Purchase demand prediction method and system based on big data analysis

Abstract

The invention discloses a purchase demand prediction method and system based on big data analysis, and relates to the technical field of supply chain management and intelligent prediction, and the method comprises the steps: collecting multi-source heterogeneous data, and carrying out the data fusion and preprocessing; based on a causal inference technology, key influence factors of purchase demand changes are analyzed, and a purchase demand prediction model is constructed; dynamically adjusting the prediction model by using real-time data, and generating a real-time or periodic purchase demand prediction result; carrying out data simulation and model optimization aiming at an extreme scene of a purchase demand; and optimizing a supply and demand matching strategy based on a prediction result, and carrying out iterative improvement on the prediction model through a feedback mechanism. Through multi-source heterogeneous data fusion, causal inference, dynamic adjustment and feedback optimization mechanisms, high-precision purchase demand prediction is realized, a supply and demand matching strategy is optimized, the adaptability of the model to market fluctuation and extreme scenes is enhanced, the purchase cost is effectively reduced, and the supply chain management efficiency and stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of supply chain management and intelligent prediction, and specifically provides a procurement demand prediction method and system based on big data analysis. Background Art

[0002] With the increasing complexity of enterprise supply chain management, procurement demand prediction technology based on big data analysis has become an important tool for optimizing enterprise resource allocation. Traditional procurement demand prediction methods usually rely on historical statistical data and use models such as linear regression or time series analysis to predict demand changes to a certain extent. However, with the rapid development of information technology, the acquisition of multi-source heterogeneous data has become more convenient, including historical procurement records, inventory data, supplier information, market dynamics, and social media comments. Integrating these data can provide more comprehensive decision-making support, but traditional technologies face significant challenges in processing such complex data. In addition, to cope with market uncertainties and volatility, modern procurement demand prediction technologies have gradually started to combine emerging technologies such as causal inference, multi-factor analysis, and real-time dynamic adjustment to achieve higher prediction accuracy and stronger adaptability.

[0003] Although existing technologies have improved the procurement demand prediction ability to a certain extent, there are still significant deficiencies in multi-source heterogeneous data fusion, dynamic adjustment models, and extreme scenario processing. First, traditional prediction methods are usually limited to the analysis of structured data and are difficult to effectively process unstructured data (such as social media comments), resulting in the lack of comprehensiveness in prediction results. Second, the model parameters of existing methods are usually static and cannot respond to market demand and environmental changes in real time, making it difficult to meet the needs of modern enterprises for dynamic adjustment and real-time prediction. In addition, extreme scenarios (such as abnormal fluctuations in market supply and demand) often significantly affect procurement decisions, and traditional methods lack effective identification and model optimization mechanisms for extreme scenarios, resulting in poor robustness of prediction models in these situations. In contrast, the present invention proposes a procurement demand prediction method based on causal inference, which can significantly improve the scientificity of prediction results by quantifying the causal weights of key factors; and combines dynamic adjustment and extreme scenario simulation to optimize the real-time adaptability and robustness of the model, thereby achieving high-precision prediction of demand changes and intelligent supply-demand matching. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: existing procurement demand prediction methods have insufficient capabilities in processing multi-source heterogeneous data, weak model dynamic adjustment capabilities, poor extreme scenario processing capabilities, and the problems of how to achieve accurate prediction and intelligent supply-demand matching.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a procurement demand prediction method based on big data analysis, including collecting multi-source heterogeneous data, and performing data fusion and preprocessing;

[0008] Based on causal inference technology, analyzing the key influencing factors of procurement demand changes, and constructing a procurement demand prediction model;

[0009] Using real-time data to dynamically adjust the prediction model to generate real-time or periodic procurement demand prediction results;

[0010] For extreme scenarios of procurement demand, performing data simulation and model optimization;

[0011] Based on the prediction results, optimizing the supply-demand matching strategy, and iteratively improving the prediction model through a feedback mechanism.

[0012] As a preferred solution of the procurement demand prediction method based on big data analysis according to the present invention, wherein: the collecting of multi-source heterogeneous data includes collecting data from historical procurement records, inventory data, supplier information, market dynamic data, and social media comments;

[0013] Performing formatting and tagging processing on the collected data, wherein:

[0014] Historical procurement records include procurement order time, material type, quantity, and price;

[0015] Inventory data includes inventory level, turnover rate, and safety inventory threshold;

[0016] Supplier information includes indicators of supplier historical performance, quotation, and delivery ability;

[0017] Market dynamic data is obtained through web crawler technology or API interface, including market supply-demand fluctuations and price trends;

[0018] Social media comments extract key sentiment information related to products or markets through natural language processing technology.

[0019] As a preferred solution of the procurement demand prediction method based on big data analysis according to the present invention, wherein: the performing of data fusion and preprocessing includes performing formatting processing on data from different data sources, and converting unstructured data into a structured form;

[0020] Performing timestamp alignment on multi-source data to unify the time dimension of the data;

[0021] Using a normalization method to perform standardization processing on the data to eliminate the dimensional difference between different data sources;

[0022] Dimensionality reduction of high-dimensional data is performed based on the principal component analysis method to extract key features;

[0023] Interpolate and fill in the missing data, and use the nearest neighbor interpolation method or collaborative regression algorithm to repair the missing part of the data;

[0024] Use the fuzzy clustering algorithm to fuse multi-source data to generate a unified data set.

