A method, system, equipment, and medium for multidimensional data analysis and supply chain collaborative decision-making.

Abstract

This invention discloses a method, system, equipment, and medium for multidimensional data analysis and supply chain collaborative decision-making, belonging to the field of intelligent economic analysis and decision optimization technology for the supply chain. The method includes: collecting multidimensional raw data; preprocessing the multidimensional raw data through regression analysis; extracting latent components of independent and dependent variables; constructing low-dimensional feature representations; purifying the output value of the power grid company through a first-level algorithm; establishing the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term errors; performing regression analysis on the purified independent and dependent variables; decomposing the output value of the power grid company; and generating supply chain collaborative adjustment decision-making suggestions. This invention provides precise support for supply chain collaborative adjustment decisions, overcoming the limitations of existing methods in terms of insufficient flexibility and insufficient consideration of the combined influence of multiple external factors when facing complex electricity demand, thus improving the accuracy and effectiveness of electricity analysis under volatile market conditions.

Description

Technical Field

[0001] This invention relates to the field of intelligent analysis and decision optimization technology for industrial chain economy, specifically to a method, system, equipment, and medium for multi-dimensional data analysis and collaborative decision-making in the industrial chain. Background Technology

[0002] Currently, while time-series-based electricity consumption analysis methods can extract useful information through techniques such as principal component analysis, they still have limitations in addressing the diverse electricity demand patterns faced by complex electricity markets. In particular, existing technologies are unable to effectively capture and respond to sudden fluctuations in the electricity market. Most existing methods focus on traditional regression analysis and principal component extraction, which lack sufficient flexibility and accuracy when dealing with complex dynamic changes and adapting to the volatile environment of the electricity market.

[0003] Furthermore, existing electricity demand analysis methods often fail to adequately consider the combined impact of multiple external factors on electricity demand. Besides factors such as electricity production costs and policy regulation, the impact of external factors like seasonal fluctuations and unforeseen events on electricity demand is also not effectively modeled and predicted. This neglect of these factors results in low accuracy of existing methods when facing complex external changes, potentially leading to significant biases in forecasts and thus failing to provide sufficiently reliable decision-making support. To address this issue, more multidimensional analytical methods must be introduced to handle the various complex factors and unforeseen events in electricity demand forecasting. Summary of the Invention

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

[0005] Therefore, the technical problem solved by this invention is that existing power analysis methods still have limitations in dealing with the complex power market demand, and lack sufficient flexibility and accuracy in dealing with complex dynamic changes and adapting to the ever-changing environment of the power market. They also fail to fully consider the comprehensive impact of multiple external factors on power demand.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for multi-dimensional data analysis and supply chain collaborative decision-making, comprising, Collect multidimensional raw data, preprocess the multidimensional raw data through regression analysis, extract the latent components of independent and dependent variables, and construct low-dimensional feature representations; Based on low-dimensional feature representation, the output value of the power grid company is purified through a first-level algorithm. The relationship between the output value of upstream and downstream enterprises and the correction relationship between long-term cointegration and short-term error is established, and regression analysis is performed on the purified independent and dependent variables. Based on the modified relationship, the output value of the power grid company is decomposed to generate decision-making suggestions for coordinated adjustment of the industrial chain.

[0007] As a preferred embodiment of the multidimensional data analysis and supply chain collaborative decision-making method described in this invention, the steps include: collecting multidimensional raw data, preprocessing the multidimensional raw data through regression analysis, extracting the latent components of independent and dependent variables, and constructing low-dimensional feature representations. Collect multidimensional raw data, preprocess the multidimensional raw data through regression analysis, and obtain the latent component score matrix; Based on the latent component score matrix, the latent components of the independent and dependent variables are extracted to obtain the weight vector on the dependent variable side. Based on the weight vector on the dependent variable side, the independent variable loading vector is calculated, and the multidimensional original data is updated by residualization to construct a low-dimensional feature representation.

[0008] As a preferred embodiment of the multidimensional data analysis and supply chain collaborative decision-making method described in this invention, the step of purifying the power grid company's output value based on low-dimensional feature representation using a first-level algorithm, establishing the relationship between upstream and downstream enterprise output value and long-term cointegration and short-term error correction, and performing regression analysis on the purified independent and dependent variables includes... Based on low-dimensional feature representation, regression coefficients are calculated, target values ​​are predicted based on regression coefficients, and the predicted target values ​​are output. Based on the predicted values, the power grid company's output value is purified through a primary algorithm; Based on the purified output value of the power grid company, we establish the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term error, and conduct regression analysis on the purified independent and dependent variables.

[0009] As a preferred embodiment of the multidimensional data analysis and supply chain collaborative decision-making method described in this invention, the step of decomposing the power grid company's output value based on the correction relationship and generating supply chain collaborative adjustment decision suggestions includes: Based on the modified relationship, an autoregressive distributed lag model is established, and the output value of the power grid company is decomposed using the autoregressive distributed lag model. Based on the decomposed output value of the power grid company, we analyze the asymmetric impact of the fluctuation of the independent variable on the dependent variable. Based on asymmetric effects, we analyze short-term changes in dependent variables and dynamically adjust the balance relationship to generate decision-making suggestions for coordinated adjustment of the industrial chain.

