Supply chain toughness optimization method based on multi-objective optimization

Abstract

The invention discloses a supply chain toughness optimization method based on multi-objective optimization, and belongs to the technical field of artificial intelligence, and the method comprises the steps: building a supply chain risk database, building a simulation and optimization model, generating an optimization strategy, and generating an optimization strategy, and achieves the purpose of dynamically optimizing a supply chain recovery strategy through cooperation of multi-source data driving and an intelligent algorithm. According to the method, multi-dimensional data of politics, public opinions, operation and accidents can be integrated, a composite stability index SSI can be constructed, risk prospective prediction can be realized, an emergency strategy is trained in a virtual environment based on a simulation-reinforcement learning hybrid framework, the actual trial and error cost is reduced, and the method is suitable for popularization and application. And generating an optimal recovery scheme through a cost-service-toughness comprehensive objective function and a dynamic programming algorithm.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to a supply chain resilience optimization method based on multi-objective optimization. Background Art

[0002] Currently, traditional supply chain risk management methods mainly rely on static risk assessment models (such as the Analytic Hierarchy Process, Risk Matrix) and single-objective optimization strategies (such as cost minimization), which have certain limitations.

[0003] Principle overview of the static risk assessment model:

[0004] Analytic Hierarchy Process (AHP): Quantifies the weights of risk factors through expert scoring, but relies on subjective experience and cannot dynamically capture sudden risks such as political unrest and public opinion outbreaks.

[0005] Risk Matrix: Static mapping of risk probability and impact level, lacking quantitative analysis of the relevance and cascading failures of supply chain nodes.

[0006] Historical data regression model: Predicts risks based on the enterprise operation history (such as delivery delay rate), but ignores real-time disturbances in the external environment (such as international sanctions and logistics network paralysis).

[0007] The deficiencies of the static risk assessment model are as follows:

[0008] Single data dimension: Only relies on enterprise internal operation data (such as inventory, orders), without integrating multi-source heterogeneous data such as political stability and public opinion events, resulting in one-sided risk assessment. For example, it cannot warn of the risk of supplier supply interruption due to geopolitical conflicts.

[0009] Rigid response strategy: Adopts predefined rules (such as "trigger replenishment when safety inventory ≥ 30%"), unable to dynamically adjust strategy parameters according to real-time risks, and is prone to strategy failure in the event of sudden grey failures (such as a 50% reduction in production capacity).

[0010] Principle overview of the single-objective optimization strategy:

[0011] Linear / Integer programming: Takes cost minimization or service maximization as a single objective, but severs the relevance of cost, timeliness, and resilience indicators, resulting in decision-making deviating from actual needs.

[0012] Monte Carlo simulation: Simulates supply chain interruption scenarios through random sampling, but lacks interaction with optimization algorithms, and the simulation results are only used for risk assessment rather than strategy generation.

[0013] Genetic Algorithm (GA): When used for multi-objective optimization, it needs to preset fixed weights and does not combine dynamic environment feedback, and the solution results are difficult to adapt to real-time changing risk states.

[0014] The deficiencies of the single-objective optimization strategy are as follows:

[0015] Multi-objective fragmentation: Traditional methods regard cost, service, and resilience as independent objectives and use simplified methods such as weighted summation to handle them, which cannot effectively characterize the non-linear trade-off relationships between objectives (e.g., "reducing costs by 10% may lead to a 30% decrease in resilience").

[0016] Simulation-decision disconnection: The simulation model and the optimization algorithm run separately. After the strategy is generated, its effectiveness cannot be dynamically verified through simulation, forming an "open-loop" decision-making process, which is difficult to cope with the complex dynamic characteristics of the supply chain network. Summary of the Invention

[0017] The objective of the present invention is to provide a supply chain resilience optimization method based on multi-objective optimization, which solves the technical problem of dynamically optimizing the supply chain recovery strategy through the collaboration of multi-source data-driven and intelligent algorithms and achieving the multi-objective balance of cost, service, and resilience.