[0025] As a preferred solution of the procurement demand prediction method based on big data analysis according to the present invention, wherein: constructing the procurement demand prediction model includes, based on the collected data, constructing a causal graph G=(V, E), where the node set V represents procurement-related variables, and the edge set E represents the causal relationship between variables, and calculating the causal effect through the causal inference formula:

[0026]

[0027] Among them, Y is the dependent variable, X is the independent variable, and Z is the potential confounding variable; P(Y|do(X)) represents the distribution of Y after intervening on X, P(Y|X, Z) represents the conditional probability distribution of Y given X and Z, and P(Z) represents the marginal probability distribution of the potential confounding variable Z;

[0028] By calculating the causal weight w i of each variable, quantifying its contribution to the demand change, and the calculation formula of the causal weight is:

[0029]

[0030] Among them, represents the change rate of the conditional probability of Y with respect to X i , reflecting the causal influence intensity of X i on Y; is the partial derivative operator, and X i is the i-th independent variable;

[0031] Construct a procurement demand prediction model based on causal weights: Based on the selected key factor set {X i}, construct a multi-factor prediction model, and the formula is:

[0032]

[0033] Among them, D t is the predicted demand value at time t, g(X i，t ) = ln(1 + X i，t ) is the non-linear transformation function of the key factor, X i，t is the value of the i-th key factor at time t, and α0 is the regression constant term, αi is the regression coefficient, ∈ t is the error term.

[0034] As a preferred solution of the procurement demand prediction method based on big data analysis according to the present invention, wherein: the dynamic adjustment of the prediction model includes, according to the fluctuation characteristics of real-time data, the regression weight α in the prediction model i is dynamically adjusted, and at the same time, combined with the causal weight w i , the adjustment formula is:

[0035]

[0036] wherein, α i (t) is the weight dynamically adjusted at time t, ΔX i，t = X i，t - X i，t-1 is the change amount of real-time data, and κ is the sensitivity adjustment coefficient, which is used to control the adjustment range of the dynamic weight;

[0037] Substitute the dynamically adjusted regression weight α i (t) and the causal weight w i into the prediction model to generate the real-time procurement demand prediction value, and the prediction formula is:

[0038]

[0039] wherein, is the real-time procurement demand prediction value at time t;

[0040] Within a fixed time interval T, smooth the real-time prediction value to generate a periodic prediction result, and the smoothing formula is:

[0041]

[0042] wherein, is the periodic procurement demand prediction value at time t, k is the time step before the current time point, T is the smoothing window length, is the real-time procurement demand prediction value at time k.

[0043] As a preferred solution of the procurement demand prediction method based on big data analysis according to the present invention, wherein: the data simulation and model optimization include, based on historical data and real-time data, combined with an anomaly detection algorithm, identifying extreme events that will cause large fluctuations in procurement demand, and the formula is:

[0044]

[0045] wherein, H(X t ) is the anomaly detection result, X tis a real-time data vector, μ X and σ X are the historical data mean and standard deviation respectively, and θ is the anomaly threshold factor; is the indicator function, when ||X t -μ X || > 0·σ X is true, it takes the value of 1, otherwise it is 0; ||X t -μ X || represents the Euclidean distance of X t from the mean μ X ;

[0046] In the identified extreme scenarios, use the Monte Carlo simulation method to generate a simulated data set The generation formula of the simulated data is:

[0047]

[0048] where, is the i-th simulated data value at time t, ξ i is Gaussian noise, represents the fluctuation range of factor i; represents the Gaussian distribution, with a mean of 0 and a variance of The generated random values are used to simulate abnormal fluctuations;

[0049] Based on the generated simulated data set Retrain the prediction model to enhance its robustness to extreme scenarios. The optimized model parameters are obtained by minimizing the following objective function:

[0050]

[0051] where, Θ * is the optimized model parameter, representing the set of optimized prediction model parameters; is the loss function, R(Θ) is the regularization term, λ is the regularization coefficient, is the target value of the simulated scenario demand at time t, is the predicted value of the prediction model for the simulated data under the parameter Θ, and m is the total number of time steps of the simulated data;

[0052] Feed the optimized model back into the real-time prediction process to update the dynamic prediction results and improve the prediction accuracy of the model.