[0010] As a preferred embodiment of the multidimensional data analysis and supply chain collaborative decision-making method described in this invention, the step of calculating the independent variable loading vector and updating the multidimensional original data using residuals to construct a low-dimensional feature representation includes: By performing linear reconstruction on the independent and dependent variables respectively, the explained information is extracted from the data matrix, leaving residual information for the next round of latent component extraction. This clarifies the projection structure of the current latent components in the original independent and dependent variable spaces, expressed as: in, The independent variable is the load vector. The sum of squares of the score vectors. The dependent variable loading vector; Let be the covariance vector of the independent variable, which is an M×1 vector; The dependent variable covariance vector is a P×1 vector. The original data matrix is ​​subjected to deflection processing, subtracting the portion reconstructed from the current components from the original matrix, retaining residual information for use in the next round of component extraction. Explanatory power is sequentially extracted from multiple mutually orthogonal latent components to obtain a low-dimensional representation system. A complete PLSR regression structure is then formed through recursion, expressed as: in, The matrix of independent variables after deflection represents the structural information remaining after removing the portion explained by the first latent component. Let be the dependent variable matrix after deflection, and represent the residuals after removing the portion explained by the first latent component. and These are matrices of independent and dependent variables, respectively. Let X be a first-order reconstruction matrix with dimensions T×M. Let be the first-order reconstruction matrix of Y.

[0011] As a preferred embodiment of the multidimensional data analysis and supply chain collaborative decision-making method described in this invention, the step of purifying the power grid company's output value based on predicted values ​​using a primary algorithm includes: Introduce exogenous instrumental variables using the instrumental variable method. purify independent variables The endogenous part is introduced by introducing exogenous variables through the instrumental variable method. Then, regression analysis can be performed to quantify the independent variables. For dependent variable The effect of regression equation is expressed as: in, Let be the observed value of the dependent variable, and represent the dependent variable at time t. The independent variable represents any variable that may affect the dependent variable. The constant term represents the baseline level of the model. This is the causal effect coefficient. For the direct influence coefficient, This is the error term.

[0012] This invention introduces exogenous variables through the instrumental variable method, removing endogeneity bias caused by bidirectional causal relationships or omitted variables in the independent variables, thereby purely identifying the causal effect of independent variables on the output value of the power grid company. This overcomes the estimation distortion problem caused by the mutual influence of dependent variables in the context of industrial chain collaboration in traditional regression analysis, and improves the causal inference ability of decision-making methods in complex power market environments.

[0013] As a preferred embodiment of the multidimensional data analysis and supply chain collaborative decision-making method described in this invention, the establishment of the autoregressive distributed lag model includes, To capture the short-term dynamic effects of independent variables, the NARDL model expression is: in, As the dependent variable, It represents the upward fluctuation of the independent variable, indicating the amount of change when the independent variable increases. It represents the downward fluctuation of the independent variable, indicating the amount of change when the independent variable decreases. and These are the coefficients for positive and negative fluctuations, respectively, representing the degree of influence of the independent variable on the dependent variable under different fluctuation conditions. This is the error term.

[0014] This invention breaks through the limitation of assuming symmetrical influence in traditional equilibrium analysis by distinguishing and quantifying the upward and downward fluctuations of independent variables. It can not only decompose the composition of power grid output value, but also identify the non-equilibrium effects of positive and negative shocks in short-term fluctuations, thereby enhancing the pertinence and robustness of decision-making recommendations.

[0015] This invention provides a system for multidimensional data analysis and collaborative decision-making across the industrial chain.

[0016] To address the aforementioned technical problems, this invention provides the following technical solution: a system for multi-dimensional data analysis and supply chain collaborative decision-making, comprising: a data acquisition and feature extraction module, a regression analysis module, and an adjustment decision-making module. Collect multidimensional raw data, preprocess the multidimensional raw data through regression analysis, extract the latent components of independent and dependent variables, and construct low-dimensional feature representations; The regression analysis module is based on low-dimensional feature representation. It purifies the output value of the power grid company through a first-level algorithm, establishes the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term error, and performs regression analysis on the purified independent and dependent variables. The adjustment decision module decomposes the power grid company's output value based on the correction relationship and generates industrial chain collaborative adjustment decision suggestions.

[0017] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the method for multidimensional data analysis and supply chain collaborative decision-making.

[0018] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the method for multidimensional data analysis and supply chain collaborative decision-making.

[0019] The beneficial effects of this invention are as follows: By introducing regression analysis preprocessing to construct a low-dimensional feature representation, the latent components of independent and dependent variables are extracted, reducing data dimensionality and complexity; further, based on the low-dimensional features, endogeneity is purified through a first-level algorithm, and the correction relationship between upstream and downstream output value and long-term cointegration and short-term errors is established, enhancing the model's adaptability and interpretive accuracy in dynamic electricity market environments; finally, by using an autoregressive distributed lag model to decompose grid output value and analyze asymmetric impacts, precise support is provided for the coordinated adjustment decision of the industrial chain, solving the limitations of existing methods in terms of insufficient flexibility and insufficient consideration of the comprehensive impact of multiple external factors when facing complex electricity demand, and improving the accuracy and decision-making effectiveness of electricity analysis under volatile market conditions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a method for multidimensional data analysis and supply chain collaborative decision-making, provided as an embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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 protection scope of the present invention.