[0018] To achieve the above objective, the present invention adopts the following technical means:

[0019] A supply chain resilience optimization method based on multi-objective optimization includes the following steps:

[0020] Step 1: Obtain data sources from the Internet at the data collection layer, including political stability data, international conflict and war public opinion data, public enterprise operation data, and accident case data;

[0021] Clean and preprocess the data sources at the data collection layer to obtain the standardized political stability risk score R politics , public opinion risk score R conflict , inventory risk score and accident risk score R accident ;

[0022] Calculate the comprehensive supplier stability index SSI, and the specific formula is as follows:

[0023]

[0024] where w1, w2, w3, and w4 are all weights;

[0025] Step 2: Establish a discrete event simulation model at the analysis and decision-making layer. Abstract the supply chain nodes as state variables, and input the comprehensive supplier stability index SSI and the predicted inventory risk as input data into the discrete event simulation model for simulation;

[0026] In the simulation, simulate events such as complete supply chain disruption by suppliers, partial supply disruption, and logistics delays, simulate the emergency strategy response process, optimize the supplier switching strategy through a reinforcement learning model, generate a simulation report, and generate a supplier switching strategy.

[0027] Step 3: Construct a multi-objective optimization model aiming to minimize recovery costs, maximize service levels, and enhance supply chain resilience in the analysis and decision-making layer, optimize the supplier switching strategy, and generate a list of recommended alternative suppliers.

[0028] Update the data source in real-time, update the comprehensive supplier stability index SSI, and feedback it to the discrete event simulation model and the multi-objective optimization model to generate an optimized strategy.

[0029] Step 4: Display the comprehensive supplier stability index SSI and the list of recommended alternative suppliers in the form of visual charts in the application and display layer.

[0030] Preferably, when executing Step 1, it specifically includes the following steps:

[0031] Step 1-1: Deploy a political data server in the data collection layer. The political data server downloads the.xlsx format index file of global governance indicators from the Internet, which includes 6 governance indicators, namely VA, PV, GE, RQ, RL, and CC for each country.

[0032] The political data server obtains the profile information of all suppliers input by the user through the client. The profile information includes the region and country where the supplier is located.

[0033] Extract the country where any one supplier is located, obtain the 6 governance indicators of this country in the index file, and calculate the standardized risk score for each indicator. The specific calculation formula is as follows:

[0034]

[0035] where, R i represents the standardized score of the i-th indicator, and i represents the number of any one of the 6 governance indicators; Estimate i is the Estimate value of the i-th indicator, and Comparator i is the Comparator reference value of the i-th indicator; σ i represents the historical standard deviation;

[0036] Generate a political scoring data table for all countries where suppliers are located based on the calculation results of the standardized risk scores.

[0037] Step 1-2: Deploy an opinion processing server at the data collection layer. The opinion processing server crawls news texts containing keywords through Google Alerts, NewsAPI or Twitter API. The keywords include sanction, tariff or export ban;

[0038] Set a preset time window, count the frequency f of the keyword appearance within the preset time window, and perform normalization calculation to obtain the public opinion risk score R conflict :

[0039]

[0040] where f is the number of times the keyword appears within the preset time period, f min and f max are the minimum and maximum frequencies respectively during the observation period;

[0041] Step 1-3: Deploy an enterprise data server at the data collection layer. The enterprise data server obtains the on-time delivery rate R on-time and order backlog rate R backlog of all suppliers from public industry reports, news reports and public delivery records on the Internet, and constructs an enterprise historical dataset;

[0042] Sort out the historical data in the enterprise historical dataset, calculate the time series average value and change rate of each index. Specifically, use the public on-time delivery rate and order backlog rate to construct a regression model to predict inventory risk:

[0043]

[0044] where, is the predicted inventory risk index, and α, β1 and β2 are regression coefficients;

[0045] Step 1-4: Deploy a case server at the data collection layer. The case server obtains the occurrence times, influence scope and severity of all suppliers' supply chain disconnection and logistics anomaly accidents in history from public accident case reports and industry research reports;

[0046] For each supplier, calculate the accident occurrence times and the influence coefficient of each accident, and construct an accident risk scoring model. The specific formula is as follows:

[0047]

[0048] where D i is the occurrence times of the i-th accident; S i is the influence coefficient of the i-th accident, and the value of S i is in the range of [1,5], and T is the statistical time window;

[0049] Steps 1 - 5: Align each data source according to the supplier, country identifier, and time dimension, and obtain a unified data set through data merging operations;

[0050] Steps 1 - 6: Calculate the comprehensive supplier stability index SSI, output the SSI score of each supplier, sort by the score, obtain a list of low - risk suppliers, and generate a list of alternative suppliers.