[0053] As a preferred solution of the procurement demand prediction method based on big data analysis according to the present invention, wherein: the iterative improvement of the prediction model through the feedback mechanism includes dividing the procurement demand into short-term demand and long-term demand based on the prediction results, and hierarchically managing the supply resources, and respectively matching them to demands with different priorities;

[0054] According to the real-time status of the supply chain, an iterative linear programming algorithm is used to adjust the procurement plan to ensure that key resources are preferentially allocated to high-priority demands while minimizing cost fluctuations;

[0055] Based on historical delivery data and real-time market conditions, combined with a weighted scoring model, automatically screen the optimal suppliers that meet the current demands;

[0056] During the procurement execution process, feedback indicators including delivery time, order completion rate, and inventory volatility are collected in real time;

[0057] Combined with supplier performance data, inventory adjustment data, and market change data, establish a multi-dimensional feedback analysis model to evaluate the deviation between the current prediction result and the actual execution result;

[0058] For scenarios with large prediction deviations, identify the main factors causing the deviations and trigger specific exception handling mechanisms, such as re-correcting the input of the prediction model or adjusting the policy rules;

[0059] Use the feedback data to dynamically optimize the causal weights and non-linear transformation function parameters in the model;

[0060] By periodically introducing new feedback data, use an adaptive regression algorithm to iteratively train the prediction model to continuously improve the prediction ability of the model;

[0061] Apply the iterated model to the simulation scenario, verify the improvement degree of its prediction results in terms of accuracy, stability, and response speed, and record the optimization metrics to support the next round of iteration.

[0062] In a second aspect, an embodiment of the present invention provides a procurement demand prediction system based on big data analysis, including:

[0063] Data acquisition and fusion module: collect multi-source heterogeneous data, and perform data fusion and preprocessing;

[0064] Model construction module: based on causal inference technology, analyze the key influencing factors of procurement demand changes, and construct a procurement demand prediction model;

[0065] Real-time dynamic adjustment module: use real-time data to dynamically adjust the prediction model to generate real-time or periodic procurement demand prediction results;

[0066] Model optimization module: For extreme scenarios of procurement requirements, conduct data simulation and model optimization;

[0067] Supply-demand matching module: Based on the prediction results, optimize the supply-demand matching strategy, and iteratively improve the prediction model through a feedback mechanism.

[0068] Thirdly, an embodiment of the present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0069] Fourthly, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described above are implemented.

[0070] Advantages of the present invention: By collecting real-time operation data in the textile production process and performing multi-source fusion and deep feature extraction, the efficient utilization and standardization of data are achieved; by dynamically adjusting the process parameters of the textile production line, the real-time performance and accuracy of the production process are improved; by performing real-time quality inspection and dynamically adjusting the production line operation based on the inspection results, the product qualification rate and the stability of the production line are significantly improved; by constructing a multi-feature fusion fault prediction model to predict the operating state of the equipment, potential faults are identified in advance, reducing equipment downtime and maintenance costs; by optimizing and adjusting the production line operation mode based on the prediction results, a real-time closed-loop intelligent control system is formed, realizing the full intelligentization, automation, and high-efficiency operation of the textile production line. Description of the Drawings

[0071] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings, where:

[0072] Figure 1 It is the overall flowchart of a procurement demand prediction method based on big data analysis provided by the first embodiment of the present invention;

[0073] Figure 2 It is the flowchart of dynamic adjustment and smoothing processing of real-time procurement demand prediction of a procurement demand prediction method based on big data analysis provided by the first embodiment of the present invention;

[0074] Figure 3 It is the module structure connection diagram of a procurement demand prediction system based on big data analysis provided by the third embodiment of the present invention. Detailed Embodiments

[0075] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0076] Example 1, referring to Figure 1 and Figure 2 , which is an embodiment of the present invention, provides a procurement demand prediction method based on big data analysis, including:

[0077] S1: Collect multi-source heterogeneous data and perform data fusion and preprocessing.

[0078] Collect data from historical procurement records, inventory data, supplier information, market dynamic data, and social media comments;

[0079] Perform formatting and tagging on the collected data, where:

[0080] Historical procurement records include the procurement order time, material type, quantity, and price;

[0081] Inventory data includes inventory level, turnover rate, and safety stock threshold;

[0082] Supplier information includes indicators of supplier historical performance, quotation, and delivery ability;

[0083] Market dynamic data is obtained through web crawler technology or API interfaces and includes market supply and demand fluctuations and price trends;

[0084] Social media comments extract key sentiment information related to products or the market through natural language processing technology.

[0085] It should be noted that the "key sentiment information" of social media comments is data that extracts the sentiment tendency (positive, negative, or neutral) of users towards products through natural language processing technology, reveals potential driving factors for changes in market demand, and provides auxiliary information for accurate prediction.

[0086] Furthermore, by collecting historical purchase records and inventory data and combining them with market dynamics, historical trends of demand and current supply-demand relationships can be revealed, supporting short-term and long-term forecasts. Integrating supplier information with social media comments can capture external driving factors of demand changes (such as user preferences or sudden market events), providing multi-dimensional references for enterprise decision-making. Through safety stock threshold and turnover rate analysis, the inventory level can be dynamically adjusted in combination with the forecast results to avoid situations of insufficient inventory or overstocking. With the help of historical performance and delivery capacity indicators in supplier information, the most suitable suppliers can be selected to ensure the efficiency and reliability of supply-demand matching.

[0087] Format the data from different data sources, converting unstructured data into a structured form. Formatting is the first step in data preprocessing, mainly addressing the issue of inconsistent formats from different data sources. Structured data (such as historical purchase records and inventory data in tabular form) can be directly applied to analysis models, but unstructured data (such as social media comments and market news) needs to be converted into a structured form. Use natural language processing techniques (such as TF-IDF, word embedding models) to convert social media comments into numerical vector forms, extracting core sentiment information and keywords. For market dynamics containing image data, use convolutional neural networks to extract features and convert them into numerical descriptions.