[0023] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for multidimensional data analysis and supply chain collaborative decision-making, including: To address the limitations of existing power analysis methods in dealing with complex power market demands, which still fail to adequately address diverse power demand patterns, lack sufficient flexibility and accuracy in responding to complex dynamic changes and adapting to the ever-changing power market environment, and fail to fully consider the comprehensive impact of multiple external factors on power demand, this invention provides a method for multidimensional data analysis and industry chain collaborative decision-making.

[0024] S1: Collect multidimensional raw data, preprocess the multidimensional raw data through regression analysis, extract the latent components of independent and dependent variables, and construct low-dimensional feature representations; S2: Based on low-dimensional feature representation, the output value of the power grid company is purified through a first-level algorithm, and the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term error is established. Regression analysis is performed on the purified independent and dependent variables. S3: Based on the correction relationship, the output value of the power grid company is decomposed to generate decision-making suggestions for the coordinated adjustment of the industrial chain.

[0025] Therefore, by preprocessing the data through regression analysis to extract latent components and constructing low-dimensional feature representations to condense core information, the instrumental variable method is then used to purify endogeneity bias and establish a reliable long-term cointegration and short-term error correction relationship between upstream and downstream output. Finally, based on the asymmetric impact decomposition mechanism, the differentiated effects of market shocks in different directions are analyzed, and dynamic adaptive supply chain collaborative adjustment decisions are generated to solve the limitations of traditional methods in dealing with dynamic changes in complex power markets, such as insufficient flexibility and unclear causal inference.

[0026] Example 2, an embodiment of the present invention, provides a method for multi-dimensional data analysis and supply chain collaborative decision-making based on the previous embodiment, including: In this embodiment of the application, step S1 involves collecting multidimensional raw data, preprocessing the multidimensional raw data through regression analysis, extracting the latent components of independent and dependent variables, and constructing a low-dimensional feature representation, including the following steps A1-A3: A1: Collect multidimensional raw data, preprocess the multidimensional raw data through regression analysis, and obtain the latent component score matrix.

[0027] After collecting multidimensional raw data and obtaining the independent and dependent variable matrices, based on the PLSR principle, a direction vector is first found in the independent variable space such that the projected scores along this direction can both represent the main changes in the explanatory variables and have the greatest linear correlation with the dependent variable. This direction vector is called the weight vector on the independent variable side, and the score vector of the first set of latent components can be calculated from it. This score vector assigns a comprehensive score to each period's sample in the time dimension, reflecting the position of that period in the comprehensive economic state most relevant to the dependent variable. While retaining high correlation with changes in output value, the dimensionality of the variables is greatly reduced. This formula transforms the original high-dimensional independent variable matrix into a latent component score matrix. t 1 By maximizing the correlation between the independent and dependent variables through principal components, the expression is as follows: in, Let be the score vector of the first principal component, with dimension T×1, where the t-th element represents the comprehensive score on that latent component in period t. The matrix of independent variables, This is the weight vector of the first principal component, with a dimension of M×1. Each element represents the weight of the corresponding explanatory variable in that component, and can be set manually based on experience.

[0028] A2: Based on the latent component score matrix, extract the latent components of the independent and dependent variables to obtain the weight vector on the dependent variable side.

[0029] To simultaneously characterize the performance of the dependent variable in the direction of this latent component, the score vector As explanatory variables, least squares regression is performed on the dependent variable matrix Y to obtain the regression weight vector on the dependent variable side. The goal of the regression is to find a direction in the dependent variable space such that the projected values ​​in that direction have the best fit with the scores on the independent variable side, thereby aligning the two matrices in the latent component space. This ensures that the extracted latent components take into account both the explanatory independent and dependent variables, obtaining a linear mapping of the dependent variable on the first latent component, expressed as: in, Let be the weight vector on the dependent variable side, with dimension P×1, where each element represents the linear coefficient of each explained variable in the direction of the first latent component. Let be the score vector of the first principal component. Let be a scalar, representing the sum of squares of the score vector; The covariance vector is a P×1 vector that represents the degree of covariance between each dependent variable and the score vector.

[0030] A3: Based on the weight vector on the dependent variable side, calculate the independent variable loading vector, and perform residual updates on the multidimensional original data to construct a low-dimensional feature representation.

[0031] By performing linear reconstruction on the independent and dependent variables respectively, the explained information is extracted from the data matrix, leaving residual information for the next round of latent component extraction. This clarifies the projection structure of the current latent components in the original independent and dependent variable spaces, expressed as: in, The independent variable is the load vector. The sum of squares of the score vectors. The dependent variable loading vector; Let be the covariance vector of the independent variable, which is an M×1 vector; Let be the dependent variable covariance vector, which is a P×1 vector.

[0032] In this embodiment of the application, the construction of low-dimensional feature representation in step S1 is specifically manifested as follows: The original data matrix is ​​subjected to deflection processing, subtracting the portion reconstructed from the current components from the original matrix, retaining residual information for use in the next round of component extraction. Explanatory power is sequentially extracted from multiple mutually orthogonal latent components to obtain a low-dimensional representation system. A complete PLSR regression structure is then formed through recursion, expressed as: in, The matrix of independent variables after deflection represents the structural information remaining after removing the portion explained by the first latent component. Let be the dependent variable matrix after deflection, and represent the residuals after removing the portion explained by the first latent component. and These are matrices of independent and dependent variables, respectively. Let X be a first-order reconstruction matrix with dimensions T×M. Let be the first-order reconstruction matrix of Y.