[0051] Preferably, when performing Step 2, a simulation server is established at the analysis and decision - making layer. The simulation server uses the discrete - event simulation DES method to simulate the operation status of the supply chain under partial interruption scenarios and evaluate the effectiveness of recovery strategies, specifically including the following steps:

[0052] Step 2 - 1: The simulation server conducts target and problem definition. By defining supply - chain nodes and state variables, a supply - chain simulation scenario affected by external risk disturbances is constructed. The supply - chain simulation scenario includes three types of interruption events: complete chain breakage, gray failure, and logistics delay;

[0053] Supply - chain nodes include suppliers, logistics links, and inventory nodes;

[0054] State variables include the supplier inventory level I, order demand D, on - time delivery rate R, R = R on-time and the comprehensive supplier stability index SSI;

[0055] Gray failure specifically refers to the decline in the supply capacity of suppliers;

[0056] Step 2 - 2: The simulation server conducts model design. Specifically, a dynamic model is established using the discrete - event simulation method, including an inventory dynamic model, an event - trigger model, and a key performance indicator model:

[0057] Inventory dynamic model:

[0058] I i (t + 1)=I i (t)+S i (t)-D i (t);

[0059] where, I i (t) is the inventory level of the i - th supplier at time t;

[0060] S i (t) is the supply volume of the i - th supplier at time t:

[0061] S i (t)=S i,0 ×f(SSI i ,E(t));

[0062] Among them, S i,0 is the normal supply volume, f(·) is a function affected by risk disturbances, and E(t) represents an event factor (such as a broken chain or a gray failure);

[0063] D i (t) is the order volume that is satisfied. If there is a logistics delay, the order fulfillment rate will decrease:

[0064] D i (t) = D0(t) × [1 - δ(t)];

[0065] Among them, D0(t) is the theoretical order volume, and δ(t) is the order loss ratio caused by logistics delay;

[0066] The event trigger model specifically uses the Poisson distribution to simulate event triggers, including broken chain events, gray failures, and logistics delay events:

[0067] Broken chain event: P(broken chain at t) = p 断链 ;

[0068] Gray failure: P(gray failure at t) = p 灰 ; If triggered, the supply capacity becomes λ × S i,0 , where λ is the failure attenuation coefficient;

[0069] Logistics delay event: P(delay at t) = p 延迟 ; If triggered, the delay time Δt is determined according to a preset distribution (such as a normal distribution);

[0070] Key performance indicator model:

[0071]

[0072] Among them, D 实际 (t) represents the actual number of orders satisfied at time t; D0(t) represents the total order demand at time t, and T represents the total statistical time period;

[0073] Record the time from the event trigger to when the inventory or order fulfillment rate recovers to the preset threshold to obtain the recovery time T 恢复 ;

[0074] Calculate the additional costs generated due to broken chains or logistics delays to obtain the recovery cost RecoveryCost, specifically by calculating the express transportation cost or the additional procurement cost of alternative suppliers;

[0075] Step 2-3: The simulation server uses the reinforcement learning model for optimization, specifically including:

[0076] Define the state s(t): s(t) = {I i(t), D0(t), SSI i , L(t), E(t)};

[0077] Define the action space a(t), which specifically includes switching to an alternative supplier, adjusting the inventory scheduling strategy, and expediting the transportation logistics;

[0078] Optimize through the reward function R(s, a):

[0079] R(s, a) = ServiceLevel - λ4 × RecoveryCost;

[0080] where λ4 is the cost penalty coefficient;

[0081] Step 2 - 4: The simulation server inputs the comprehensive supplier stability index SSI, on - time delivery rate R on-time , order backlog rate R backlog , predicted inventory risk indicators , the probabilities or statistical distribution parameters of chain breakage, grey failures, and logistics delays into the model in Step 2 - 2, runs the model in Step 2 - 2 for simulation, generates a simulation result report of the supply chain under various shock scenarios, and generates a supplier switching strategy based on the simulation result report, which specifically includes:

[0082] The recovery capabilities of each supplier and the overall service level of the supply chain;

[0083] Emergency strategy recommendations based on simulation and reinforcement learning, including alternative supplier switching and inventory scheduling strategies;

[0084] Performance indicators, namely order fulfillment rate, recovery time, and recovery cost.

