Product supply chain comprehensive optimization system and method based on big data

Abstract

The invention belongs to the technical field of intelligent supply chain management and optimization, discloses a product supply chain comprehensive optimization system and method based on big data, and aims to solve the problem that a traditional supply chain management system based on historical data and a static model is difficult to quickly adapt to changes. Forming a real-time evaluation vector; integrating the environment data, constructing an environment feature vector, and training to obtain a risk prediction model in combination with a real-time evaluation vector, thereby obtaining a risk probability of each node in the future, and obtaining a risk evaluation vector; based on the risk assessment vector, in combination with the inventory level and the transportation path, performing joint optimization on the inventory configuration and the distribution path to generate an optimization strategy; updating each node of the supply chain according to the optimization strategy, monitoring the state of the updated supply chain in real time, and generating a cost prediction vector when abnormity is monitored; and taking corrective measures on the supply chain according to the cost prediction vector to ensure stable operation of the supply chain.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent supply chain management and optimization. More specifically, the present invention relates to an integrated optimization system and method for a product supply chain based on big data. Background Art

[0002] In current supply chain management, enterprises rely on a variety of information systems and technologies to optimize operations, but there are some limitations in the existing technologies. Traditional supply chain management systems usually include ERP (Enterprise Resource Planning), CRM (Customer Relationship Management), and simple Internet of Things (IoT) device applications. Although these systems provide effective support in their respective fields, they still have deficiencies in terms of integration, real-time performance, and intelligence.

[0003] Existing supply chain management systems generally face challenges such as data islands and response lags, rely on static and local optimization methods, and it is difficult to achieve comprehensive integration and dynamic adaptation, resulting in inaccurate risk prediction, untimely exception response, and a lack of effective real-time warning and closed-loop management mechanisms, thus affecting the overall operation efficiency and cost control. Summary of the Invention

[0004] To overcome the above-mentioned defects of the existing technology and to achieve the above object, the present invention provides the following technical solutions: An integrated optimization system for a product supply chain based on big data, comprising:

[0005] An intelligent evaluation module: used to obtain market changes, customer demand fluctuations, and internal operation data, conduct intelligent evaluation, and form a real-time evaluation vector;

[0006] A dynamic risk prediction module: used to integrate weather forecasts and traffic conditions, analyze their impacts on the supply chain, and construct an environmental feature vector; combine the real-time evaluation vector and the environmental feature vector, train to obtain a risk prediction model, predict the risk probabilities of each node within a future period based on the risk prediction module, and splice the risk probability levels of all nodes into a risk assessment vector;

[0007] A strategy optimization module: based on the risk assessment vector, in combination with the current inventory level and the existing transportation routes, jointly optimize the inventory configuration and distribution routes, and generate an optimization strategy;

[0008] A real-time monitoring and exception response module: used to update the inventory configuration and transportation route arrangements of each node in the supply chain according to the optimization strategy, real-time monitor the updated supply chain status, when an exception is detected, generate an exception detection vector, and generate a cost prediction vector according to the exception detection vector;

[0009] A correction module: used to take corrective measures on the supply chain according to the cost prediction vector to ensure the stable operation of the supply chain.

[0010] Furthermore, the method for forming the real-time evaluation vector includes:

[0011] The market changes include the industry trend index and the economic health score; the customer demand fluctuations include the customer satisfaction score and the order fulfillment rate; the internal operation data includes the inventory turnover rate, the average transportation time, and the production efficiency score.

[0012] Based on the industry trend index and the economic health score, generate a market sensitivity index; based on the customer satisfaction score and the order fulfillment rate, generate a customer satisfaction index; based on the inventory turnover rate, the average transportation time, and the production efficiency score, generate an operation efficiency index.

[0013] Horizontally concatenate the market sensitivity index, the customer satisfaction index, and the operation efficiency index to form a real-time evaluation vector.

[0014] Furthermore, the method for generating the market sensitivity index based on the industry trend index and the economic health score includes:

[0015] Obtain the sales growth rate, the number of new product launches, and the market share change in the t period from an industry analysis agency. Discretize the t period into ct time points, and use the weighted average method to calculate the sales growth rate, the number of new product launches, and the market share change at the last time point within the t period as the industry trend index for the t period.

[0016] Obtain the macroeconomic indicators for the t period and the previous cv periods from the National Bureau of Statistics, including the GDP growth rate, the inflation rate, the unemployment rate, the CCI index, and the PMI index, and perform standardization processing on all macroeconomic indicators.

[0017] According to the standardized macroeconomic indicators, construct an original data matrix with each macroeconomic indicator as a column and the corresponding values at each time point as rows.

[0018] According to the original data matrix, perform analysis using the principal component analysis method. Obtain XY structural quantities and the corresponding latent structures according to the Kaiser criterion method. Construct an unrotated load matrix with the latent structures as rows, the macroeconomic indicators as columns, and the relationship strength of the macroeconomic indicators on the corresponding latent structures as the values at each position.

[0019] According to the unrotated load matrix, rotate the unrotated load matrix using the orthogonal rotation method to obtain a load matrix.

[0020] Multiply the load matrix by the original data matrix to obtain a structure score matrix. Among them, each row in the obtained structure score matrix represents the corresponding time point in the original data matrix, and each position represents the score of the corresponding macroeconomic indicator on the corresponding latent structure.

[0021] According to the structural score matrix, the weighted average method is used to calculate all potential structural scores at each time point, and the economic health score at each time point is obtained. The economic health score of the last time point in the t period is used as the economic health score of the t period;

[0022] Standardize the industry trend index and economic health score of the t period and the previous cv periods, and calculate the weights of the standardized industry trend index and economic health score respectively;

[0023]

[0024] Among them, QZ JJ represents the weight of the economic health score, QZ HY represents the weight of the industry trend index, JJ t represents the standardized economic health score of the t period, JJ 1 represents the average value of the economic health scores of the t period and the previous cv periods after standardization, HY t represents the standardized industry trend index of the t period, HY 1 represents the average value of the industry trend indexes of the t period and the previous cv periods;

[0025] According to the industry trend index, economic health score, and the corresponding weights, the market sensitivity index is calculated using the weighted summation formula.

[0026] Furthermore, the method for generating the customer satisfaction index based on the customer satisfaction score and order fulfillment rate includes:

[0027] Obtain the customer scores from the CRM system, and take the average of all customer scores within the t period as the customer satisfaction score of the t period;

[0028] Obtain the number of orders completed on time and the total number of orders from the ERP system, and take the ratio of the number of orders completed on time to the total number of orders within the t period as the order fulfillment rate of the t period;

[0029] Standardize the customer satisfaction scores and order fulfillment rates of the t period and the previous cv periods to the range [0, 1], and define the maximum value 1 within the range [0, 1] as the theoretical maximum value of the standardized customer satisfaction scores and order fulfillment rates;

[0030] For the standardized customer satisfaction scores and order fulfillment rates, calculate the correlation coefficients with the reference sequence respectively

[0031] Among them, hg represents the correlation coefficient of the customer satisfaction score or order fulfillment rate, bl cv+t,pkDenote the pk-th value among the standardized values of the customer satisfaction score or order fulfillment rate within cv time periods up to and including time period t. bl represents the standardized value of the customer satisfaction score or order fulfillment rate at time period t, min cv+t |bl cv+t -1| represents the minimum absolute difference between the customer satisfaction score or order fulfillment rate within cv time periods up to and including time period t and the theoretical maximum value, max cv+t |bl cv+t,pk -1| represents the maximum absolute difference between the customer satisfaction score or order fulfillment rate within cv time periods up to and including time period t and the theoretical maximum value;

[0032] Multiply the standardized customer satisfaction score at time period t by the correlation coefficient of the customer satisfaction score, and multiply the order fulfillment rate by the correlation coefficient of the order fulfillment rate, then sum them up to obtain the customer satisfaction indicator;

[0033] The method for generating the operation efficiency indicator based on the inventory turnover rate, average transportation time, and production efficiency score includes:

[0034] Extract the cost of goods sold and inventory value data from the ERP system. Take the ratio of the cost of goods sold to the average inventory value within time period t as the inventory turnover rate; take the average of the inventory value at the first time point and the inventory value at the last time point within time period t as the average inventory value;

[0035] Obtain the transportation time data from the logistics system. Take the average of all transportation times within time period t as the average transportation time of time period t;

[0036] Obtain the output speed and equipment operation status of the production line from the IoT device. Calculate the ratio of the actual output quantity within time period t to the theoretical maximum output quantity of the equipment, and multiply it by the equipment availability rate to obtain the production efficiency score of time period t; take the ratio of the actual operation time of the equipment to the planned operation time within time period t as the equipment availability rate;

[0037] Standardize the inventory turnover rate, average transportation time, and production efficiency score. Define the ideal solution of the standardized inventory turnover rate and production efficiency score as 1, the negative ideal solution as 0, the ideal solution of the average transportation time as 0, and the negative ideal solution as 1;

[0038] Calculate the distances between the standardized inventory turnover rate, average transportation time, and production efficiency score and the ideal solution and negative ideal solution;

[0039]

[0040] Among them, d + represents the distance from the inventory turnover rate kcz, average transportation time scx, and production efficiency score yss to the ideal solution, d -Denote the distances of kcz, scx, and yss to the negative ideal solution;

[0041] For d - and d + Sum them up to get the total distance, and take the ratio of d - to the total distance as the operation efficiency index.