[0033] In an optional implementation, the construction of the low-dimensional feature representation in step S1 can also employ principal component regression. After standardizing the independent variable matrix X, principal component analysis (PCA) is performed. The first k principal components are extracted. These components are linear combinations of the column vectors of X and are orthogonal to each other. The score matrix of the k principal components is the constructed low-dimensional feature representation.

[0034] In another optional implementation, the construction of the low-dimensional feature representation in step S1 can also employ canonical correlation analysis to find a linear combination, respectively from X and Y, that maximizes the correlation coefficient between the canonical variables. This process can be performed continuously to extract multiple pairs of uncorrelated canonical variables. The canonical variable score U(T × d) on the X side can then be used as the low-dimensional feature representation extracted from X that has the greatest correlation with Y.

[0035] In this embodiment of the application, in step S1, a low-dimensional feature representation is constructed. By successively eliminating the explained information and extracting orthogonal latent components, the data dimensionality is reduced while retaining the core explanatory power, thus overcoming the multicollinearity problem that is common in the original electricity market data.

[0036] In this embodiment of the application, step S2 is based on low-dimensional feature representation, using a first-level algorithm to purify the output value of the power grid company, establishing the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term errors, and performing regression analysis on the purified independent and dependent variables, including the following steps B1-B3: B1: Based on low-dimensional feature representation, calculate regression coefficients, predict the target value based on the regression coefficients, and output the predicted target value.

[0037] Principal component extraction (PCE) yields score matrices for both independent and dependent variables. These score matrices maximize the covariance between the independent and dependent variables in the latent space, establishing a linear relationship between them. The calculation of the regression coefficients requires the use of the loading matrices and score matrices of these two matrices to transform the latent components of the independent variables into predictive effects on the dependent variable. This ensures that the PLSR model not only captures the main changes in the data but also improves the accuracy of dependent variable prediction. The regression coefficient B describes the influence of the latent components on the dependent variable. The process of calculating the regression coefficients is expressed as follows: in, The regression coefficient matrix is ​​the core of the PLSR model. It maps the latent components of the independent variables to changes in the dependent variable. The regression coefficients determine the weight and influence of each latent component in predicting the dependent variable. Let P be the score matrix of the independent variables, which contains the projections of the independent variables into the latent space. Each element represents the score of the independent variable on that latent component. P is the loading matrix of the independent variables, which represents the loading of each independent variable in the direction of the latent component, i.e., how the independent variable is associated with the latent component. Let be the transpose of the loading matrix of the dependent variable, representing the projection of the dependent variable onto the directions of the latent components. Through A matrix maps the latent components of independent variables to the relationships between dependent variables. It is the inverse matrix of the product of the independent variable loading matrix and the score matrix. Its function is to adjust the regression coefficients so that the regression model can more accurately reflect the relationship between the independent and dependent variables.

[0038] By extracting principal components and calculating regression coefficients, the relationship between independent and dependent variables is clarified in the form of latent components. The regression coefficient B, calculated using a formula, represents the influence of the latent components of the independent variables on the changes in the dependent variable. At this point, a simplified latent structure has been extracted from the high-dimensional independent and dependent variables, and regression coefficients have been obtained, providing a theoretical basis for subsequent predictions. In the PLSR model, the final step is to apply the calculated regression coefficient B to new data to predict the value of the dependent variable. The predicted dependent variable Y is calculated using the regression coefficient matrix B combined with the independent variable matrix X. The expression for predicting the dependent variable using known independent variable data and regression coefficients is: in, This is the dependent variable matrix, representing the model's prediction objective. It contains the values ​​of the dependent variable that are desired to be predicted. In regression analysis, the dependent variable typically represents the result or output of interest in this invention. This is the independent variable matrix, containing the input data. Each column represents a feature or variable, and each row represents a feature of a sample. Using these independent variables, the PLSR model will infer the value of the dependent variable. This is the regression coefficient matrix, representing the influence of the latent components of the independent variables on the dependent variable. The error matrix represents the difference between the model's predicted values ​​and the actual observed values. Ideally, the error should be as small as possible, indicating that the model's predictions are relatively accurate. The error terms reflect the parts of the data that the model cannot fully capture, usually caused by noise or model fitting errors.

[0039] B2: Based on the predicted values, the power grid company's output value is purified through a primary algorithm.

[0040] To further analyze causal relationships and asymmetric effects, this invention introduces the IV-LP-NARDL model. The IV-LP model eliminates endogeneity issues by using exogenous instrumental variables, ensuring the accuracy of causal inferences. The NARDL model, on the other hand, can analyze the positive and negative changes of independent variables separately, revealing their short-term and long-term effects on the dependent variable. Through these two steps, the IV-LP-NARDL model can provide more accurate causal relationships and dynamic feedback for each link in the industrial chain, optimizing the decision-making process.