[0085] Preferably, when executing Step 3, deploy a policy server at the analysis and decision - making layer. The policy server constructs a multi - objective function model, which specifically includes the following steps:

[0086] Step 3 - 1: Construct a recovery cost function, including logistics expediting cost, inventory shortage or surplus cost, and alternative supplier startup cost. The specific formula is as follows:

[0087]

[0088] where L(x, t) represents the logistics cost in the t - th period, S(x, t) represents the supplier switching cost, Cost(x) represents the recovery cost, c logistics , c inventory and c switch are all cost coefficients, x is the supplier switching strategy, T is the time window; I(t) is the current inventory, and I target (t) is the target inventory level;

[0089] Step 3-2: Construct the service level function ServiceLevel(x) to measure the order fulfillment rate and delivery timeliness. The specific formula is as follows:

[0090]

[0091] Among them, D fulfilled (x,t) represents the actual number of orders fulfilled in the t-th period under strategy x; D0(t) represents the theoretical order demand;

[0092] Step 3-3: Construct the supply chain resilience index model Resilience(x), which is specifically defined as the time required to recover to normal operation after a shock or the degree of improvement in the system state. The formula is as follows:

[0093]

[0094] Among them, T recovery (x) represents the time required to recover to the preset service level under measurement x. The shorter the recovery time, the higher the resilience index;

[0095] Step 3-4: Construct a multi-objective function model. The specific formula is as follows:

[0096] min f(x) = w7 × Cost(x) + w8 × ServiceLevel(x) + w9 × Resilience(x);

[0097] Among them, w7, w8, and w9 are all weights;

[0098] Step 3-5: Solve according to the multi-objective function model to obtain the optimal decision x, and generate a list of recommended alternative suppliers.

[0099] A supply chain resilience optimization method based on multi-objective optimization according to the present invention solves the technical problem of dynamically optimizing the supply chain recovery strategy through the collaboration of multi-source data-driven and intelligent algorithms to achieve the multi-objective balance of cost, service, and resilience. The present invention can integrate multi-dimensional data of politics, public opinion, operation, and accidents, construct a composite stability index SSI, realize forward-looking risk prediction, train emergency strategies in a virtual environment based on a simulation-reinforcement learning hybrid framework, reduce the actual trial-and-error cost, and generate an optimal recovery plan through a cost-service-resilience comprehensive objective function and a dynamic programming algorithm. Brief Description of the Drawings

[0100] Figure 1 is the system architecture diagram of the present invention;

[0101] Figure 2 is the main flow chart of the present invention;

[0102] Figure 3 It is the flowchart of Step 1 of the present invention;

[0103] Figure 4 It is the flowchart of Step 2 of the present invention;

[0104] Figure 5 It is the flowchart of Step 3 of the present invention. Detailed implementation manners

[0105] by Figures 1 - 5 A supply chain resilience optimization method based on multi-objective optimization shown as follows includes the following steps:

[0106] Step 1: Obtain data sources from the Internet at the data collection layer, including political stability data, international conflict and war public opinion data, public enterprise operation data, and accident case data;

[0107] Clean and preprocess the data sources at the data collection layer to obtain the standardized political stability risk score R politics , public opinion risk score R conflict , inventory risk score and accident risk score R accident ;

[0108] Calculate the comprehensive supplier stability index SSI, and the specific formula is as follows:

[0109]

[0110] wherein, w1, w2, w3, and w4 are all weights.