[0088] Furthermore, formatting eliminates semantic and format differences between data sources by standardizing the data form, enabling subsequent analysis to be carried out in a consistent form. By processing unstructured data, the scope of analyzed data is expanded, and the model's responsiveness to market dynamic changes is improved.

[0089] Align the timestamps of multi-source data to unify the time dimension of the data. Multi-source data usually has different collection time intervals (such as inventory data updated daily, while market dynamics may change in real time). Timestamp alignment synchronizes data with different time dimensions to the same timeline through methods such as interpolation, resampling, or time window aggregation. For missing time points, linear interpolation or nearest neighbor interpolation is used to supplement the data. Downsample high-frequency data (such as second-level real-time data) to a lower frequency (such as minute or hour level) to match other data sources.

[0090] It should be noted that timestamp alignment ensures the temporal consistency between data from different sources, enabling the data to be analyzed according to a unified time dimension. This processing is particularly crucial for time series analysis and can significantly improve the prediction accuracy of the model.

[0091] The normalization method is used to standardize the data and eliminate the dimensional differences between different data sources. The dimensional differences of data sources (such as price represented in currency units while inventory level is in quantity) can lead to uneven impacts of different variables on the model results. Normalization ensures a consistent numerical range of variables by converting the data into a dimensionless form.

[0092] Principal component analysis method is used to reduce the dimension of high-dimensional data and extract key features. Multi-source data usually has high dimensionality and redundant information, and directly using it may lead to overfitting of the model or low computational efficiency. Principal component analysis (PCA) projects the high-dimensional data into a low-dimensional space through linear transformation, retaining the main variance information.

[0093] Interpolate and fill in the missing data, and use the nearest neighbor interpolation method or collaborative regression algorithm to repair the missing parts of the data. Missing values are inevitable during the data collection process, and directly deleting the missing data may result in the loss of important information. For example, the following methods can be used to fill in the missing values:

[0094] Nearest neighbor interpolation method: Use the known data points closest to the missing value to fill in the missing value, which is suitable for time series data.

[0095] Collaborative regression algorithm: By training a regression model, use the known data of other variables to predict the missing value, which is suitable for multi-variable data.

[0096] Interpolating and filling repairs the problem of missing data, ensuring the integrity and continuity of the data. In particular, the collaborative regression algorithm can utilize the internal correlation of multi-source data to improve the accuracy of filling.

[0097] Use the fuzzy clustering algorithm to fuse multi-source data and generate a unified data set. The fuzzy clustering algorithm divides the data into multiple fuzzy sets, assigns membership degrees to each data point, reflecting the diversity and uncertainty of the data. The fuzzy clustering algorithm effectively integrates the characteristics of multi-source data, generates a unified data set, and improves the robustness and adaptability of the subsequent analysis model.

[0098] S2: Based on causal inference technology, analyze the key influencing factors of the change in procurement demand and construct a procurement demand prediction model.

[0099] Based on the collected data, construct a causal graph G=(V, E), where the node set V represents procurement-related variables, such as historical procurement records, inventory levels, supplier delivery capabilities, etc. The edge set E represents the causal relationships between variables. For example, the inventory level may directly affect the procurement demand, while the supplier delivery capability may indirectly affect the demand through the inventory. Calculate the causal effect through the causal inference formula:

[0100]

[0101] Among them, Y is the dependent variable (target procurement demand), X is the independent variable (key influencing factor), and Z is the potential confounding variable; P(Y|do(X)) represents the distribution of Y after intervening on X, P(Y|X, Z) represents the conditional probability distribution of Y given X and Z, and P(Z) represents the marginal probability distribution of the potential confounding variable Z. This formula eliminates its interference by marginalizing the potential confounding variable Z, and obtains the pure causal effect of Y when X is intervened (do operation).

[0102] By calculating the causal weight w of each variable i , quantify its contribution to the demand change. The calculation formula of the causal weight is as follows:

[0103]

[0104] Among them, represents the rate of change of the conditional probability of Y with respect to X i , reflecting the causal influence intensity of X i on Y; is the partial derivative operator, and X i is the i-th independent variable. The formula calculates the contribution of the variable X i to the change of the demand Y, and quantifies the influence intensity by taking the derivative, avoiding the limitation that traditional correlation analysis cannot distinguish causality from correlation.

[0105] Construct a procurement demand prediction model based on causal weights: Based on the set of key factors {X i} selected, construct a multi-factor prediction model. The formula is as follows:

[0106]

[0107] Among them, D t is the predicted demand value at time t, g(X i，t ) = ln(1 + X i，t ) is the non-linear transformation function of the key factor, X i，t is the value of the i-th key factor at time t, α0 is the regression constant term, α i is the regression coefficient, and ∈ t is the error term.

[0108] It should be noted that by calculating the true influence of the key factor on the target variable through the causal effect formula and eliminating the interference factors, the reliability of the prediction model is significantly improved. For example, the inventory level may be affected by both price fluctuations and market demand at the same time. Traditional correlation analysis may wrongly attribute the change in inventory level to price fluctuations, while causal inference technology can accurately identify the real driving factors of demand through causal diagrams.