[0041] The Instrumental Variable-Local Projection (IV-LP) model combines the instrumental variable method (IV) and the local projection method (LP). The instrumental variable method addresses endogeneity issues between independent and dependent variables, particularly when bidirectional causal relationships exist or variables are omitted. It can purify the endogeneity of independent variables through exogenous instrumental variables, ensuring the validity of causal inferences. The local projection method, on the other hand, extracts the dynamic effects between variables through regression analysis, helping to quantify the impact of variable changes at various points in time series data on other variables. The IV-LP model effectively identifies causal relationships in economic systems, avoids biases caused by endogeneity, and analyzes short-term dynamic effects, providing relatively accurate estimates of causal relationships for economics and finance research.

[0042] The Nonlinear Autoregressive Distributed Lag (NARDL) model is an extension of the traditional Autoregressive Distributed Lag (ARDL) model, capable of handling the nonlinear relationship between independent and dependent variables. The core advantage of NARDL lies in its ability to classify changes in independent variables into positive and negative fluctuations, analyzing their different impacts on the dependent variable. This method is particularly suitable for economic variables with asymmetric effects, such as oil prices and exchange rates, whose impacts on the economy typically involve different upward and downward effects. NARDL models not only capture the short-term dynamic effects of independent variables but also reveal long-term equilibrium relationships, making them a powerful tool for studying nonlinear economic relationships and dynamic adjustment mechanisms.

[0043] B3: Based on the purified output value of the power grid company, establish the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term error, and conduct regression analysis on the purified independent and dependent variables.

[0044] To further analyze the long-term impact of the independent variable on the dependent variable, it is necessary not only to focus on the current value of the independent variable, but also to introduce the lagged terms of the independent variable. Therefore, the lagged terms of the independent variable are introduced. The lagged terms help analyze how changes in past independent variables affect changes in the current dependent variable. Meanwhile, the introduction of seasonal and policy dummy variables helps eliminate the impact of external fluctuations, thereby enhancing the model's explanatory power and stability.

[0045] In this embodiment of the application, the instrumental variable method is specifically used in step S2 to purify the power grid company's output value through a first-level algorithm: Introducing exogenous instrumental variables using the instrumental variable method purify independent variables The endogenous part is introduced by introducing exogenous variables through the instrumental variable method. Then, regression analysis can be performed to quantify the independent variables. For dependent variable The effect of regression equation is expressed as: in, Let be the observed value of the dependent variable, and represent the dependent variable at time t. The independent variable represents any variable that may affect the dependent variable. The constant term represents the baseline level of the model. This is the causal effect coefficient. The coefficient that has a direct impact can be set manually based on experience. This is the error term.

[0046] In an optional implementation, the purification of the power grid company's output value in step S2 using a first-level algorithm can also employ a difference-in-differences method. The sample is divided into a treatment group (power grid companies / regions affected by the policy) and a control group (power grid companies / regions not affected by the policy). By comparing the differences in the output value changes of upstream and downstream enterprises in the treatment group and the control group before and after the policy impact, the net causal effect of the power grid output value change (caused by the policy) on the upstream and downstream output value is estimated. The time-invariant inter-group differences and the common time trend of the two groups are controlled, thereby purifying endogeneity.

[0047] In another alternative implementation, the purification of the power grid company's output value through the first-level algorithm in step S2 can also adopt a regression breakpoint design. Near the threshold of the operating variable, individuals can be considered to be approximately randomly assigned on both sides of the threshold, because the characteristics of companies right above and below the threshold are very similar. By comparing the differences in the output value of upstream and downstream enterprises on both sides of the threshold, this difference is attributed to the changes in the state of the power grid company determined exogenously by the threshold rule, thereby identifying the local average treatment effect (LATE) of the power grid company's state on the output value of upstream and downstream enterprises.

[0048] In this embodiment of the application, step S2 uses a first-level algorithm to purify the output value of the power grid company, eliminating endogeneity bias caused by bidirectional causal relationships or omitted variables in the independent variables, thereby estimating the true causal effect of the independent variables on the output value of the power grid company, solving the fundamental problem of estimation distortion in traditional regression analysis in the context of industrial chain linkage, and improving the causal inference capability and policy evaluation accuracy of the entire decision-making model in the complex power market environment.

[0049] In this application's implementation, step B3, which involves performing regression analysis on the purified independent and dependent variables, is specifically manifested as follows: To fully capture the lagged effect of the independent variable on the dependent variable, the expression is: in, It is the current value of the dependent variable, representing the dependent variable at time t. The constant term represents the baseline level of the model. It is the lagged term of the independent variable, representing the value of the independent variable in the past i periods, and helps to capture the long-term effect of the independent variable. This is the coefficient of the lagged term, representing the degree of influence of each lagged independent variable on the dependent variable. It can be manually set based on experience. Seasonal adjustment item The regression coefficients, It is a seasonal adjustment term that eliminates the impact of seasonal fluctuations on the regression model. This is the error term, representing the random fluctuations that the model failed to explain.

