risk on procurement decisions. By analyzing historical procurement prices and external market price data, the price fluctuation range is transformed into a calculable risk level, and further mapped into a coefficient used to correct procurement parameters.

[0096] S604: Based on the supply safety factor and price safety factor, and combined with the cumulative demand variance, the purchase quantity is adjusted to obtain the final purchase quantity: in, Indicates the final purchase quantity. Indicates the basic purchase quantity. This indicates the cumulative variance of demand.

[0097] The basic procurement quantity is used to characterize the demand forecast within the forecast period, without considering supply risk and price fluctuation risk, and is the procurement quantity required after deducting current available inventory and inventory in transit.

[0098] It should be noted that, to avoid generating excessively large procurement quantities under multiple risk correction conditions, a procurement quantity threshold is further set as a constraint when determining the final procurement quantity. After obtaining the procurement quantity calculated based on demand forecasts, supply security factors, and price security factors, the calculated procurement quantity is compared with the preset procurement quantity threshold, and the smaller of the two is selected as the final procurement quantity to be implemented. This ensures that the procurement plan has risk redundancy capabilities while effectively limiting the procurement scale, preventing over-procurement due to the amplification of risks, and improving the safety and controllability of procurement decisions in the actual execution process. The procurement quantity threshold is set by procurement personnel based on experience and actual conditions.

[0099] In this embodiment of the invention, by introducing a supply safety factor and a price safety factor on the basis of the basic purchase quantity to jointly correct the purchase quantity, and further superimposing the demand variance accumulation as an uncertainty compensation term, the supply interruption risk, price fluctuation risk and demand uncertainty can be uniformly incorporated into the purchase quantity calculation process, so that the final determined purchase quantity not only meets the demand forecast results, but also has the redundancy capability to cope with multiple risk factors.

[0100] S605: Decompose the final purchase quantity into a scheme structure regarding suppliers, supplier shares, and supplier delivery dates to generate a preliminary purchase scheme.

[0101] Specifically, after obtaining the final purchase quantity, candidate suppliers are screened based on their historical performance data, price levels, and supply capabilities. The final purchase quantity is then divided according to preset allocation rules. These rules comprehensively consider each supplier's supply capacity ceiling, historical performance stability, and pricing level to allocate corresponding purchase shares to different suppliers, thereby determining the purchase quantity for each supplier. Simultaneously, based on each supplier's historical average delivery cycle and current supply status, reasonable delivery times or lead times are matched to the corresponding purchase quantities. This ultimately forms a scheme structure that includes supplier identification, purchase share, and delivery time, serving as a preliminary procurement plan for subsequent risk verification and execution.

[0102] In this embodiment of the invention, the supply security factor can quantify the instability of supply fulfillment and affect the procurement decision, while the price security factor can correct the cost uncertainty caused by market price fluctuations. Thus, the generated preliminary procurement plan can meet demand forecasting and inventory constraints while taking into account supply continuity, price stability and demand uncertainty. This significantly improves the robustness, feasibility and risk resistance of the procurement plan under complex supply environments and volatile market conditions, and avoids the procurement plan from failing due to supply interruption or abnormal price fluctuations during actual implementation.

[0103] S7: Perform risk constraint verification on the preliminary procurement plan to determine if it passes the verification. If yes, retain the verified preliminary procurement plan as the final procurement plan and proceed to S8. Otherwise, adjust the procurement parameters and return to S6 to regenerate the procurement plan.

[0104] Specifically, after generating a preliminary procurement plan, a risk constraint verification is performed to determine whether the plan meets risk control requirements. This verification includes a comprehensive assessment of the reasonableness of the procurement quantity, the level of supply risk, and price fluctuation risk. If the preliminary procurement plan passes the risk constraint verification, it indicates that the plan is feasible under current demand forecasts and risk conditions, and it is retained as the final procurement plan. If the preliminary procurement plan fails the risk constraint verification, relevant parameters in the plan are adjusted, such as modifying the procurement quantity, supply safety factor, or price safety factor. After parameter adjustments, the procurement plan generation process is repeated to generate a plan that meets the risk constraints, thereby ensuring the final procurement plan has good stability and reliability during actual implementation.

[0105] S8: Convert the final procurement plan into standardized procurement instructions, and automatically send the standardized procurement instructions to the procurement execution system through the system interface.

[0106] In one possible implementation, S8 specifically includes: S801: Extract procurement items from the final procurement plan.