[0109] Through the causal weight w iWith the introduction of , the model parameters are directly associated with the actual impact intensity of factors on demand. When predicting demand, the weight allocation of supplier delivery capacity and inventory level can reflect their true contributions, avoiding unreasonable settings of model parameters.

[0110] Through the non - linear transformation function g(X i，t ), the non - linear association between key factors and demand is captured. For example, the relationship between demand and inventory level may intensify as the inventory approaches the threshold, and the non - linear transformation can effectively reflect this dynamic change.

[0111] S3: Dynamically adjust the prediction model using real - time data to generate real - time or periodic purchase demand prediction results.

[0112] According to the fluctuation characteristics of real - time data, dynamically adjust the regression weight α i in the prediction model, and at the same time combine the causal weight w i . The adjustment formula is:

[0113]

[0114] where α i (t) is the weight dynamically adjusted at time t, ΔX i，t = X i，t - X i，t-1 is the change amount of real - time data, and κ is the sensitivity adjustment coefficient used to control the adjustment amplitude of the dynamic weight. By adjusting the regression weight in real - time, the model can respond to market changes. Especially when short - term demand fluctuates violently, it can dynamically capture the change trend of key variables, thus enhancing the real - time prediction ability of the model.

[0115] Substitute the dynamically adjusted regression weight α i (t) and the causal weight w i into the prediction model to generate the real - time purchase demand prediction value. The prediction formula is:

[0116]

[0117] where is the real - time purchase demand prediction value at time t;

[0118] Within a fixed time interval T, smooth the real - time prediction values to generate periodic prediction results. The smoothing formula is:

[0119]

[0120] where is the periodic purchase demand prediction value at time t, k is the time step before the current time point, T is the smoothing window length, is the predicted value of real-time procurement demand at time k. Periodic smoothing eliminates short-term fluctuations in real-time forecasts, making the forecast results more suitable for long-term procurement planning and resource allocation.

[0121] It should be noted that by combining real-time forecasting and periodic smoothing, not only can short-term dynamic demands be met, but also long-term strategic planning can be supported, making the enterprise's procurement decisions more comprehensive and scientific.

[0122] S4: For extreme scenarios of procurement demand, conduct data simulation and model optimization.

[0123] Based on historical data and real-time data, combined with anomaly detection algorithms, identify extreme events that can cause significant fluctuations in procurement demand. The formula is:

[0124]

[0125] where, H(X t ) is the anomaly detection result, X t is the real-time data vector, μ X and σ X are the mean and standard deviation of historical data respectively, and θ is the anomaly threshold factor; is the indicator function, which takes the value of 1 when ||X t -μ X ||>θ·σ X is true, otherwise it is 0; ||X t -μ X || represents the Euclidean distance between X t and the mean μ X ; by combining the historical mean μ X and the standard deviation σ X , define the range of anomaly events (θ·σ X beyond the normal fluctuation range) to ensure the accurate identification of sudden and abnormal events and avoid misjudging ordinary fluctuations as anomalies.

[0126] In the identified extreme scenarios, use the Monte Carlo simulation method to generate a simulated data set to provide more diverse data for model training and make up for the deficiencies in real data. The formula for generating simulated data is:

[0127]

[0128] where, is the i-th simulated data value at time t, ξ i is Gaussian noise, represents the fluctuation range of factor i; represents the Gaussian distribution, with a mean of 0 and a variance of The generated random values are used to simulate abnormal fluctuations. The simulated data can expand the coverage of the existing dataset, enabling the prediction model to have better robustness in extreme scenarios. The noise of the Gaussian distribution simulates the random fluctuation characteristics in the real scenario, providing more realistic samples for model training.

[0129] Based on the generated simulated dataset Retrain the prediction model, and the optimized model parameters are obtained by minimizing the following objective function:

[0130]

[0131] where, Θ * is the optimized model parameter, representing the set of optimized prediction model parameters; is the loss function, R(Θ) is the regularization term, λ is the regularization coefficient, is the target value of the simulated scenario demand at time t, is the predicted value of the simulated data by the prediction model under the parameter Θ , and m is the total number of time steps of the simulated data. Adding the regularization term R(Θ) to the objective function can control the model complexity and improve the generalization ability in extreme scenarios. The addition of the simulated data enables the model training to cover abnormal situations that do not occur but may occur in the real scenario.

[0132] Feed the optimized model back into the real-time prediction process to update the dynamic prediction results to improve the prediction accuracy of the model.

[0133] S5: Based on the prediction results, optimize the supply-demand matching strategy and iteratively improve the prediction model through the feedback mechanism.

[0134] Based on the prediction results, divide the procurement demand into short-term demand and long-term demand, and hierarchically manage the supply resources, and match them to the demands of different priorities respectively. For example:

[0135] Short-term demand: Based on the real-time market conditions and prediction data, define the procurement demand that needs to be met within a short period (such as daily or weekly). It is mainly used to meet the rapidly changing market demand.

[0136] Long-term demand: Based on historical data and trend prediction results, define the procurement demand within a longer period (such as monthly or quarterly). It is used to guide strategic resource allocation and reserve.