[0050] In an optional implementation, step B3, which involves regression analysis of the purified independent and dependent variables, can also employ an autoregressive distributed lag model. This model considers not only the historical influence of upstream variables but also the historical values ​​of downstream variables (upstream and downstream output values) to estimate the long-term equilibrium coefficient. Simultaneously, it automatically derives an error correction term to measure the speed and strength of the reverse correction in the current period when the actual relationship deviates from the long-term equilibrium in the previous period, in order to bring the equilibrium back to a normal state. In another alternative implementation, the regression analysis of the purified independent and dependent variables in step B3 can also employ a vector autoregression model, treating the grid output and upstream and downstream output as two equal members within a system. An equation is established for each variable, and the explanatory variables of each equation include the historical values ​​of both variables themselves. When an unexpected shock occurs to the grid output, the upstream and downstream outputs will dynamically respond over several subsequent periods.

[0051] In the embodiments of this application, step B3 performs regression analysis on the purified independent and dependent variables. By introducing lagged terms of the independent variables, the time lag effect and long-term cumulative effect of the economic variables in the industrial chain are captured, overcoming the limitation of static models that only reflect the correlation of the same period, and can more accurately identify the structural long-term cointegration relationship between the independent and dependent variables.

[0052] It should be noted that the instrumental variable method was used to eliminate endogeneity interference and identify the causal effects between upstream and downstream output values; and by incorporating lagged terms and seasonal adjustments, the problems of unclear causal inference and separation of long-term and short-term mechanisms in traditional industrial chain analysis were solved.

[0053] In this embodiment of the application, step S3 involves collecting multidimensional raw data, preprocessing the multidimensional raw data through regression analysis, extracting the latent components of independent and dependent variables, and constructing a low-dimensional feature representation, including the following steps C1-C3: C1: Based on the modified relationship, an autoregressive distributed lag model is established, and the output value of the power grid company is decomposed using the autoregressive distributed lag model.

[0054] Simple linear relationships and lag effects are insufficient to capture the asymmetric effects of independent variable fluctuations in different directions. To more comprehensively analyze the impact of independent variables on the dependent variable, especially in the presence of asymmetric effects, the NARDL model is introduced to distinguish between positive and negative fluctuations of the independent variable. By modeling positive and negative fluctuations of the independent variable separately, the NARDL model can more accurately capture the different impacts of independent variable fluctuations in different directions. This helps to reveal the different economic impacts of market fluctuations, especially during periods of high market volatility, where positive and negative fluctuations may have different effects on the dependent variable. It can not only capture the short-term dynamic effects of the independent variable but also analyze its long-term asymmetric impact on the dependent variable, further revealing the dynamic adjustment mechanisms in the economic system.

[0055] C2: Based on the decomposed output value of the power grid company, analyze the asymmetric impact of the fluctuation of the independent variable on the dependent variable.

[0056] The NARDL model is used to model the positive and negative fluctuations of the independent variable separately, capturing the asymmetric impact of independent variable fluctuations on the dependent variable in different directions. The different economic impacts of market fluctuations are analyzed, especially during periods of high market volatility, where positive and negative fluctuations may have different effects on the dependent variable. However, while this provides the short-term impact of independent variable fluctuations on the dependent variable, further verification is needed to confirm whether a long-term equilibrium relationship exists between the independent and dependent variables, particularly since the impact of independent variable fluctuations on the dependent variable may not fully reflect long-term dynamics. Therefore, a long-term cointegration test will be conducted to further confirm whether a long-term equilibrium relationship exists between the independent and dependent variables, expressed as: in, As the dependent variable, Let Y be the lagged value of the dependent variable Y in the i-th period. Let X be the lagged value of the independent variable X in the i-th period. This is the error correction term, representing the long-term deviation between the independent and dependent variables. It is a constant term, representing the baseline level of the long-term equilibrium relationship. and These are the coefficients of the lagged terms of the independent and dependent variables, respectively.

[0057] By checking the error correction item Whether it is significant is used to determine whether a long-term cointegration relationship exists between the independent and dependent variables. If the deviation is significant and the system exhibits long-term deviation, it indicates that there is a long-term equilibrium relationship between the independent and dependent variables.

[0058] C3: Based on asymmetric effects, analyze the short-term changes of dependent variables and dynamically adjust the balance relationship to generate decision-making suggestions for coordinated adjustment of the industrial chain.

[0059] By testing the significance of the error correction term, we can confirm whether there is a long-term cointegration relationship between variables. However, knowing only whether there is a long-term equilibrium is not enough to guide policy or corporate decisions. Markets fluctuate frequently in the short term, and the system often deviates from the long-term equilibrium. Whether and how quickly the system can return to equilibrium is also related to resource allocation efficiency and risk control. After verifying cointegration, we further construct an error correction model (ECM) to embed the long-term equilibrium into the short-term dynamic equation, which describes the speed and strength of the adjustment of variables in the deviation-correction process.

[0060] in, and These are the differences between the dependent and independent variables, respectively. This is the error correction term, representing the degree of deviation from long-run equilibrium, which helps in understanding how the dependent variable adjusts back to long-run equilibrium. It is the error correction coefficient, representing the speed at which the system reverts to long-term equilibrium. This is the error term, representing random fluctuations that cannot be explained in the short term.

[0061] Short-term fluctuations are captured using a difference method, the degree of deviation from equilibrium in the previous period is quantified using an error correction term, and the strength of the system's pullback to equilibrium is revealed using a correction coefficient.