[0107] Specifically, the procurement items include material identification, supplier identification, procurement quantity, agreed unit price, delivery cycle, and planned arrival time.

[0108] Optionally, each procurement item is: in, Indicates the first j One procurement item, Indicates the first j The material identifier corresponding to each purchase item Indicates the first j The supplier identifier corresponding to each procurement item Indicates the first j The quantity of each purchase item corresponds to the purchase quantity. Indicates the first j The agreed unit price for each procurement item. Indicates the first j Delivery cycle for each procurement item Indicates the first j The planned delivery time for each procurement item.

[0109] S802: Map procurement items to a unified data structure according to the procurement instruction template to form standardized procurement instructions.

[0110] It should be noted that a procurement instruction template is a unified descriptive model used to standardize the data structure and field meanings of procurement instructions. It standardizes and encapsulates procurement decision results by pre-defining the key fields that should be included in the procurement instruction, their data types, value rules, and semantic meanings. By adopting procurement instruction templates, it is possible to ensure that different procurement items follow a consistent data format during generation, transmission, and execution, reducing the complexity of system interface integration and data parsing. This provides stable and reliable structured support for the automatic generation, system issuance, and execution feedback of procurement instructions.

[0111] Optionally, the standardized procurement instructions are as follows: in, Indicates the first j Standardized procurement instructions for each procurement item. A unique identifier number representing a procurement order. This indicates the material code corresponding to the purchase order. This indicates the supplier identifier corresponding to the procurement order. This indicates the quantity to be purchased corresponding to the purchase order. This indicates the unit purchase price stipulated in the procurement order. This indicates the delivery time corresponding to the procurement order.

[0112] It should be noted that the delivery cycle and planned arrival time are both reflected in the... middle.

[0113] In this embodiment of the invention, by uniformly encapsulating each procurement item in the final procurement plan into a standardized procurement instruction containing fields such as procurement identifier, material information, supplier information, procurement quantity, procurement price, and delivery time, the invention realizes the structured expression of procurement decision results into executable instructions, enabling procurement information to have a unified data format and clear semantic definition, thereby improving the accuracy and consistency of procurement instructions in the system interface transmission and execution process.

[0114] S803: Calls the standard interface of the procurement execution system to send standardized procurement instructions to the procurement execution system in the form of interface data packets.

[0115] Specifically, through a pre-configured system interface call module, the standardized procurement instruction generated in step S802 is encapsulated into an interface data packet according to the data interaction protocol supported by the procurement execution system. Using synchronous or asynchronous calling methods, the interface data packet is sent to the receiving interface of the procurement execution system. After the interface call is completed, the processing status or receipt information returned by the procurement execution system is obtained and recorded to indicate whether the procurement instruction has been successfully received and entered the execution process. If unsuccessful, the instruction is re-encapsulated and the interface resend is retried, the procurement instruction parameters are adjusted and resent, or the procurement instruction is transferred to the manual confirmation process.

[0116] In this embodiment of the invention, by transforming the procurement items in the final procurement plan into standardized procurement instructions with unified structure and standardized fields, and automatically issuing them to the procurement execution system based on the system interface, the automatic connection between the procurement decision results and the procurement execution process is realized, reducing delays and errors caused by manual intervention. At the same time, through the status feedback and exception handling mechanism for the procurement instruction issuance results, retrying, parameter adjustment, or manual confirmation can be performed in a timely manner when the instruction issuance fails, improving the reliability and stability of the procurement instruction issuance process. This ensures that the procurement plan can enter the execution process efficiently and accurately, enhancing the automation level and closed-loop management capability of the overall procurement process.

Claims

1. A method for intelligent procurement decision-making and automated execution based on demand forecasting, characterized in that, include: S1: Obtain multi-source data related to procurement needs; S2: Based on the multi-source data, construct a demand feature vector set, and use principal component analysis to select key demand feature vectors from the demand feature vector set; S3: Based on the key demand feature vector, predict future procurement demand using a pre-built extreme learning machine demand prediction model, and output the predicted demand amount and demand fluctuation trend within the prediction period. S4: Based on the inventory data in the multi-source data, the demand forecast, and the demand fluctuation trend, dynamically calculate the inventory safety index; S5: Determine whether the inventory safety index is greater than the preset inventory safety index; If so, after the preset time, return to S1 to continue monitoring; Otherwise, proceed to S6; S6: Based on the demand forecast, the demand fluctuation trend, and the inventory data and historical supply-side fulfillment data from the multi-source data, and taking into account both supply security and price security factors, a preliminary procurement plan is generated. S7: Perform risk constraint verification on the preliminary procurement plan and determine whether the preliminary procurement plan passes the verification. If so, retain the preliminary procurement plan that has passed verification as the final procurement plan and proceed to S8; Otherwise, after adjusting the procurement parameters, return to S6 to regenerate the procurement plan; S8: Convert the final procurement plan into a standardized procurement instruction, and automatically send the standardized procurement instruction to the procurement execution system through the system interface.

2. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 1, characterized in that, The multi-source data includes historical sales data, order data, inventory data, historical fulfillment data from the supply side, historical purchase price data, external price data, and supplier delivery data.

3. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 1, characterized in that, S2 specifically includes: S201: Align the historical sales data, order data, inventory data, historical fulfillment data from the supply side, external price data, and supplier delivery data in the multi-source data according to the time dimension to obtain aligned multi-source data; S202: Based on the aligned multi-source data, extract the basic features related to the procurement requirements; S203: Based on the aligned multi-source data, extract time-related temporal features; S204: Based on the historical demand characteristics in the aforementioned basic features, extract the demand change trend characteristics; S205: Normalize the basic features, the time-series features, and the demand change trend features, and then, according to a unified time index, splice and combine the normalized basic features, time-series features, and demand change trend features to construct the demand feature vector set. S206: Using the principal component analysis method, the key demand feature vectors are selected from the set of demand feature vectors.

4. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 1, characterized in that, S3 specifically includes: S301: Normalize the key demand feature vectors at each time point within the prediction period to obtain normalized key demand feature vectors. S302: Based on the normalized key requirement feature vector, construct a kernel similarity vector for each prediction time. S303: Based on the kernel similarity vector, predict future procurement demand using the extreme learning machine demand prediction model, and output the model demand prediction quantity; S304: Perform inverse normalization on the demand forecast of the model to obtain the demand forecast; S305: Calculate the rolling standard deviation of the demand forecast and use the rolling standard deviation as the demand fluctuation trend.

5. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 1, characterized in that, The specific calculation method for the inventory safety index includes: Obtain inventory data at the assessment time from the multi-source data and calculate available inventory; Obtain procurement and supply side parameters, wherein the procurement and supply side parameters include average lead time and target service level coefficient; Based on the demand forecast, calculate the average demand forecast within the average lead time. Based on the aforementioned demand fluctuation trend, calculate the cumulative demand variance within the average lead time. Calculate the safety stock based on the target service level coefficient and the accumulated demand variance; Calculate the reorder point based on the average demand forecast and the safety stock. The inventory safety index is dynamically calculated by combining the available inventory and the reorder point.

6. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 1, characterized in that, S6 specifically includes: S601: Obtain the supply-side historical performance data, historical procurement price data, and external price data from the multi-source data, wherein the supply-side historical performance data includes on-time delivery rate, historical number of interruptions, and average interruption duration; S602: Calculate the supply safety factor based on the on-time delivery rate, the number of historical interruptions, and the average interruption duration; S603: Calculate the price safety factor based on the historical procurement price data and the external price data; S604: Based on the supply safety factor and the price safety factor, and combined with the cumulative demand variance, the purchase quantity is adjusted to obtain the final purchase quantity; S605: Decompose the final purchase quantity into a scheme structure regarding suppliers, supplier shares, and supplier delivery dates to generate the preliminary purchase scheme.

7. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 6, characterized in that, Specifically, S602 includes: S6021: Normalize the historical performance data of the supply side, and calculate the supply interruption risk index based on the normalized historical performance data of the supply side; S6022: Based on the supply disruption risk index, calculate the supply security coefficient using a mapping function.

8. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 6, characterized in that, Specifically, S603 includes: S6031: Calculate the price fluctuation intensity based on the historical procurement price data and the external price data; S6032: Calculate the price volatility risk index based on the aforementioned price volatility intensity; S6033: Based on the price volatility risk index, calculate the price safety coefficient through a mapping function.

9. The intelligent procurement decision-making and automated execution method based on demand forecasting according to claim 1, characterized in that, S8 specifically includes: S801: Extract the procurement items from the final procurement plan; S802: Map the procurement items to a unified data structure according to the procurement instruction template to form the standardized procurement instruction; S803: Call the standard interface of the procurement execution system and send the standardized procurement instruction to the procurement execution system in the form of an interface data packet.

Images