[0137] Short-term demand directly drives the immediate procurement plans of enterprises, while long-term demand provides the basis for the overall planning of the supply chain. By managing them at different levels, the relationship between rapid response and strategic stability can be effectively balanced. High-priority demands are matched with suppliers with high stability and fast response speed to ensure timely satisfaction. Low-priority demands are matched with suppliers with lower costs or longer delivery cycles to reduce cost fluctuations.

[0138] Furthermore, hierarchical management optimizes the resource allocation strategy, ensuring that critical demands are met first while reducing the procurement costs of non-critical demands.

[0139] According to the real-time status of the supply chain, an iterative linear programming algorithm is used to adjust the procurement plan, ensuring that critical resources are preferentially allocated to high-priority demands while minimizing cost fluctuations. Through iterative optimization, this algorithm can adjust the procurement plan in real time, especially in the case of demand fluctuations or changes in supplier capabilities, ensuring the flexibility and robustness of the supply chain.

[0140] Based on historical delivery data and real-time market conditions, a weighted scoring model is combined to automatically screen the optimal suppliers that meet the current demands. Sorting by scores from high to low, high-scoring suppliers are preferentially selected; scores are updated in real time to ensure the dynamic adaptability of supplier selection. During the procurement execution process, feedback metrics including delivery time, order completion rate, and inventory volatility are collected in real time.

[0141] Combining supplier performance data, inventory adjustment data, and market change data, a multi-dimensional feedback analysis model is established to evaluate the deviation between the current prediction results and the actual execution results.

[0142] For scenarios with large prediction deviations, identify the main factors causing the deviations and trigger specific exception handling mechanisms, such as re-correcting the input of the prediction model or adjusting the policy rules. Real-time feedback and exception handling mechanisms can quickly discover and solve problems, ensuring the consistency between procurement demand prediction and actual execution results.

[0143] Using the feedback data, dynamically optimize the causal weights and non-linear transformation function parameters in the model;

[0144] By periodically introducing new feedback data, an adaptive regression algorithm is used to iteratively train the prediction model, continuously improving the prediction ability of the model. Through continuous iterative optimization, the adaptive regression algorithm ensures that the model can maintain high prediction accuracy as the market and supply chain conditions change.

[0145] Apply the iterated model to the simulation scenario to verify the improvement degree of its prediction results in terms of accuracy, stability, and response speed, and record the optimization metrics to support the next round of iteration. Through simulation verification and metric recording, ensure the sustainability and scientific nature of model improvement, providing technical support for long-term optimization.

[0146] Example 2, which is the second embodiment of the present invention, is different from the previous embodiment in that:

[0147] If the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention, in essence, or the part that makes a contribution to the prior art or the part of the current technical solution, can be embodied in the form of a software product. The current computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0148] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0149] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part with one or more wirings (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or processing it in other suitable ways as necessary, and then storing it in a computer memory.

[0150] Example 3, refer to Figure 3, as an embodiment of the present invention, provides a procurement demand prediction system based on big data analysis, including a data collection and fusion module, a model construction module, a real-time dynamic adjustment module, a model optimization module, and a supply-demand matching module.

[0151] Data collection and fusion module: Collect multi-source heterogeneous data, and perform data fusion and preprocessing;

[0152] Model construction module: Based on causal inference technology, analyze the key influencing factors of procurement demand changes, and construct a procurement demand prediction model;

[0153] Real-time dynamic adjustment module: Use real-time data to dynamically adjust the prediction model, and generate real-time or periodic procurement demand prediction results;

[0154] Model optimization module: For extreme scenarios of procurement demand, perform data simulation and model optimization;

[0155] Supply-demand matching module: Based on the prediction results, optimize the supply-demand matching strategy, and iteratively improve the prediction model through a feedback mechanism.

[0156] Example 4, as an embodiment of the present invention, provides a procurement demand prediction method based on big data analysis. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / contrast experiments.

[0157] This experiment aims to verify the practical application effect of the procurement demand prediction method based on multi-source heterogeneous data collection, causal inference analysis, and dynamic adjustment. The experiment focuses on the annual procurement optimization problem of a large manufacturing enterprise, and mainly solves the problems of dynamic market demand prediction, supply chain resource allocation optimization, and emergency response.

[0158] The experiment first collected historical procurement demand, inventory level, market dynamic score, and supplier delivery capacity data from multi-source heterogeneous platforms such as the enterprise's ERP system, supplier database, and market monitoring system. To ensure the availability and consistency of the data, data cleaning, formatting, and normalization processing were performed, including filling in missing data, timestamp alignment, and standardization processing.

[0159] Based on the cleaned data, a procurement demand prediction model was constructed using causal inference technology. Through causal analysis, key factors affecting procurement demand, such as inventory level, market dynamic score, and delivery capacity, were identified, and causal weights were assigned to each factor to quantify its impact on procurement demand. During the prediction process, the dynamic adjustment algorithm enables the model to respond to market changes in real time and generate more accurate prediction results.