[0042] Furthermore, the method for constructing the environmental feature vector includes:

[0043] Use GIS technology to map each node and transportation route in the supply chain onto a map, where the nodes are key locations or facilities in the complete supply chain from the supplier to the end consumer, including but not limited to suppliers, manufacturing plants, warehouses, and distribution centers;

[0044] Obtain the weather forecast data and traffic condition data of the regions where all nodes and transportation routes are located in real time from the China Meteorological Administration and third-party navigation platforms, and convert them into a structured form through structured processing, where the structured processing includes field extraction, standardization, and timestamp alignment;

[0045] Field extraction is to identify and extract key information, including temperature, precipitation, wind speed, traffic congestion index, and environmental warning information; environmental warning information includes meteorological risk warning, geological risk warning, biological risk warning, invasive pest, and road closure warning;

[0046] Standardization is to unify the units and formats of data from different sources; timestamp alignment is to make the timestamps of all data consistent and use interpolation method to fill in the missing time points;

[0047] Take the temperature, precipitation, and wind speed obtained after structured processing as environmental factors, and integrate them with environmental warning information and traffic congestion index into an environmental dataset;

[0048] According to the environmental dataset, for each node, count the frequency of environmental warning information appearing at the location of each node within cv time periods before the t time period

[0049] where II(RV(v)) represents the indicator function, taking 1 when there is any type of environmental warning in the v time period, and taking 0 if there is no environmental warning of any type in the v time period, and YJP represents the frequency of environmental warning information at the location of the node within cv time periods before the t time period;

[0050] According to the environmental factors in the environmental dataset, use the weighted summation method to calculate the severe weather index at the location of each node within the t time period;

[0051] Combined with the environmental warning information frequency and the severe weather index, the meteorological exposure degree value of each node at time t is calculated using the weighted summation method;

[0052] For each node, the meteorological exposure degree value within time t is concatenated with the environmental factors, traffic congestion index, and all environmental warning information levels in the environmental dataset to obtain the environmental feature vector of each node at time t;

[0053] Among them, in the multi-dimensional feature vector, if there is any kind of environmental warning information for a node at time t, the corresponding environmental warning information is taken as 1, and for the non-existent environmental warning information, it is taken as 0.

[0054] Furthermore, the training method of the risk prediction model includes:

[0055] A1: Horizontally concatenate the real-time evaluation vector and the environmental feature vector to form a comprehensive feature vector;

[0056] A2: Use the comprehensive feature vectors of all nodes as the input and the risk probabilities of each node within the next time period as the output to construct a risk prediction model, and the risk prediction model is a CatBoost model;

[0057] A3: Collect a sample set, including pp groups of samples. Each group of samples includes the comprehensive feature vectors of all nodes in a time period and the label corresponding to each node;

[0058] Among them, the label is whether a risk event has occurred at the corresponding node in the corresponding time period. If a risk event has occurred, the sample belongs to the positive class and the label is taken as 1. If no risk event has occurred, the sample belongs to the negative class and the label is taken as 0. The risk events include transportation delays, supply chain node interruptions, and market demand fluctuations;

[0059] A4: Divide the sample set into a training set and a validation set according to a certain proportion;

[0060] A5: Initialize the hyperparameters of the CatBoost model, use Bayesian optimization for hyperparameter tuning, use k-fold cross-validation to evaluate the cross-validation scores of the model under different hyperparameter combinations, and select the optimal parameter combination;

[0061] A6: Take the optimal parameter combination as the initial parameters of the model, and define the logarithmic loss function as the loss function for evaluating the prediction accuracy of the model Among them, NZ represents the total number of samples, bq uv represents the label of the uv-th sample, pg uv represents the predicted value of the risk probability that the uv-th sample belongs to the positive class;

[0062] A7: Adjust the model parameters according to the gradient using the training set and the loss function in each iteration of the model;

[0063] For each iteration, update the model parameters using the training set, use AUC as the evaluation metric, and calculate the AUC value on the validation set;

[0064] According to the AUC value on the validation set, calculate the difference between the AUC value after the current iteration and the AUC value of the previous iteration, denoted as the iteration difference;

[0065] A8: Set an iteration difference threshold. If the iteration difference is greater than the iteration difference threshold, it is determined that the performance of the model has improved;

[0066] If the iteration difference is less than or equal to the iteration difference threshold, it is determined that the performance of the model has not improved;

[0067] If the performance of the model does not improve on the validation set in consecutive DD iterations, stop training to obtain the trained risk prediction model.

[0068] Furthermore, the method of jointly optimizing the inventory configuration and the distribution path according to the risk assessment vector, combined with the current inventory level and the existing transportation path, to generate an optimization strategy includes:

[0069] Obtain the inventory quantity, inventory holding cost, and product demand of each node at the current time period, as well as the existing transportation path arrangement through the enterprise's financial system and ERP system. The existing transportation path arrangement includes the existing transportation path, vehicle scheduling for each transportation path, total amount of goods transported, transportation cost, and transportation time;

[0070] Define the objective function as maximizing the overall efficiency of the supply chain while minimizing the inventory cost and transportation cost: min∑ i cb i kc i +∑ j ysc j ysl j ;

[0071] Among them, cb i represents the inventory holding cost of node i, kc i represents the inventory quantity of node i, ysc j represents the transportation cost of transportation path j, and ysl j represents the total amount of goods transported through transportation path j;

[0072] Introduce the constraint condition that the inventory quantity of each node is restricted to be greater than or equal to the minimum safety inventory level and less than the inventory warehouse capacity;

[0073] Based on the objective function and constraints, use an integer programming solver to obtain the optimal inventory configuration;

[0074] Based on the risk assessment vector, use a reinforcement learning algorithm to dynamically optimize the delivery route, including:

[0075] Define the state space, where the state includes the inventory level at each node in the current period, the existing transportation route arrangement, the risk assessment vector, and the product demand;

[0076] Define the action space, including selecting the optimal route from the existing transportation routes, changing the number of transportation vehicles, and adjusting the priorities of different transportation tasks;

[0077] Define the reward function to encourage fast and efficient delivery while punishing delays and unnecessary transportation costs JLH = α×DDS - β×YSC - γ×YCD;

[0078] Where, JLH represents the reward, DDS represents the number of orders completed on time, YSC represents the total transportation cost on all transportation routes, YCD represents the number of orders not completed on time, the difference between the total number of orders and the number of orders completed on time is used as the number of orders not completed on time, α, β, and γ represent weight parameters, and the best combination of α, β, and γ is found using hyperparameter optimization techniques;

[0079] Collect the historical dataset of the reinforcement learning model, including the inventory records, transportation route arrangements, risk probability values, and product demands of each node in the PC group;

[0080] Use the historical dataset and the simulation environment to train the reinforcement learning model to obtain a reinforcement learning model that selects the optimal delivery route in different states;

[0081] Based on the trained reinforcement learning model, according to the risk assessment vector in the current period, combined with the inventory levels, inventory holding costs, and product demands of each node, as well as the existing transportation route arrangement, output the optimal delivery route for the current period;

[0082] Combine the optimal inventory configuration and the optimal delivery route to form a comprehensive optimization strategy.