[0111] When implementing Step 1, it specifically includes the following steps:

[0112] Step 1-1: Deploy a political data server at the data collection layer. The political data server downloads the.xlsx format index file of global governance indicators from the Internet, which includes 6 governance indicators of VA (Voice and Accountability), PV (Political Stability and Absence of Violence), GE (Government Effectiveness), RQ (Regulatory Quality), RL (Rule of Law), and CC (Control of Corruption) for each country;

[0113] In this embodiment, the.xlsx format indicator file of the global governance indicators is obtained by downloading from the official website of the World Bank Group. The data is shown in Table 1. Table 1 shows the PV values of some countries, and each item includes the estimate value, the lower value, and the upper value. In this embodiment, only the estimate value is used to participate in the model calculation, and the COMPARATOR value is the reference value:

[0114]

[0115] Table 1

[0116] The political data server obtains the profile information of all suppliers input by the user through the client. The profile information includes the region and country where the supplier is located;

[0117] Extract the country where any one supplier is located, obtain the 6 governance indicators of this country in the indicator file, and calculate the standardized risk score for each indicator. The specific calculation formula is as follows:

[0118]

[0119] Among them, R i represents the standardized score of the i-th indicator, and i represents the number of any one of the 6 governance indicators; Estimate i is the Estimate estimated value of the i-th indicator, and Comparator i is the Comparator reference value of the i-th indicator; σ i represents the historical standard deviation;

[0120] In this embodiment, according to the suppliers shown in the supply chain, the country where they are located is selected, and the corresponding governance indicators are obtained in the.xlsx format indicator file of the global governance indicators.

[0121] According to the calculation results of the standardized risk scores, a political score data table for all countries where the suppliers are located is generated;

[0122] Step 1-2: Deploy an opinion handling server at the data collection layer. The opinion handling server grabs news texts containing keywords through GoogleAlerts, NewsAPI or Twitter API. The keywords include sanction, tariff or export ban;

[0123] In this embodiment, in the actual application process, real-time news is obtained by subscribing to emails of Google Alerts, NewsAPI, and TwitterAPI, and keywords are scraped from the news text. To count the number of occurrences of keywords using regular expressions or simple text matching, regardless of the specific vocabulary, as long as "sanction", "tariff", or "export ban" appears, it is considered as one occurrence. Then, the keywords of the country referred to in the news are screened to obtain the frequency of occurrence of the keywords of the country where the supplier is located. In this embodiment, a monthly time window is used to count the frequency of occurrence of keywords.

[0124] Set a preset time window, count the frequency f of keyword occurrences within the preset time window, and perform normalization calculation to obtain the public opinion risk score R conflict :

[0125]

[0126] where f is the number of keyword occurrences within the preset time period, f min and f max are the minimum and maximum frequencies within the observation period respectively;

[0127] Step 1-3: Deploy an enterprise data server at the data collection layer. The enterprise data server obtains the on-time delivery rate R on-time and the order backlog rate R backlog of all suppliers from public industry reports, news reports, and public delivery records on the Internet, and constructs an enterprise historical dataset;

[0128] Sort out the historical data in the enterprise historical dataset, and calculate the time series average and change rate of each indicator. Specifically, use the public on-time delivery rate and order backlog rate to construct a regression model to predict inventory risk:

[0129]

[0130] where is the predicted inventory risk indicator, and α, β1, and β2 are regression coefficients;

[0131] Step 1-4: Deploy a case server at the data collection layer. The case server obtains the occurrence times, impact scope, and severity of all suppliers' supply chain disconnection and logistics exception accidents in history from public accident case reports and industry research reports;

[0132] For each supplier, calculate the number of accident occurrences and the impact coefficient of each accident, and construct an accident risk scoring model. The specific formula is as follows:

[0133]

[0134] Among them, D i is the number of occurrences of the i-th accident; S i is the impact coefficient of the i-th accident, and S i takes values in the range of [1, 5], and T is the statistical time window;

[0135] In this embodiment, the value of S i is determined by the user himself / herself, and generally the user is the purchaser.

[0136] Step 1-5: Align each data source according to the supplier, country identifier, and time dimension, and obtain a unified data set through data merging operations;

[0137] In this embodiment, data alignment needs to align each data according to the supplier, country, and time (such as monthly, quarterly). Each record should contain the following fields: country / supplier identifier (such as ADO), political stability risk score R politics , public opinion risk score R conflict , inventory risk score and accident risk score R accident .

[0138] When unifying into a data set, the data merging method is adopted, such as the SQL JOIN method for merging. The data set should actually be a multi-dimensional data set.

[0139] Step 1-6: Calculate the comprehensive supplier stability index SSI, output the SSI score of each supplier, sort by the score, obtain a list of low-risk suppliers, and generate a list of alternative suppliers.