[0062] Based on the power grid company's output changes (such as growth or decline), specific quantitative recommendations are generated. These recommendations cover aspects such as upstream raw material supply and downstream sales strategies, helping companies optimize supply chain management and market strategies. The outputs at this stage are not limited to forecast data but also provide practical operational suggestions for the company's strategic decision-making, thereby promoting efficient collaboration and market-oriented operational optimization across the industrial chain.

[0063] In this embodiment of the application, the establishment of an autoregressive distributed lag model in step C1 is specifically manifested as follows: To capture the short-term dynamic effects of independent variables, the NARDL model expression is: in, As the dependent variable, It represents the upward fluctuation of the independent variable, indicating the amount of change when the independent variable increases. It represents the downward fluctuation of the independent variable, indicating the amount of change when the independent variable decreases. and These are the coefficients for positive and negative fluctuations, respectively, representing the degree of influence of the independent variable on the dependent variable under different fluctuation conditions. This is the error term.

[0064] In an optional implementation, the autoregressive distributed lag model established in step C1 can also adopt a threshold autoregressive model. Instead of simply decomposing the changes of the independent variable into positive and negative, it divides the entire sample period into two or more state intervals based on whether the independent variable itself exceeds a certain preset or data-driven threshold value. Independent regression equations are estimated for each state interval. By comparing the significance and magnitude of the regression coefficients in different intervals, it is determined whether there is a nonlinear interval effect in the influencing mechanism.

[0065] In another alternative implementation, the autoregressive distributed lag model established in step C1 can also adopt a Markov regime switching model, assuming that the system randomly switches between several unobservable states, each state corresponding to a different set of model parameters. The maximum likelihood estimation simultaneously infers which state is most likely to be in each period and estimates the influence of the independent variable on the dependent variable in each state.

[0066] In this embodiment, step C1 establishes an autoregressive distributed lag model and introduces an asymmetric autoregressive distributed lag model. By deconstructing the changes in independent variables into two states—upward and downward—and modeling them separately, this breaks through the traditional model's assumption of symmetry in the impact of shocks, capturing and quantifying the differentiated transmission effects of external economic factors on the upstream and downstream output value of the industrial chain. This can more realistically reflect the actual impact path under the complex dynamics of the power market and enhance the foresight and pertinence of coordinated adjustment decisions.

[0067] In summary, by constructing low-dimensional features through partial least squares regression, the core latent components of independent and dependent variables are extracted, reducing data dimensionality and overcoming multicollinearity. Furthermore, by combining instrumental variable methods, endogeneity bias in the industrial chain output analysis is eliminated, and long-term cointegration and short-term error correction relationships between upstream and downstream enterprises are established. A nonlinear autoregressive distributed lag model is introduced to achieve a refined decomposition of the power grid company's output value by distinguishing the asymmetric impact of upward and downward fluctuations in independent variables. Finally, the dynamic adjustment speed from short-term deviation to long-term equilibrium is quantified through the error correction model. This addresses the lack of flexibility and accuracy in traditional power demand analysis when dealing with dynamic changes in complex market environments.

[0068] Example 3 is an embodiment of the present invention, which provides a system for multi-dimensional data analysis and supply chain collaborative decision-making, including a data acquisition and feature extraction module, a regression analysis module, and an adjustment decision module. The data acquisition and feature extraction module collects multidimensional raw data, preprocesses the multidimensional raw data through regression analysis, extracts the latent components of independent and dependent variables, and constructs low-dimensional feature representations. The regression analysis module is based on low-dimensional feature representation. It uses a first-level algorithm to purify the output value of the power grid company, establishes the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term error, and performs regression analysis on the purified independent and dependent variables. The adjustment decision module is based on the correction relationship, decomposes the output value of the power grid company, and generates decision suggestions for coordinated adjustment of the industrial chain.

[0069] This embodiment also provides an electronic device applicable to a method for multidimensional data analysis and supply chain collaborative decision-making, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for multidimensional data analysis and supply chain collaborative decision-making as proposed in the above embodiment.

[0070] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements a method for multidimensional data analysis and supply chain collaborative decision-making as proposed in the above embodiments.

[0071] The storage medium proposed in this embodiment and the method for implementing multidimensional data analysis and industrial chain collaborative decision-making proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0072] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0073] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for multidimensional data analysis and supply chain collaborative decision-making, characterized in that: include, Collect multidimensional raw data, preprocess the multidimensional raw data through regression analysis, extract the latent components of independent and dependent variables, and construct low-dimensional feature representations; Based on low-dimensional feature representation, the output value of the power grid company is purified through a first-level algorithm. The relationship between the output value of upstream and downstream enterprises and the correction relationship between long-term cointegration and short-term error is established, and regression analysis is performed on the purified independent and dependent variables. Based on the modified relationship, the output value of the power grid company is decomposed to generate decision-making suggestions for coordinated adjustment of the industrial chain.

2. The method for multidimensional data analysis and supply chain collaborative decision-making as described in claim 1, characterized in that: The process involves collecting multidimensional raw data, preprocessing it through regression analysis to extract latent components of independent and dependent variables, and constructing low-dimensional feature representations. Collect multidimensional raw data, preprocess the multidimensional raw data through regression analysis, and obtain the latent component score matrix; Based on the latent component score matrix, the latent components of the independent and dependent variables are extracted to obtain the weight vector on the dependent variable side. Based on the weight vector on the dependent variable side, the independent variable loading vector is calculated, and the multidimensional original data is updated by residualization to construct a low-dimensional feature representation.