[0160] During the experiment, an abnormal scenario simulation technique was also introduced. Extreme market volatility scenarios were generated based on historical data analysis, and the dataset was extended through the Monte Carlo method to improve the model's robustness to abnormal situations. Finally, the prediction results were dynamically optimized through a feedback mechanism, and the model parameters were updated using real-time feedback data (such as order completion rate, inventory changes) to achieve iterative improvement of the prediction.

[0161] For specific experimental data, please refer to Table 1.

[0162] Table 1 Experimental Data Reference Table

[0163]

[0164] It can be seen from the experimental data that the method of the present invention has significant advantages in terms of prediction accuracy, supply-demand matching optimization, and adaptability to extreme scenarios.

[0165] The predicted demand and the optimized procurement demand are closer to the actual situation compared to the historical data. For example, the historical procurement demand of Supplier C was 503 pieces, the predicted demand was 508 pieces, and the optimized procurement demand was 502 pieces, which is consistent with the actual market demand change trend. This shows that the dynamic adjustment algorithm effectively responds to market fluctuations.

[0166] The data shows that Supplier E has the highest delivery capacity score (95 points), so Supplier E is preferentially selected in the allocation of high-priority demands. The inventory level of Supplier B is relatively low (77 pieces), but by allocating an appropriate amount of procurement demand (612 pieces), resource waste is avoided. The hierarchical management strategy effectively balances the efficiency and cost of resource allocation.

[0167] For suppliers with a relatively low market dynamic score (such as Supplier D with a score of 7.0), the optimized procurement demand through abnormal scenario simulation is 527 pieces, which is more reasonable than the predicted value (529 pieces), indicating that the model has strong stability in dealing with extreme scenarios.

[0168] The model parameters were dynamically optimized by collecting indicators such as delivery time and inventory change rate in real time. After adjustment, the optimized procurement demand of Supplier A (538 pieces) is closer to its actual delivery capacity (535 pieces), significantly improving the model's adaptability.

[0169] Traditional procurement prediction techniques rely on a single data source, cannot dynamically respond to real-time market changes, and lack the ability to handle extreme scenarios. The present invention effectively solves the above problems through multi-source heterogeneous data fusion, causal inference, and dynamic optimization techniques, making the prediction more accurate and the optimization more efficient.

[0170] The test results prove that the method of the present invention can significantly improve the accuracy of procurement forecasting and the rationality of supply-demand matching in a complex market environment, providing a scientific basis for enterprises to reduce procurement costs and improve supply chain efficiency. In particular, through the combination of abnormal scenario simulation and feedback mechanism, it provides better adaptability to future market changes.

[0171] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A procurement demand forecasting method based on big data analysis, characterized in that: include: Collect multi-source heterogeneous data and perform data fusion and preprocessing; Based on causal inference technology, analyze the key factors affecting changes in procurement demand and build a procurement demand forecasting model; Use real-time data to dynamically adjust the forecast model and generate real-time or periodic procurement demand forecast results; Conduct data simulation and model optimization for extreme scenarios of procurement needs; Based on the forecast results, the supply and demand matching strategy is optimized, and the forecast model is iteratively improved through the feedback mechanism.

2. The procurement demand forecasting method based on big data analysis according to claim 1, characterized in that: The collecting of multi-source heterogeneous data includes collecting data from historical purchase records, inventory data, supplier information, market dynamics data, and social media comments; The collected data is formatted and labeled, including: Historical purchase records include purchase order time, material type, quantity and price; Inventory data includes inventory levels, turnover rates, and safety stock thresholds; Supplier information includes indicators of supplier historical performance, quotes, and delivery capabilities; Market dynamics data is obtained through web crawler technology or API interface, including market supply and demand fluctuations and price trends; Social media comments are used to extract key sentiment information related to products or markets through natural language processing technology.

3. The procurement demand forecasting method based on big data analysis according to claim 2, characterized in that: The data fusion and preprocessing includes formatting the data from different data sources and converting the unstructured data into a structured form; Align timestamps of multi-source data to unify the time dimension of the data; The normalization method is used to standardize the data and eliminate the dimensional differences between different data sources; Reduce the dimension of high-dimensional data based on principal component analysis and extract key features; Interpolate and fill in missing data, using the nearest neighbor interpolation method or collaborative regression algorithm to repair missing data; Fuzzy clustering algorithm is used to fuse multi-source data to generate a unified data set.

4. The procurement demand forecasting method based on big data analysis according to claim 3 is characterized in that: The construction of the procurement demand forecasting model includes, based on the collected data, constructing a causal graph G=(V, E), wherein the node set V represents procurement-related variables, the edge set E represents the causal relationship between the variables, and calculating the causal effect through the causal inference formula: Where Y is the dependent variable, X is the independent variable, and Z is the potential confounding variable; P(Y|do(X)) represents the distribution of Y after intervention on X, P(Y|X, Z) represents the conditional probability distribution of Y given X and Z, and P(Z) represents the marginal probability distribution of the potential confounding variable Z; By calculating the causal weight w of each variable i , quantify its contribution to demand changes, and the calculation formula of causal weight is: in, represents the conditional probability of Y with respect to X i The rate of change of X i the strength of the causal influence on Y; is the partial derivative operator, X i is the ith independent variable; Construct a procurement demand forecasting model based on causal weights: Based on the selected key factor set {X i }, construct a multi-factor prediction model, the formula is: Among them, D t is the predicted demand value at time t, g(X i，t )=ln(1+X i，t ) is the nonlinear transformation function of the key factor, X i，t is the i-th key factor value at time t, α0 is the regression constant term, α i is the regression coefficient, and ∈t is the error term.