[0140] In this embodiment, a risk report of the country where the suppliers included therein are located can also be generated according to the instructions of the supply chain and presented in the form of a chart.

[0141] Step 2: Establish a discrete event simulation model in the analysis and decision-making layer, abstract the supply chain nodes as state variables, and input the comprehensive supplier stability index SSI and the predicted inventory risk as input data into the discrete event simulation model for simulation;

[0142] In the simulation, simulate events such as complete supply chain disconnection, partial supply interruption, and logistics delay of suppliers, simulate the emergency strategy response process, optimize the supplier switching strategy through a reinforcement learning model, generate a simulation report, and generate a supplier switching strategy;

[0143] When executing Step 2, establish a simulation server in the analysis and decision-making layer. The simulation server adopts the discrete event simulation DES method to simulate the operation status of the supply chain under partial interruption scenarios and evaluate the effect of recovery strategies, specifically including the following steps:

[0144] Step 2-1: The simulation server conducts target and problem definition. By defining supply chain nodes and state variables, a supply chain simulation scenario affected by external risk disturbances is constructed. The supply chain simulation scenario includes three types of interruption events: complete chain breakage, gray failure, and logistics delay.

[0145] The supply chain nodes include suppliers, logistics links, and inventory nodes.

[0146] The state variables include the supplier inventory level I, the order demand quantity D, the on-time delivery rate R, where R = R on-time and the comprehensive supplier stability index SSI.

[0147] The gray failure specifically refers to the decline in the supplier's supply capacity.

[0148] In this embodiment, the supplier inventory level I is calculated by using a prediction model:

[0149]

[0150] where I estimate represents the current inventory trend:

[0151] I estimate = γ 10 × D hist + γ 11 × S perf ;

[0152] D hist represents the delivery volume in the past period, which is specifically obtained by scraping from trade statistics and industry analysis reports. S perf represents the reliability score of the supplier, which is statistically obtained from customer evaluations. γ 10 and γ 11 are both weights. If the supplier had a large and stable delivery volume in the past, it means that its inventory management is good and there will be no inventory depletion in the short term.

[0153] Step 2-2: The simulation server conducts model design. Specifically, a discrete event simulation method is used to establish a dynamic model, including an inventory dynamic model, an event trigger model, and a key performance indicator model:

[0154] Inventory dynamic model:

[0155] I i (t + 1) = I i (t) + S i (t) - D i (t);

[0156] where I i (t) is the inventory level of the i-th supplier at time t;

[0157] S i The supply quantity of the \(i\)-th supplier at time \(t\) is:

[0158] S i (t) = S i,0 × f(SSI i , E(t));

[0159] Among them, S i,0 is the normal supply quantity, f(·) is a function affected by risk disturbances, and E(t) represents an event factor (such as a broken chain or a gray failure);

[0160] In this embodiment, in the case of no risk, the normal supply quantity of supplier \(i\) is S i,0 , but in actual situations, the supply quantity will be affected by the supply chain risk index SSI i and the external environmental factor E(t), making the actual supply quantity S i (t) lower than the normal level. Specifically, when in use, f(·) is a preset input value.

[0161] In this embodiment, the specific formula of f(·) is as follows:

[0162]

[0163] Among them, is the exponential decay model of SSI. The higher the SSI, the lower the supply quantity; λ represents the risk impact coefficient.

[0164] D′(t) represents the market demand index. If the demand rises, the supplier may try to maintain the supply, and this value is a fixed value preset by the user;

[0165] M′(t) represents the macroeconomic fluctuation. If the raw material price or exchange rate is unstable, it may affect the supply, and this value is a fixed value preset by the user;

[0166] L′(t) represents the logistics obstruction index. When the logistics is affected, the supply quantity decreases, and this value is a fixed value preset by the user.

[0167] α1, α2, and α3 are all regression coefficients.