[0083] Furthermore, the method of generating the anomaly detection vector and generating the cost prediction vector according to the anomaly detection vector includes:

[0084] According to the optimization strategy, update the inventory configuration and transportation route arrangement of each node;

[0085] Define a set of anomaly detection rules for the supply chain state, including inventory level anomalies, transportation delays, and environmental factor anomalies;

[0086] In the current period, when the inventory quantity is less than the minimum safety inventory capacity or greater than or equal to the maximum inventory capacity, it is determined to be an abnormal inventory level;

[0087] When the transport vehicle fails to arrive at or leave the designated node on time, it is considered a transport delay;

[0088] When the environment and traffic have an impact on transportation, the environmental factors are judged to be abnormal;

[0089] Use the monitoring system to monitor the supply chain status in real time. When a situation that exists within the anomaly detection rule is detected, an anomaly detection vector is generated, including the anomaly type, the affected node, and the timestamp.

[0090] Define the categories of abnormal costs based on anomaly detection rules, including additional inventory costs caused by abnormal inventory levels; delayed delivery costs caused by transportation delays; and increased transportation costs caused by abnormal environmental factors;

[0091] For each category of abnormal cost, estimate the abnormal cost for each category, including:

[0092] According to the abnormal inventory level, when the inventory level is greater than or equal to the maximum inventory capacity, the difference between the inventory level and the maximum inventory capacity is multiplied by the unit inventory holding cost to obtain the additional inventory cost;

[0093] According to the transportation delay, all affected orders are counted to obtain the total late delivery penalty as the late delivery cost;

[0094] According to environmental factors, the difference between the unit transportation cost of the new transportation route used for substitution and the unit transportation cost of the blocked original transportation route is calculated, and multiplied by the transportation volume on the original transportation route to obtain the transportation cost of an alternative route. The transportation costs of all alternative routes are counted to obtain the total transportation cost increase;

[0095] All estimated abnormal cost values ​​are aggregated to form a cost prediction vector.

[0096] Furthermore, the method of taking corrective measures on the supply chain according to the cost forecast vector includes:

[0097] Set a cost acceptance threshold. If the sum of abnormal costs in the cost prediction vector is less than the cost acceptance threshold, the loss cost is considered to be within an acceptable range and no corrective measures are required.

[0098] If the sum of abnormal costs in the cost prediction vector is greater than or equal to the cost acceptance threshold, the loss cost is considered unacceptable and corrective measures need to be taken;

[0099] For situations where corrective actions are required, the corrective actions taken include:

[0100] Adjust inventory levels: If the inventory level of a node is less than the minimum safety inventory capacity, transfer inventory from other nodes with sufficient inventory to this node. If the inventory level of a node is greater than or equal to the maximum safety inventory capacity, transfer the inventory of this node to other nodes with low inventory levels;

[0101] Re-plan transportation routes: Use real-time traffic data and navigation systems to dynamically adjust transportation routes and avoid congested sections. If there are multiple routes from the current node to the destination, use the shortest path algorithm to select the optimal route and use the optimal route to reduce transportation time and costs;

[0102] Temporarily increase transportation capacity: Deploy additional vehicles into the transportation route and increase the number of transportation shifts to ensure timely delivery of goods;

[0103] According to the corrective measures taken, re-evaluate the risk assessment vector. Based on the new risk assessment vector, adjust the inventory levels of each node, re-plan the transportation routes, and apply the new inventory levels and transportation routes.

[0104] Furthermore, the integrated optimization method for a product supply chain based on big data is characterized by including:

[0105] Obtain market changes, customer demand fluctuations, and internal operation data, conduct intelligent evaluation, and form a real-time evaluation vector;

[0106] Integrate weather forecasts and traffic conditions, analyze their impacts on the supply chain, and construct an environmental feature vector; Combine the real-time evaluation vector and the environmental feature vector, train to obtain a risk prediction model, predict the risk probabilities of each node within a future period based on the risk prediction module, and splice the risk probability levels of all nodes into a risk assessment vector;

[0107] Based on the risk assessment vector, combine the current inventory levels and existing transportation routes, jointly optimize the inventory configuration and distribution routes, and generate an optimization strategy;

[0108] Update the inventory configuration and transportation route arrangements of each node within the supply chain according to the optimization strategy, monitor the status of the updated supply chain in real time, when an anomaly is detected, generate an anomaly detection vector, and generate a cost prediction vector based on the anomaly detection vector;

[0109] Take corrective measures for the supply chain according to the cost prediction vector to ensure the stable operation of the supply chain.

[0110] The technical effects and advantages of the integrated optimization system and method for a product supply chain based on big data of the present invention:

[0111] The present invention integrates multi-source data (such as market changes, customer needs, internal operations, etc.) to form a real-time evaluation vector, comprehensively reflecting the actual state of each link in the supply chain and eliminating the problem of data islands. Secondly, by combining external factors such as weather forecasts and traffic conditions, an environmental feature vector is constructed and combined with the real-time evaluation vector to train a risk prediction model, generating a risk probability vector for each node within a future period, improving the accuracy and timeliness of risk prediction, helping enterprises identify potential risk points in advance, and enhancing the risk resistance ability of the supply chain. Then, based on the risk assessment vector, the inventory configuration and transportation routes are jointly optimized to generate a global optimal solution, achieving the global optimization and flexible adjustment of the supply chain, reducing unnecessary costs, and improving the overall operation efficiency and economic benefits. Next, the supply chain status is continuously monitored, anomalies are detected in real time and a cost prediction vector is generated, establishing a closed-loop management mechanism from monitoring to response to ensure that all anomalies can be processed in a timely and effective manner, preventing problems from expanding, and reducing the impact of anomalies on the supply chain. Finally, according to the cost prediction vector, precise corrective measures are taken to ensure the stable operation of the supply chain, maximize the resource utilization efficiency, reduce the operating costs, and be able to proactively prevent the occurrence of potential problems, enhancing the stability and reliability of the supply chain and ensuring instant response and high-precision anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] Figure 1 is a schematic diagram of the comprehensive optimization system for the product supply chain based on big data according to the present invention;

[0113] Figure 2 is a schematic diagram of the comprehensive optimization method for the product supply chain based on big data according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0114] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the 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 protection scope of the present invention.

[0115] Embodiment 1

[0116] Please refer to Figure 1 as shown, the comprehensive optimization system for the product supply chain based on big data in this embodiment includes:

[0117] An intelligent evaluation module: used to obtain market changes, customer demand fluctuations, and internal operation data, perform intelligent evaluation, and form a real-time evaluation vector;

[0118] Dynamic Risk Prediction Module: It is used to integrate weather forecasts and traffic conditions, analyze their impacts on the supply chain, and construct an environmental feature vector; combine the real-time evaluation vector and the environmental feature vector, train to obtain a risk prediction model, predict the risk probabilities of each node within a future time period based on the risk prediction module, and splice the risk probability levels of all nodes into a risk assessment vector;

[0119] Strategy Optimization Module: Based on the risk assessment vector, combined with the current inventory level and existing transportation routes, jointly optimize the inventory configuration and distribution routes to generate an optimization strategy;

[0120] Real-time Monitoring and Exception Response Module: It is used to update the inventory configuration and transportation route arrangements of each node within the supply chain according to the optimization strategy, real-time monitor the updated supply chain status, generate an exception detection vector when an exception is detected, and generate a cost prediction vector according to the exception detection vector;

[0121] Correction Module: It is used to take corrective measures on the supply chain according to the cost prediction vector to ensure the stable operation of the supply chain;

[0122] The method for forming the real-time evaluation vector includes:

[0123] Market changes include industry trend index and economic health score; customer demand fluctuations include customer satisfaction score and order fulfillment rate; internal operation data includes inventory turnover rate, average transportation time, and production efficiency score;

[0124] Based on the industry trend index and economic health score, generate a market sensitivity index; based on the customer satisfaction score and order fulfillment rate, generate a customer satisfaction index; based on the inventory turnover rate, average transportation time, and production efficiency score, generate an operation efficiency index;

[0125] Horizontally splice the market sensitivity index, customer satisfaction index, and operation efficiency index to form a real-time evaluation vector;

[0126] The method for generating the market sensitivity index based on the industry trend index and economic health score includes:

[0127] Obtain the sales growth rate, number of new product launches, and market share changes in the t period from an industry analysis agency. Discretize the t period into ct time points (the start time of the t period is the first time point, and the time point before the first time point of the next period is the last time point of the t period). Use the weighted average method to calculate the sales growth rate, number of new product launches, and market share changes at the last time point within the t period as the industry trend index of the t period;

[0128] The method for calculating the industry trend by the weighted average method includes:

[0129] Among them, HYQ represents the industry trend index, HYS represents 3, that is, three categories: sales growth rate, number of new product releases, and change in market share, YS represents the index, and HY YS represents the preset weight of the YS-th category, and YH YS represents the value of the YS-th category, that is, any one of the three values: sales growth rate, number of new product releases, and change in market share;

[0130] Obtain the macroeconomic indicators for the t period and the previous cv periods (such as one month or one quarter) from the National Bureau of Statistics, including GDP growth rate, inflation rate, unemployment rate, CCI index (Consumer Confidence Index, reflecting consumers' confidence level in the future economic environment), and PMI index (Purchasing Managers' Index, reflecting the economic activities of manufacturing and service industries), and perform standardization processing on all macroeconomic indicators;

[0131] According to the standardized macroeconomic indicators, construct an original data matrix with each macroeconomic indicator as a column and the corresponding values at each time point as rows;

[0132] According to the original data matrix, perform analysis using the principal component analysis method, and obtain XY structural quantities according to the Kaiser criterion method (XY is the number of structures to be extracted, that is, the number of structures to be extracted selected by the Kaiser criterion method, and the corresponding number of latent structures are extracted according to the number of structures), and extract the latent structures (the extracted latent structures can explain most of the changes in the original macroeconomic indicators). Take the latent structures as rows, the macroeconomic indicators as columns, and the relationship strength of the macroeconomic indicators on the corresponding latent structures as the values at each position to construct an unrotated load matrix;

[0133] Among them, the latent structure represents the common trend or pattern hidden behind multiple macroeconomic indicators (these patterns are not directly visible, but are latent factors revealed by analyzing the data. The structure classifies multiple related data points into several main trends to better understand and interpret the data; for example, assume that two latent structures are found through macroeconomic indicators, an economic growth structure mainly contributed by GDP growth rate and PMI, and an economic stability structure mainly contributed by inflation rate and unemployment rate);

[0134] According to the unrotated load matrix, use the orthogonal rotation method to rotate the unrotated load matrix to obtain a load matrix (rotation does not change the total variance explained by the latent structure, but redistributes this variance so that each latent structure is more concentrated on certain specific variables. The rotated load matrix is easier to interpret because each latent structure now more clearly reflects certain specific economic phenomena);

[0135] Multiply the load matrix by the original data matrix to obtain a structural score matrix. Each row in the structural score matrix represents the corresponding time point in the original data matrix, and each position represents the score of the corresponding macroeconomic indicator on the corresponding latent structure;

[0136] According to the structural score matrix, use the weighted average method to calculate all the latent structure scores at each time point, obtain the economic health score at each time point, and take the economic health score of the last time point in the t period as the economic health score of the t period;

[0137] The way of using the weighted average method to calculate all the latent structure scores at each time point includes:

[0138] Among them, QZDF represents the economic health score at each time point, XY represents the number of structures, that is, the number of latent structure scores is equal to the number of structures, SYZ represents the index, and SQZ SYZ represents the preset weight of the SYZ-th latent structure score at each time point, and YQZ SYZ represents the SYZ-th latent structure score at each time point;

[0139] Standardize the industry trend index and economic health score of the t period and the previous cv periods, and calculate the weights of the standardized industry trend index and economic health score respectively;

[0140]

[0141] Among them, QZ JJ represents the weight of the economic health score, QZ HY represents the weight of the industry trend index, JJ t represents the standardized economic health score of the t period, JJ 1 represents the average value of the economic health scores of the t period and the previous cv periods after standardization, HY t represents the standardized industry trend index of the t period, HY 1 represents the average value of the industry trend indexes of the t period and the previous cv periods;

[0142] According to the industry trend index, economic health score, and the corresponding weights, use the weighted summation formula to calculate the market sensitivity index;

[0143] The way of generating the customer satisfaction index based on the customer satisfaction score and order fulfillment rate includes:

[0144] Obtain the customer scores from the CRM system (Customer Relationship Management, which is software specifically designed to manage the interactions and relationships between enterprises and customers), and take the average of all customer scores within the t period as the customer satisfaction score for the t period;

[0145] Obtain the number of orders completed on time and the total number of orders from the ERP system (Enterprise Resource Planning, which is a set of integrated software applications used to manage and optimize various business processes within an enterprise), and take the ratio of the number of orders completed on time to the total number of orders within the t period as the order fulfillment rate for the t period;

[0146] Normalize the customer satisfaction scores and order fulfillment rates for the t period and the previous cv periods to the range [0, 1], and define the maximum value 1 within the range [0, 1] as the theoretical maximum value of the normalized customer satisfaction scores and order fulfillment rates;

[0147] For the normalized customer satisfaction scores and order fulfillment rates, calculate the correlation coefficients with the reference sequence respectively

[0148] Among them, hg represents the correlation coefficient of the customer satisfaction score or order fulfillment rate, and bl cv+t,pk represents the pk-th value among the normalized values of the customer satisfaction score or order fulfillment rate within the t period and the previous cv periods, bl represents the normalized value of the customer satisfaction score or order fulfillment rate at the t period, and min cv+t |bl cv+t -1| represents the minimum absolute difference between the customer satisfaction score or order fulfillment rate within the t period and the previous cv periods and the theoretical maximum value, and max cv+t |bl cv+t,pk -1| represents the maximum absolute difference between the customer satisfaction score or order fulfillment rate within the t period and the previous cv periods and the theoretical maximum value;

[0149] Multiply the normalized customer satisfaction score at the t period by the correlation coefficient of the customer satisfaction score, and multiply the order fulfillment rate by the correlation coefficient of the order fulfillment rate, and then sum them up to obtain the customer satisfaction index;

[0150] It should be noted that the customer satisfaction index may be affected by various factors, and these effects may be non-linear. The customer satisfaction index calculated by this method can effectively handle this complex relationship, provide more accurate evaluation results, and help quickly identify the key factors affecting customer satisfaction;

[0151] The method for generating the operation efficiency index based on the inventory turnover rate, average transportation time, and production efficiency score includes:

[0152] Extract the cost of goods sold and inventory value data from the ERP system, and take the ratio of the cost of goods sold to the average inventory value within the t period as the inventory turnover rate; take the average value of the inventory value at the first time point and the inventory value at the last time point within the t period as the average inventory value;

[0153] Obtain the transportation time data from the logistics system, and take the average value of all transportation times within the t period as the average transportation time within the t period;

[0154] Obtain the output speed and equipment operation status of the production line from IoT devices (Internet of Things, IoT refers to connecting various physical devices (such as sensors, controllers, smart home appliances, etc.) through the Internet, enabling these devices to communicate with each other and exchange data. These devices are usually embedded with sensors and network interfaces, and can automatically collect environmental data and send it to the cloud for processing and analysis), calculate the ratio of the actual output within the t period to the theoretical maximum output of the equipment, and multiply it by the equipment availability rate to obtain the production efficiency score within the t period; take the ratio of the actual operation time of the equipment to the planned operation time within the t period as the equipment availability rate;

[0155] Standardize the inventory turnover rate, average transportation time, and production efficiency score. Define the ideal solution of the standardized inventory turnover rate and production efficiency score as 1, the negative ideal solution as 0, the ideal solution of the average transportation time as 0, and the negative ideal solution as 1;

[0156] Calculate the distances between the standardized inventory turnover rate, average transportation time, and production efficiency score and the ideal solution and negative ideal solution;

[0157]

[0158] Among them, d + represents the distance from the inventory turnover rate kcz, average transportation time scx, and production efficiency score yss to the ideal solution, and d - represents the distance from kcz, scx, and yss to the negative ideal solution;

[0159] Sum d - and d + to obtain the total distance, and take the ratio of d - to the total distance as the operation efficiency index;

[0160] The method for constructing the environmental feature vector includes:

[0161] Using GIS (Geographic Information System) technology, map each node and transportation route in the supply chain onto a map, where the nodes are key locations or facilities in the complete supply chain from the supplier to the end consumer (for example, assuming a complete supply chain from the supplier to the end consumer includes a supplier, a manufacturing plant, a warehouse, a distribution center, a retail store, a cross-dock, a rail freight station, and a port, then these key locations or building facilities are respectively analogized into corresponding nodes);

[0162] Obtain real-time weather forecast data and traffic condition data of the regions where all nodes and transportation routes are located from the China Meteorological Administration and third-party navigation platforms, and convert them into a structured form through structured processing, where the structured processing includes field extraction, standardization, and timestamp alignment;

[0163] Field extraction is to identify and extract key information, including temperature, precipitation, wind speed, traffic congestion index, and environmental warning information; environmental warning information includes meteorological risk warnings (including heavy rain, flood, strong wind, ice and snow weather, extreme temperature), geological risk warnings (including earthquake, landslide, and debris flow), ecological risk warnings (including air pollution, water resource shortage (especially serious for industries relying on water resources such as food processing plants and some chemical plants)), biological risk warnings (including disease outbreaks (such as avian influenza, African swine fever, seriously affecting agricultural production and food supply chain safety), pest invasions (affecting agriculture, aquaculture, and forest resources)), and road closure warnings (including road closures, traffic congestion);

[0164] Standardization is to unify the units and formats of data from different sources; timestamp alignment is to make the timestamps of all data consistent and use interpolation method to fill in the missing time points;

[0165] Take the temperature, precipitation, and wind speed obtained after structured processing as environmental factors, and integrate them with environmental warning information and traffic congestion index into an environmental dataset;

[0166] According to the environmental dataset, for each node, statistically obtain the frequency of environmental warning information occurring at the location of each node within cv time periods before the t time period

[0167] Among them, II(RV(v)) represents an indicator function, which takes 1 when there is any type of environmental warning in the v time period, and takes 0 if there is no environmental warning of any type in the v time period, and YJP represents the frequency of environmental warning information at the location of the node within cv time periods before the t time period;

[0168] According to the environmental factors in the environmental dataset, use the weighted summation method to calculate the severe weather index at the location of each node in the t time period;

[0169] Combined with the environmental warning information frequency and the severe weather index, the weighted summation method is used to calculate the meteorological exposure degree value of each node at time period t;

[0170] For each node, the meteorological exposure degree value within time period t is concatenated with the environmental factors, traffic congestion index, and all environmental warning information levels in the environmental dataset to obtain the environmental feature vector of each node at time period t;

[0171] Among them, in the multi-dimensional feature vector, if there is any kind of environmental warning information for a node at time period t, the corresponding environmental warning information is taken as 1, and for the non-existent environmental warning information, it is taken as 0;

[0172] The training method of the risk prediction model includes:

[0173] A1: Horizontally concatenate the real-time evaluation vector and the environmental feature vector to form a comprehensive feature vector;

[0174] A2: Use the comprehensive feature vectors of all nodes as inputs and the risk probabilities of each node within the next time period as outputs to construct a risk prediction model, and the risk prediction model is a CatBoost model;

[0175] A3: Collect a sample set, including pp groups of samples. Each group of samples includes the comprehensive feature vectors of all nodes in a time period and the label corresponding to each node;

[0176] Among them, the label is whether a risk event has occurred at the corresponding node in the corresponding time period. If a risk event has occurred, the sample belongs to the positive class and the label is taken as 1. If no risk event has occurred, the sample belongs to the negative class and the label is taken as 0. Risk events include transportation delays (the reasons for occurrence such as traffic congestion, severe weather, and logistics facility failures, and the impacts include the inability of goods to arrive on time, resulting in inventory shortages or surpluses, affecting production plans), supply chain node interruptions (the reasons for occurrence such as natural disasters (such as earthquakes, floods), equipment failures, power outages, and the impacts such as key nodes (such as warehouses, manufacturing plants) being unable to operate normally, affecting production and distribution), and market demand fluctuations (the reasons for occurrence such as market trend changes, emergencies (sudden changes in customer evaluations of products, mutations in operations), and the impacts include sudden increases or decreases in demand, requiring rapid adjustment of inventory and distribution strategies);

[0177] A4: Divide the sample set into a training set and a validation set according to a certain proportion;

[0178] A5: Initialize the hyperparameters of the CatBoost model, use Bayesian optimization for hyperparameter tuning, use k-fold cross-validation to evaluate the cross-validation scores of the model under different hyperparameter combinations, and select the optimal parameter combination;

[0179] A6: Use the optimal parameter combination as the initial parameters of the model, and define the logarithmic loss function as the loss function for evaluating the prediction accuracy of the model. Where NZ represents the total number of samples, bq uv represents the label of the uv-th sample, pg uv represents the predicted value of the risk probability that the uv-th sample belongs to the positive class, pg uv = [0, 1];

[0180] A7: Use the training set and the loss function to adjust the model parameters according to the gradient in each iteration of the model.

[0181] For each iteration, update the model parameters using the training set, use AUC as the evaluation metric, and calculate the AUC value on the validation set.

[0182] According to the AUC value on the validation set, calculate the difference between the AUC value after the current iteration and the AUC value of the previous iteration, denoted as the iteration difference.

[0183] A8: Set the iteration difference threshold. If the iteration difference is greater than the iteration difference threshold, it is determined that the performance of the model has improved.

[0184] If the iteration difference is less than or equal to the iteration difference threshold, it is determined that the performance of the model has not improved.

[0185] If the performance of the model does not improve on the validation set in consecutive DD iterations, stop training to obtain the trained risk prediction model.

[0186] The method of jointly optimizing the inventory configuration and distribution path according to the risk assessment vector, combined with the current inventory level and the existing transportation path, to generate an optimization strategy includes:

[0187] Obtain the inventory quantity, inventory holding cost, and product demand at each node, as well as the existing transportation path arrangement at the current time period through the enterprise's financial system and ERP system. The existing transportation path arrangement includes the existing transportation path, vehicle scheduling for each transportation path, total amount of goods transported, transportation cost, and transportation time.

[0188] Define the objective function as maximizing the overall efficiency of the supply chain while minimizing the inventory cost and transportation cost: min∑ i cb i kc i +∑ j ysc j ysl j ;

[0189] Where cb i represents the inventory holding cost of node i, kc iRepresents the inventory level of node i, ysc j Represents the transportation cost of transportation route j, ysl j Represents the total quantity of goods transported through transportation route j;

[0190] Introduce a constraint to limit the inventory level of each node to be greater than or equal to the minimum safety inventory level and less than the inventory warehouse capacity;

[0191] According to the objective function and constraints, use an integer programming solver (such as Gurobi, CPLEX) to obtain the optimal inventory allocation;

[0192] According to the risk assessment vector, use a reinforcement learning algorithm (such as DQN or PPO) to dynamically optimize the distribution path, including:

[0193] Define the state space, where the state includes the inventory level of each node at the current time period, the existing transportation route arrangement, the risk assessment vector, and the product demand;

[0194] Define the action space, including selecting the optimal path from the existing transportation routes, changing the number of transportation vehicles, and adjusting the priorities of different transportation tasks;

[0195] Define the reward function to encourage fast and efficient distribution, while punishing delays and unnecessary transportation costs JLH = α × DDS - β × YSC - γ × YCD;

[0196] Where, JLH represents the reward, DDS represents the number of orders completed on time, YSC represents the total transportation cost on all transportation routes, YCD represents the number of orders not completed on time, the difference between the total number of orders and the number of orders completed on time is used as the number of orders not completed on time, α, β, and γ represent weight parameters, and use hyperparameter optimization techniques (such as Bayesian optimization, random search, or grid search) to find the best combination of α, β, and γ;

[0197] Collect the historical data set of the reinforcement learning model, including the inventory records, transportation route arrangements, risk probability values, and product demands of each node in the PC group;

[0198] Use the historical data set and the simulation environment to train the reinforcement learning model to obtain a reinforcement learning model that selects the optimal distribution route in different states;

[0199] Based on the trained reinforcement learning model, according to the risk assessment vector at the current time period, combined with the inventory level, inventory holding cost, and product demand of each node, and the existing transportation route arrangement, output the optimal distribution route for the current time period;

[0200] Combine the optimal inventory allocation and the optimal distribution route to form a comprehensive optimization strategy;