3. The method for multidimensional data analysis and supply chain collaborative decision-making as described in claim 2, characterized in that: The method, based on low-dimensional feature representation, purifies the power grid company's output value using a first-level algorithm, establishes the relationship between upstream and downstream enterprise output value and long-term cointegration and short-term error correction, and performs regression analysis on the purified independent and dependent variables, including... Based on low-dimensional feature representation, regression coefficients are calculated, target values ​​are predicted based on regression coefficients, and the predicted target values ​​are output. Based on the predicted values, the power grid company's output value is purified through a primary algorithm; Based on the purified output value of the power grid company, we establish the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term error, and conduct regression analysis on the purified independent and dependent variables.

4. The method for multidimensional data analysis and supply chain collaborative decision-making as described in claim 3, characterized in that: The process of decomposing the power grid company's output value based on the modified relationship and generating decision-making suggestions for coordinated adjustment of the industrial chain includes: Based on the modified relationship, an autoregressive distributed lag model is established, and the output value of the power grid company is decomposed using the autoregressive distributed lag model. Based on the decomposed output value of the power grid company, we analyze the asymmetric impact of the fluctuation of the independent variable on the dependent variable. Based on asymmetric effects, we analyze short-term changes in dependent variables and dynamically adjust the balance relationship to generate decision-making suggestions for coordinated adjustment of the industrial chain.

5. The method for multidimensional data analysis and supply chain collaborative decision-making as described in claim 4, characterized in that: The process of calculating the independent variable loading vector and updating the multidimensional original data using residuals to construct a low-dimensional feature representation includes: By performing linear reconstruction on the independent and dependent variables respectively, the explained information is extracted from the data matrix, leaving residual information for the next round of latent component extraction. This clarifies the projection structure of the current latent components in the original independent and dependent variable spaces, expressed as: in, The independent variable is the load vector. The sum of squares of the score vectors. The dependent variable loading vector; Let be the covariance vector of the independent variable, which is an M×1 vector; The dependent variable covariance vector is a P×1 vector. The original data matrix is ​​subjected to deflection processing, subtracting the portion reconstructed from the current components from the original matrix, retaining residual information for use in the next round of component extraction. Explanatory power is sequentially extracted from multiple mutually orthogonal latent components to obtain a low-dimensional representation system. A complete PLSR regression structure is then formed through recursion, expressed as: in, The matrix of independent variables after deflection represents the structural information remaining after removing the portion explained by the first latent component. Let be the dependent variable matrix after deflection, and represent the residuals after removing the portion explained by the first latent component. and These are matrices of independent and dependent variables, respectively. Let X be a first-order reconstruction matrix with dimensions T×M. Let be the first-order reconstruction matrix of Y.

6. The method for multidimensional data analysis and supply chain collaborative decision-making as described in claim 5, characterized in that: The process of purifying the power grid company's output value based on predicted values ​​using a primary algorithm includes, Introduce exogenous instrumental variables using the instrumental variable method. purify independent variables The endogenous part is introduced by introducing exogenous variables through the instrumental variable method. Then, regression analysis can be performed to quantify the independent variables. For dependent variable The effect of regression equation is expressed as: in, Let be the observed value of the dependent variable, and represent the dependent variable at time t. The independent variable represents any variable that may affect the dependent variable. The constant term represents the baseline level of the model. This is the causal effect coefficient. For the direct influence coefficient, This is the error term.

7. The method for multidimensional data analysis and supply chain collaborative decision-making as described in claim 6, characterized in that: The establishment of the autoregressive distributed lag model includes, To capture the short-term dynamic effects of independent variables, the NARDL model expression is: in, As the dependent variable, It represents the upward fluctuation of the independent variable, indicating the amount of change when the independent variable increases. It represents the downward fluctuation of the independent variable, indicating the amount of change when the independent variable decreases. and These are the coefficients for positive and negative fluctuations, respectively, representing the degree of influence of the independent variable on the dependent variable under different fluctuation conditions. This is the error term.

8. A system for multidimensional data analysis and supply chain collaborative decision-making, employing the method for multidimensional data analysis and supply chain collaborative decision-making as described in any one of claims 1 to 7, characterized in that, It includes a data acquisition and feature extraction module, a regression analysis module, and an adjustment decision module. The data acquisition and feature extraction module acquires multidimensional raw data, preprocesses the multidimensional raw data through regression analysis, extracts the latent components of independent and dependent variables, and constructs low-dimensional feature representations. The regression analysis module is based on low-dimensional feature representation. It purifies the output value of the power grid company through a first-level algorithm, establishes the correction relationship between the output value of upstream and downstream enterprises and long-term cointegration and short-term error, and performs regression analysis on the purified independent and dependent variables. The adjustment decision module decomposes the power grid company's output value based on the correction relationship and generates industrial chain collaborative adjustment decision suggestions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for multidimensional data analysis and supply chain collaborative decision-making as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for multidimensional data analysis and supply chain collaborative decision-making as described in any one of claims 1 to 7.

Images