[0168] D i (t) is the order quantity satisfied. If there is a logistics delay, the order fulfillment rate will decrease:

[0169] D i (t) = D0(t) × [1 - δ(t)];

[0170] Among them, D0(t) is the theoretical order quantity, and δ(t) is the order loss ratio caused by the logistics delay;

[0171] The event trigger model specifically uses the Poisson distribution to simulate event triggers, including disconnection events, grey failures, and logistics delay events:

[0172] Disconnection event: P(disconnection at t) = p 断链 ;

[0173] Grey failure: P(grey failure at t) = p 灰 ; If triggered, the supply capacity becomes λ × S i,0 , where λ is the failure attenuation coefficient;

[0174] Logistics delay event: P(delay at t) = p 延迟 ; If triggered, the delay time Δt is determined according to a preset distribution, such as a normal distribution;

[0175] Key performance indicator model:

[0176]

[0177] Among them, D 实际 (t) represents the actual number of orders satisfied at time t; D0(t) represents the total order demand at time t, and T represents the total statistical time period;

[0178] Record the time from the event trigger to when the inventory or order fulfillment rate recovers to the preset threshold to obtain the recovery time T 恢复 ;

[0179] Calculate the additional costs incurred due to disconnection or logistics delay to obtain the recovery cost RecoveryCost, specifically by calculating the emergency transportation cost or the additional procurement cost of alternative suppliers;

[0180] Step 2-3: The simulation server uses the reinforcement learning model for optimization, specifically including:

[0181] Define the state s(t): s(t) = {I i (t), D0(t), SSI i , L(t), E(t)};

[0182] Define the action space a(t), specifically including switching to an alternative supplier, adjusting the inventory scheduling strategy, and expediting the logistics of transportation;

[0183] Optimize through the reward function R(s, a):

[0184] R(s, a) = ServiceLevel - λ4 × RecoveryCost;

[0185] Among them, λ4 is the cost penalty coefficient;

[0186] Step 2-4: The simulation server inputs the comprehensive supplier stability index SSI, on-time delivery rate R on-time , order backlog rate R backlog , and predicted inventory risk indicators , the probabilities or statistical distribution parameters of chain breakage, grey failures, and logistics delays, runs the model in Step 2-2 for simulation, generates a simulation result report of the supply chain under various shock scenarios, and generates a supplier switching strategy based on the simulation result report, specifically including:

[0187] The recovery capabilities of each supplier and the overall service level of the supply chain;

[0188] Emergency strategy suggestions based on simulation and reinforcement learning, including backup supplier switching and inventory scheduling strategies;

[0189] Performance indicators, namely order fulfillment rate, recovery time, and recovery cost.

[0190] Step 3: Construct a multi-objective optimization model aiming to minimize the recovery cost, maximize the service level, and maximize the supply chain resilience in the analysis and decision-making layer, optimize the supplier switching strategy, and generate a list of recommended backup suppliers;

[0191] Update the data source in real time, update the comprehensive supplier stability index SSI, and feedback it to the discrete event simulation model and the multi-objective optimization model to generate an optimized strategy;

[0192] When executing Step 3, deploy a policy server in the analysis and decision-making layer. The policy server constructs a model of the multi-objective function, specifically including the following steps:

[0193] Step 3-1: Construct a recovery cost function, including logistics expediting cost, inventory shortage or surplus cost, and backup supplier startup cost. The specific formula is as follows:

[0194]

[0195] Among them, L(x, t) represents the logistics cost in the t-th period, S(x, t) represents the supplier switching cost, Cost(x) represents the recovery cost, c logistics , c inventory and c switch are all cost coefficients, x is the supplier switching strategy, T is the time window; I(t) is the current inventory, and I target (t) is the target inventory level;

[0196] Step 3-2: Construct a service level function ServiceLevel(x) to measure the order fulfillment rate and delivery timeliness. The specific formula is as follows:

[0197]

[0198] Among them, D fulfilled (x, t) represents the actual order volume satisfied in the t-th period under the strategy x; D0(t) represents the theoretical order demand;

[0199] Step 3-3: Construct a supply chain resilience index model Resilience(x), which is specifically defined as the time required to resume normal operation after a shock or the degree of improvement of the system state. The formula is as follows:

[0200]

[0201] Among them, T recovery (x) represents the time required to resume the preset service level under the measurement x. The shorter the recovery time, the higher the resilience index;

[0202] Step 3-4: Construct a model of a multi-objective function. The specific formula is as follows:

[0203] min f(x) = w7 × Cost(x) + w8 × ServiceLevel(x) + w9 × Resilience(x);

[0204] Among them, w7, w8, and w9 are all weights;

[0205] Step 3-5: Solve according to the model of the multi-objective function to obtain the optimal decision x and generate a list of recommended backup suppliers.