[0201] Update the inventory configuration and transportation route arrangement of each node in the supply chain according to the optimization strategy, and monitor the status of the updated supply chain in real time. When an anomaly is detected, generate an anomaly detection vector. The method of generating a cost prediction vector based on the anomaly detection vector includes:

[0202] Update the inventory configuration and transportation route arrangement of each node according to the optimization strategy;

[0203] Define a set of anomaly detection rules for the supply chain status, including inventory level anomalies, transportation delays, and environmental factor anomalies;

[0204] During the current period, when the inventory level is less than the minimum safety inventory capacity or greater than or equal to the maximum inventory capacity, it is determined as an inventory level anomaly;

[0205] When the transportation vehicle fails to arrive at or leave the designated node on time, it is determined as a transportation delay;

[0206] When the environment and traffic affect transportation, it is determined as an environmental factor anomaly;

[0207] Use a monitoring system to monitor the supply chain status in real time. When a situation within the anomaly detection rules is detected, generate an anomaly detection vector, including the anomaly type, affected node, and timestamp;

[0208] Define the categories of anomaly costs according to the anomaly detection rules, including additional inventory costs caused by inventory level anomalies; late delivery costs caused by transportation delays (fines or other losses that may occur due to late order delivery); increased transportation costs caused by environmental factor anomalies (additional costs brought about by adjusting the transportation route or increasing the transportation frequency due to the transportation route being blocked by environmental factors);

[0209] For each category of anomaly cost, estimate the anomaly cost of each category, including:

[0210] According to the inventory level anomaly, when the inventory level is greater than or equal to the maximum inventory capacity, multiply the difference between the inventory level and the maximum inventory capacity by the unit inventory holding cost to obtain the additional inventory cost;

[0211] According to the transportation delay, count all affected orders to obtain the total late delivery fine as the late delivery cost;

[0212] According to the environmental factors, calculate the difference between the unit transportation cost of the new transportation route used for replacement and the unit transportation cost of the original blocked transportation route, and multiply it by the transportation volume on the original transportation route to obtain the transportation cost of one replacement route. Count the transportation costs of all replacement routes to obtain the total increase in transportation costs;

[0213] Summarize all the estimated abnormal cost values to form a cost prediction vector;

[0214] The ways to take corrective measures on the supply chain according to the cost prediction vector include:

[0215] Set a cost acceptable threshold (the acceptable cost loss amount set by professionals in the industry after comprehensive consideration based on the market environment and enterprise internal information). If the total abnormal cost in the cost prediction vector is less than the cost acceptable threshold, it is considered that the lost cost is within the acceptable range, and it is determined that no corrective measures need to be taken;

[0216] If the total abnormal cost in the cost prediction vector is greater than or equal to the cost acceptable threshold, it is considered that the lost cost is unacceptable, and it is determined that corrective measures need to be taken;

[0217] For the situation where corrective measures need to be taken, the corrective measures include:

[0218] Adjust the inventory level: If the inventory level of a node is less than the minimum safety inventory capacity, transfer inventory from other nodes with sufficient inventory to this node. If the inventory level of a node is greater than or equal to the maximum safety inventory capacity, transfer the inventory of the node to other nodes with low inventory levels;

[0219] Re-plan the transportation route: Use real-time traffic data and navigation systems to dynamically adjust the transportation route to avoid congested sections. If there are multiple routes from the current node to the destination, use the shortest path algorithm to select the optimal path and use the optimal path to reduce transportation time and costs;

[0220] Temporarily increase the transportation capacity: Deploy additional vehicles into the transportation route and increase the transportation frequency to ensure the timely delivery of goods;

[0221] According to the corrective measures taken, re-evaluate the risk assessment vector. According to the new risk assessment vector, adjust the inventory levels of each node, re-plan the transportation route, and apply the new inventory levels and transportation routes.

[0222] In this embodiment, by integrating market changes, customer needs, and internal operations, a real-time evaluation vector is formed to comprehensively reflect the actual status of each link in the supply chain, eliminating the problem of data islands. Secondly, by combining external factors such as weather forecasts and traffic conditions, an environmental feature vector is constructed and combined with the real-time evaluation vector to train a risk prediction model, generating a risk probability vector for each node within a future time period, improving the accuracy and timeliness of risk prediction, helping enterprises identify potential risk points in advance, and enhancing the risk resistance ability of the supply chain. Then, based on the risk assessment vector, the inventory configuration and transportation route are jointly optimized to generate a global optimal solution, achieving the global optimization and flexible adjustment of the supply chain, reducing unnecessary costs, and improving the overall operation efficiency and economic benefits. Next, continuously monitor the supply chain status, detect anomalies in real-time and generate a cost prediction vector, establishing a closed-loop management mechanism from monitoring to response to ensure that all anomalies can be processed in a timely and effective manner, preventing problems from expanding, and reducing the impact of anomalies on the supply chain. Finally, according to the cost prediction vector, precise corrective measures are taken to ensure the stable operation of the supply chain, maximize the resource utilization efficiency, reduce the operation cost, and be able to proactively prevent the occurrence of potential problems, enhancing the stability and reliability of the supply chain, and ensuring instant response and high-precision anomaly detection.

[0223] Embodiment 2

[0224] Please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. A comprehensive optimization method for a product supply chain based on big data is provided, including:

[0225] S1. Obtain market changes, customer demand fluctuations, and internal operation data, conduct intelligent evaluation, and form a real-time evaluation vector;

[0226] S2. Integrate weather forecasts and traffic conditions, analyze their impacts on the supply chain, and construct an environmental feature vector; combine the real-time evaluation vector and the environmental feature vector, train to obtain a risk prediction model, predict the risk probability of each node within a future time period based on the risk prediction module, and splice the risk probability levels of all nodes into a risk assessment vector;

[0227] S3. Based on the risk assessment vector, combine the current inventory level and the existing transportation route, jointly optimize the inventory configuration and distribution route, and generate an optimization strategy;

[0228] S4. Update the inventory configuration and transportation route arrangements of each node in the supply chain according to the optimization strategy, continuously monitor the status of the updated supply chain, when an anomaly is detected, generate an anomaly detection vector, and generate a cost prediction vector according to the anomaly detection vector;

[0229] S5: Based on the cost prediction vector, take corrective measures on the supply chain to ensure the stable operation of the supply chain.

[0230] Embodiment III

[0231] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-provided comprehensive optimization method for the product supply chain based on big data.

[0232] Since the electronic device introduced in this embodiment is the electronic device adopted for implementing the comprehensive optimization method for the product supply chain based on big data in the embodiments of the present application, based on the comprehensive optimization method for the product supply chain based on big data introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variation forms of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device adopted for the comprehensive optimization method for the product supply chain based on big data in the embodiments of the present application, it falls within the scope of protection of the present application.

[0233] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters and threshold selections in the formulas are set by those skilled in the art according to the actual situation.

[0234] The above is only the preferred embodiment of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A comprehensive optimization system for product supply chain based on big data, characterized by: include: Intelligent evaluation module: used to obtain market changes, customer demand fluctuations and internal operation data, conduct intelligent evaluation and form real-time evaluation vectors; Dynamic risk prediction module: used to integrate weather forecasts and traffic conditions, analyze their impact on the supply chain, and construct an environmental feature vector; combine the real-time assessment vector and the environmental feature vector to train and obtain a risk prediction model, and predict the risk probability of each node in a future period based on the risk prediction module, and splice the risk probability levels of all nodes into a risk assessment vector; Strategy Optimization Module: Based on the risk assessment vector, combined with the current inventory level and existing transportation routes, the inventory configuration and distribution route are jointly optimized to generate an optimization strategy; Real-time monitoring and abnormal response module: used to update the inventory configuration and transportation route arrangement of each node in the supply chain according to the optimization strategy, monitor the updated supply chain status in real time, generate an abnormality detection vector when an abnormality is detected, and generate a cost prediction vector based on the abnormality detection vector; Correction module: used to take corrective measures on the supply chain based on the cost forecast vector to ensure the stable operation of the supply chain.