5. The procurement demand forecasting method based on big data analysis according to claim 4, characterized in that: The dynamic adjustment of the prediction model includes adjusting the regression weight α in the prediction model according to the fluctuation characteristics of the real-time data. i Dynamically adjust and combine causal weight w i , the adjustment formula is: Among them, α i (t) is the weight after dynamic adjustment at time t, ΔX i，t =X i，t -X i，t-1 is the real-time data change, κ is the sensitivity adjustment coefficient, which is used to control the adjustment range of the dynamic weight; The dynamically adjusted regression weight α i (t) and causal weight w i Substitute into the forecast model to generate real-time procurement demand forecast value. The forecast formula is: in, is the real-time purchase demand forecast value at time t; In a fixed time interval T, the real-time prediction value is smoothed to generate a periodic prediction result. The smoothing formula is: in, is the forecast value of periodic purchase demand at time t, k is the time step before the current time point, T is the smoothing window length, is the real-time purchase demand forecast value at time k.

6. The method for predicting procurement demand based on big data analysis according to claim 5, characterized in that: The data simulation and model optimization include identifying extreme events that may cause significant fluctuations in purchasing demand based on historical data and real-time data in combination with anomaly detection algorithms. The formula is: Among them, H(X t ) is the abnormal detection result, X t is the real-time data vector, μ X and σ X are the mean and standard deviation of historical data respectively, and θ is the abnormal threshold factor; is the indicator function, when ||X t -μ X ||>θ·σ X If true, the value is 1, otherwise it is 0; ||X t -μ X || represents X t Mean μ X The Euclidean distance of Monte Carlo simulation method is used to generate simulated data sets under the identified extreme scenarios. The formula for generating simulated data is: in, is the ith simulated data value at time t, ξ i is Gaussian noise, represents the fluctuation range of factor i; represents a Gaussian distribution with a mean of 0 and a variance of The generated random values ​​are used to simulate abnormal fluctuations; Based on the generated simulated data set Retrain the prediction model, and the optimized model parameters are obtained by minimizing the following objective function: Among them, Θ * is the optimized model parameter, which represents the optimized prediction model parameter set; l(·) is the loss function, R(Θ) is the regularization term, λ is the regularization coefficient, is the target value of the simulation scenario demand at time t, is the prediction model for the simulated data under parameter Θ The predicted value of , m is the total number of time steps of the simulated data; The optimized model is fed back into the real-time prediction process, and the dynamic prediction results are updated to improve the prediction accuracy of the model.

7. The method for predicting procurement demand based on big data analysis according to claim 6, characterized in that: The iterative improvement of the forecast model through the feedback mechanism includes dividing the procurement demand into short-term demand and long-term demand based on the forecast results, and hierarchically managing the supply resources to match them to demands of different priorities; According to the real-time status of the supply chain, an iterative linear programming algorithm is used to adjust the procurement plan to ensure that key resources are allocated to high-priority needs first while minimizing cost fluctuations; Based on historical delivery data and real-time market conditions, combined with a weighted scoring model, automatically select the best supplier to meet current needs; During the procurement execution process, feedback indicators including delivery time, order completion rate, and inventory fluctuation rate are collected in real time; Combine supplier performance data, inventory adjustment data, and market change data to establish a multi-dimensional feedback analysis model to evaluate the deviation between the current forecast results and the actual execution results; For scenarios with large prediction deviations, identify the main factors causing the deviations and trigger specific exception handling mechanisms, such as re-correcting the prediction model input or adjusting the strategy rules; Using feedback data, the causal weights and nonlinear transformation function parameters in the model are dynamically optimized; By periodically introducing new feedback data and using an adaptive regression algorithm to iteratively train the prediction model, the model's prediction capability is continuously improved; The iterated model is applied to simulation scenarios to verify the degree of improvement in the accuracy, stability, and response speed of its prediction results, and the optimization indicators are recorded to support the next round of iterations.

8. A procurement demand forecasting system based on big data analysis, used to implement the procurement demand forecasting method based on big data analysis as claimed in any one of claims 1 to 7, characterized in that: include: Data acquisition and fusion module: collects multi-source heterogeneous data and performs data fusion and preprocessing; Model building module: Based on causal inference technology, analyze the key influencing factors of changes in procurement demand and build a procurement demand forecasting model; Real-time dynamic adjustment module: Use real-time data to dynamically adjust the forecast model and generate real-time or periodic procurement demand forecast results; Model optimization module: performs data simulation and model optimization for extreme scenarios of procurement needs; Supply and demand matching module: Based on the forecast results, optimize the supply and demand matching strategy and iteratively improve the forecast model through the feedback mechanism.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the procurement demand forecasting method based on big data analysis described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the procurement demand forecasting method based on big data analysis described in any one of claims 1 to 7 are implemented.

Images