2. The product supply chain comprehensive optimization system based on big data according to claim 1 is characterized in that: The method of forming the real-time evaluation vector includes: Market changes include industry trend index and economic health score; customer demand fluctuations include customer satisfaction score and order fulfillment rate; internal operational data include inventory turnover rate, average shipping time and production efficiency score; Generate market sensitivity indicators based on industry trend index and economic health score; generate customer satisfaction indicators based on customer satisfaction score and order fulfillment rate; generate operational efficiency indicators based on inventory turnover rate, average shipping time and production efficiency score; Market sensitivity indicators, customer satisfaction indicators and operational efficiency indicators are horizontally spliced ​​to form a real-time evaluation vector.

3. The product supply chain comprehensive optimization system based on big data according to claim 2 is characterized in that: The method of generating the market sensitivity index based on the industry trend index and the economic health score includes: Obtain the sales growth rate, number of new product launches and market share changes in period t from industry analysis institutions, discretize period t into ct time points, and use the weighted average method to calculate the sales growth rate, number of new product launches and market share changes at the last time point in period t as the industry trend index for period t; Obtain macroeconomic indicators for period t and the previous cv periods from the National Bureau of Statistics, including GDP growth rate, inflation rate, unemployment rate, CCI index and PMI index, and standardize all macroeconomic indicators; According to the standardized macroeconomic indicators, the original data matrix is ​​constructed with each macroeconomic indicator as a column and the corresponding value at each time point as a row; According to the original data matrix, the principal component analysis method is used for analysis. According to the Kaiser criterion method, the number of XY structures and the latent structure corresponding to the number of structures are obtained. The unrotated loading matrix is ​​constructed with the latent structure as the row and the macroeconomic indicator as the column. The relationship strength of the macroeconomic indicator on the corresponding latent structure is the value of each position. According to the unrotated load matrix, the unrotated load matrix is ​​rotated using the orthogonal rotation method to obtain the load matrix; The product of the loading matrix and the original data matrix is ​​taken to obtain the structural score matrix, where each row in the obtained structural score matrix represents the corresponding time point in the original data matrix, and each position represents the score of the corresponding macroeconomic indicator on the corresponding latent structure; According to the structure score matrix, all potential structure scores at each time point are calculated using the weighted average method to obtain the economic health score at each time point, and the economic health score at the last time point of period t is taken as the economic health score of period t; Standardize the industry trend index and economic health score of period t and the previous cv periods, and calculate the weights of the standardized industry trend index and economic health score respectively; Among them, QZ JJ represents the weight of the economic health score, QZ HY Indicates the weight of the industry trend index, JJ t represents the standardized economic health score for period t, JJ 1 It represents the average value of the economic health score of the standardized period t and the previous cv periods, HY t represents the standardized industry trend index for the t period, HY 1 It represents the average value of the industry trend index in the t period and the previous cv periods; Based on the industry trend index and economic health score, as well as the corresponding weights, the market sensitivity index is calculated using the weighted summation formula.

4. The product supply chain comprehensive optimization system based on big data according to claim 3 is characterized in that: The method of generating the customer satisfaction index based on the customer satisfaction score and the order fulfillment rate includes: Obtain customer ratings from the CRM system and take the mean of all customer ratings in period t as the customer satisfaction score for period t; Obtain the number of orders completed on time and the total number of orders from the ERP system, and take the ratio of the number of orders completed on time to the total number of orders in period t as the order fulfillment rate in period t; Normalize the customer satisfaction scores and order fulfillment rates in period t and the previous cv periods to the range of [0,1]; For the standardized customer satisfaction score and order fulfillment rate, calculate the correlation coefficient with the reference series respectively Where hg represents the correlation coefficient of customer satisfaction score or order fulfillment rate, bl cv+t,pk represents the pkth value in the standardized value of the customer satisfaction score or order fulfillment rate in period t and the previous cv periods, bl represents the standardized value of the customer satisfaction score or order fulfillment rate in period t, min cv+t |bl cv+t -1| represents the minimum absolute difference between the customer satisfaction score or order fulfillment rate in period t and the previous cv periods and the theoretical maximum value, max cv+t |bl cv+t,pk -1| represents the maximum absolute difference between the customer satisfaction score or order fulfillment rate in period t and the previous cv periods and the theoretical maximum value; Take the product of the standardized customer satisfaction score and the correlation coefficient of the customer satisfaction score in period t, and the product of the order fulfillment rate and the correlation coefficient of the order fulfillment rate, and sum them to obtain the customer satisfaction index; The method of generating the operational efficiency index based on the inventory turnover rate, average transportation time and production efficiency score includes: Extract sales cost and inventory value data from the ERP system, take the ratio of sales cost to average inventory value in period t as inventory turnover rate; take the average of the inventory value at the first time point and the inventory value at the last time point in period t as average inventory value; Obtain transportation time data from the logistics system and take the mean of all transportation times in period t as the average transportation time in period t; Obtain the production line output speed and equipment operating status from the IoT equipment, calculate the ratio of the actual output in period t to the theoretical maximum output of the equipment, and multiply it by the equipment availability to obtain the production efficiency score for period t; take the ratio of the actual operating time of the equipment in period t to the planned operating time as the equipment availability; The inventory turnover rate, average transportation time and production efficiency score are standardized, and the ideal solution of the standardized inventory turnover rate and production efficiency score is defined as 1, and the negative ideal solution is defined as 0, and the ideal solution of the average transportation time is defined as 0, and the negative ideal solution is defined as 1; Calculate the distance between the normalized inventory turnover rate, average transportation time, and production efficiency score and the ideal solution and the negative ideal solution; Among them, d + represents the distance from the inventory turnover rate kcz, average transportation time scx and production efficiency score yss to the ideal solution, d - represents the distances of kcz, scx and yss to the negative ideal solution; Right - and d + Sum the distance and take d - The ratio to the total distance is used as an indicator of operational efficiency.

5. The product supply chain comprehensive optimization system based on big data according to claim 4 is characterized in that: The method of constructing and obtaining the environmental feature vector includes: Use GIS technology to map the nodes and transportation routes in the supply chain, where the nodes are key locations or facilities in the complete supply chain from suppliers to end consumers; Obtain weather forecast data and traffic condition data for all nodes and transportation routes in real time from the China Meteorological Administration and third-party navigation platforms, and convert them into a structured form through structured processing, including field extraction, standardization, and timestamp alignment; Field extraction is to identify and extract key information, including temperature, precipitation, wind speed, traffic congestion index and environmental warning information; environmental warning information includes meteorological risk warning, geological risk warning, biological risk warning, harmful biological invasion and road closure warning; Standardization is to unify the units and formats of data from different sources; timestamp alignment is to make the timestamps of all data consistent and use interpolation to fill in missing time points; The temperature, precipitation and wind speed obtained after structured processing are used as environmental factors and integrated with environmental warning information and traffic congestion index into an environmental data set; According to the environmental data set, for each node, the frequency of environmental warning information appearing at each node in the cv time periods before the t period is statistically obtained. Where, II(RV(v)) represents the indicator function, which takes the value of 1 when there is any type of environmental warning in period v, and takes the value of 0 if there is no environmental warning in period v. YJP represents the frequency of environmental warning information at the location of the node within cv periods before period t. According to the environmental factors in the environmental data set, the weighted summation method is used to calculate the severe weather index of each node in the t period; Combining the frequency of environmental warning information and the severe weather index, the weighted summation method is used to calculate the meteorological exposure value of each node in period t; For each node, the meteorological exposure value in period t is spliced ​​with the environmental factors, traffic congestion index and all environmental warning information levels in the environmental data set to obtain the environmental feature vector of each node in period t; Among them, in the multi-dimensional feature vector, if the node has any kind of environmental warning information in time period t, the corresponding environmental warning information will be taken as 1, and the non-existent environmental warning information will be taken as 0.

6. The product supply chain comprehensive optimization system based on big data according to claim 5 is characterized in that: The training method of the risk prediction model includes: A1: horizontally concatenate the real-time evaluation vector and the environmental feature vector to form a comprehensive feature vector; A2: Taking the comprehensive feature vectors of all nodes as input and the risk probability of each node in a future period as output, a risk prediction model is constructed, and the risk prediction model is a CatBoost model; A3: Collect sample sets, including pp groups of samples. Each group of samples includes the comprehensive feature vectors of all nodes in a period and the labels corresponding to each node. The label is whether a risk event has occurred in the node during the corresponding period. If a risk event occurs, the sample belongs to the positive class and the label is 1. If no risk event occurs, the sample belongs to the negative class and the label is 0. Risk events include transportation delays, supply chain node interruptions, and market demand fluctuations. A4: Divide the sample set into training set and validation set in proportion; A5: Initialize the hyperparameters of the CatBoost model, use Bayesian optimization to tune the hyperparameters, use k-fold cross validation to evaluate the cross-validation scores of the model under different hyperparameter combinations, and select the optimal parameter combination; A6: Use the optimal parameter combination as the initial parameters of the model and define the logarithmic loss function as the loss function to evaluate the prediction accuracy of the model Among them, NZ represents the total number of samples, bq uv Indicates the label of the uvth sample, pg uv Represents the predicted value of the risk probability that the uvth sample belongs to the positive class; A7: Use the training set and loss function to adjust the model parameters according to the gradient in each iteration of the model; For each iteration, the model parameters are updated using the training set, and the AUC value is calculated on the validation set using the AUC as the evaluation metric; According to the AUC value on the validation set, calculate the difference between the AUC value after the current iteration and the AUC value of the previous iteration, which is recorded as the iteration difference; A8: Set the iteration difference threshold. If the iteration difference is greater than the iteration difference threshold, the performance of the model is judged to be improved. If the iteration difference is less than or equal to the iteration difference threshold, it is determined that the performance of the model has not been improved; If the performance of the model on the validation set does not improve in consecutive DD iterations, the training is stopped to obtain the trained risk prediction model.

7. The product supply chain comprehensive optimization system based on big data according to claim 6 is characterized in that: The method of jointly optimizing inventory configuration and distribution path based on the risk assessment vector, combined with the current inventory level and the existing transportation path, and generating the optimization strategy includes: Obtain the inventory volume, inventory holding cost and product demand of each node in the current period through the company's financial system and ERP system, as well as the existing transportation route arrangement. The existing transportation route arrangement includes the existing transportation routes, vehicle scheduling for each transportation route, the total amount of goods transported, transportation cost and transportation time; Define the objective function as maximizing the overall efficiency of the supply chain while minimizing inventory costs and transportation costs: minΣ i cb i kc i +∑ j ysc j ysl j ; Among them, cb i represents the inventory holding cost of node i, kc i represents the inventory of node i, ysc j represents the transportation cost of transportation path j, ysl j represents the total amount of goods transported through transportation route j; The introduced constraint condition is to limit the inventory quantity of each node to be greater than or equal to the minimum safety stock level and less than the inventory warehouse capacity; According to the objective function and constraints, the integer programming solver is used to obtain the optimal inventory configuration; Based on the risk assessment vector, the delivery path is dynamically optimized using reinforcement learning algorithms, including: Define the state space, where the state includes the inventory of each node in the current period, the existing transportation route arrangement, the risk assessment vector, and the product demand; Define the action space, including selecting the best route from existing transport routes, changing the number of transport vehicles, and adjusting the priority of different transport tasks; The reward function is defined to encourage fast and efficient delivery while penalizing delayed and unnecessary transportation costs JLH = α × DDS - β × YSC - γ × YCD; Wherein, JLH represents the reward, DDS represents the number of orders completed on time, YSC represents the total transportation cost on all transportation routes, YCD represents the number of orders that are not completed on time, and the difference between the total number of orders and the number of orders completed on time is taken as the number of orders that are not completed on time. α, β and γ represent weight parameters, and the hyperparameter optimization technology is used to find the best combination of α, β and γ. Collect historical data sets for the reinforcement learning model, including inventory records, transportation route arrangements, risk probability values, and product demand at each node of the PC group; Use historical data sets and simulation environments to train the reinforcement learning model, and obtain a reinforcement learning model that selects the optimal delivery route under different conditions; Based on the trained reinforcement learning model, according to the risk assessment vector of the current period, combined with the inventory of each node, inventory holding cost and product demand, as well as the existing transportation route arrangement, the optimal delivery route for the current period is output; Combine the optimal inventory configuration and the optimal delivery path to form a comprehensive optimization strategy.

8. The product supply chain comprehensive optimization system based on big data according to claim 7 is characterized in that: The method of generating an anomaly detection vector and generating a cost prediction vector according to the anomaly detection vector includes: Update the inventory configuration and transportation route arrangement of each node according to the optimization strategy; Define a set of anomaly detection rules for supply chain status, including abnormal inventory levels, transportation delays, and abnormal environmental factors; In the current period, when the inventory quantity is less than the minimum safety inventory capacity or greater than or equal to the maximum inventory capacity, it is determined to be an abnormal inventory level; When the transport vehicle fails to arrive at or leave the designated node on time, it is considered a transport delay; When the environment and traffic have an impact on transportation, the environmental factors are judged to be abnormal; Use the monitoring system to monitor the supply chain status in real time. When a situation that exists within the anomaly detection rule is detected, an anomaly detection vector is generated, including the anomaly type, the affected node, and the timestamp. Define the categories of abnormal costs based on anomaly detection rules, including additional inventory costs caused by abnormal inventory levels; delayed delivery costs caused by transportation delays; and increased transportation costs caused by abnormal environmental factors; For each category of abnormal cost, estimate the abnormal cost for each category, including: According to the abnormal inventory level, when the inventory level is greater than or equal to the maximum inventory capacity, the difference between the inventory level and the maximum inventory capacity is multiplied by the unit inventory holding cost to obtain the additional inventory cost; According to the transportation delay, all affected orders are counted to obtain the total late delivery penalty as the late delivery cost; According to environmental factors, the difference between the unit transportation cost of the new transportation route used for substitution and the unit transportation cost of the blocked original transportation route is calculated, and multiplied by the transportation volume on the original transportation route to obtain the transportation cost of an alternative route. The transportation costs of all alternative routes are counted to obtain the total transportation cost increase; All estimated abnormal cost values ​​are aggregated to form a cost prediction vector.

9. The product supply chain comprehensive optimization system based on big data according to claim 8 is characterized in that: The methods of taking corrective measures on the supply chain based on the cost forecast vector include: Set a cost acceptance threshold. If the sum of abnormal costs in the cost prediction vector is less than the cost acceptance threshold, the loss cost is considered to be within an acceptable range and no corrective measures are required. If the sum of abnormal costs in the cost prediction vector is greater than or equal to the cost acceptance threshold, the loss cost is considered unacceptable and corrective measures need to be taken; For situations where corrective actions are required, the corrective actions taken include: Adjust inventory levels: If the inventory level of a node is less than the minimum safety stock capacity, the inventory will be transferred from other nodes with sufficient inventory to this node. If the inventory level of a node is greater than or equal to the maximum safety stock capacity, the inventory of the node will be transferred to other nodes with low inventory levels. Replanning transportation routes: Using real-time traffic data and navigation systems, dynamically adjust transportation routes to avoid congested roads. If there are multiple paths from the current node to the destination, use the shortest path algorithm to select the optimal path, and use the optimal path to reduce transportation time and cost; Temporary increase of transport capacity: deploy additional vehicles to the transport routes and increase the transport frequency to ensure timely delivery of goods; Based on the corrective measures taken, the risk assessment vector is re-evaluated, and based on the new risk assessment vector, the inventory level of each node is adjusted, the transportation route is re-planned, and the new inventory level and transportation route are applied.

10. A method for comprehensive optimization of product supply chain based on big data, applied to the implementation of the comprehensive optimization system for product supply chain based on big data according to any one of claims 1 to 9, characterized in that: include: Obtain market changes, customer demand fluctuations, and internal operation data, conduct intelligent evaluation, and form a real-time evaluation vector; Integrate weather forecasts and traffic conditions, analyze their impact on the supply chain, and construct an environmental feature vector; combine the real-time assessment vector and the environmental feature vector to train and obtain a risk prediction model, and predict the risk probability of each node in the future period based on the risk prediction module, and splice the risk probability levels of all nodes into a risk assessment vector; Based on the risk assessment vector, combined with the current inventory level and existing transportation routes, the inventory configuration and distribution route are jointly optimized to generate an optimization strategy; Update the inventory configuration and transportation route arrangement of each node in the supply chain according to the optimization strategy, monitor the updated supply chain status in real time, generate anomaly detection vectors when anomalies are detected, and generate cost prediction vectors based on the anomaly detection vectors; Based on the cost forecast vector, corrective measures are taken on the supply chain to ensure the stable operation of the supply chain.

Images