[0206] Step 4: Display the comprehensive supplier stability index SSI and the list of recommended backup suppliers in the form of visual charts on the application display layer.

[0207] A supply chain resilience optimization method based on multi-objective optimization according to the present invention solves the technical problem of dynamically optimizing the supply chain recovery strategy through the collaboration of multi-source data-driven and intelligent algorithms to achieve the multi-objective balance of cost, service, and resilience. The present invention can integrate multi-dimensional data of politics, public opinion, operation, and accidents, construct a composite stability index SSI, realize forward-looking risk prediction, train emergency strategies in a virtual environment based on a simulation-reinforcement learning hybrid framework, reduce the actual trial-and-error cost, and generate an optimal recovery plan through a cost-service-resilience comprehensive objective function and a dynamic programming algorithm.

Claims

1. A supply chain resilience optimization method based on multi-objective optimization, characterized in that: The steps are as follows: Step 1: Obtain data sources from the Internet at the data collection layer, including political stability data, international conflict and war public opinion data, public enterprise operation data, and accident case data; Clean and preprocess the data source at the data collection layer to obtain the standardized political stability risk score R politics , public opinion risk score R conflict , inventory risk score and accident risk score R accident ; Calculate the comprehensive supplier stability index SSI, and the specific formula is as follows: Among them, w1, w2, w3, and w4 are all weights; Step 2: Establish a discrete event simulation model at the analysis and decision-making layer. Abstract the supply chain nodes as state variables, and input the comprehensive supplier stability index SSI and the predicted inventory risk as input data into the discrete event simulation model for simulation; In the simulation, simulate events such as complete supplier chain breakage, partial supply interruption, and logistics delay, simulate the emergency strategy response process, optimize the supplier switching strategy through a reinforcement learning model, generate a simulation report, and generate a supplier switching strategy; Step 3: Construct a multi-objective optimization model with the goal of minimizing recovery costs, maximizing service levels, and supply chain resilience at the analysis and decision-making layer, optimize the supplier switching strategy, and generate a list of recommended alternative suppliers; Update the data source in real time, update the comprehensive supplier stability index SSI, and feedback it to the discrete event simulation model and the multi-objective optimization model to generate an optimized strategy; Step 4: Display the comprehensive supplier stability index SSI and the list of recommended alternative suppliers in the form of visual charts at the application display layer.

2. The supply chain resilience optimization method based on multi-objective optimization according to claim 1, characterized in that: When implementing Step 1, it specifically includes the following steps: Step 1-1: Deploy a political data server at the data collection layer. The political data server downloads the.xlsx format index file of the global governance indicators from the Internet, which includes 6 governance indicators of VA, PV, GE, RQ, RL, and CC for each country; The political data server obtains the profile information of all suppliers input by the user through the client. The profile information contains the regions and countries where the suppliers are located; Extract the country where any one supplier is located, obtain the 6 governance indicators of this country in the index file, and calculate the standardized risk score for each indicator. The specific calculation formula is as follows: Among them, R i represents the standardized score of the i-th indicator, where i represents the number of any one of the 6 governance indicators; Estimate i is the Estimate value of the i-th indicator, and Comparator i is the Comparator reference value of the i-th indicator; σ i represents the historical standard deviation; Generate a political score data table for all countries where the suppliers are located according to the calculation results of the standardized risk scores; Step 1-2: Deploy a public opinion processing server at the data collection layer. The public opinion processing server grabs news texts containing keywords through Google Alerts, NewsAPI, or Twitter API. The keywords include sanction, tariff, or export ban; Set a preset time window, count the frequency f of the keyword appearing within the preset time window, perform normalization calculation, and obtain the public opinion risk score R conflict : where f is the number of occurrences of the keyword within a preset time period, f min and f max are the minimum and maximum frequencies during the observation period, respectively; Step 1-3: Deploy an enterprise data server at the data collection layer. The enterprise data server obtains the on-time delivery rate R and the order backlog rate R of all suppliers from public industry reports, news reports, and public delivery records on the Internet, and constructs an enterprise historical data set; on-time and the order backlog rate R backlog , and constructs an enterprise historical data set; Sort out the historical data in the enterprise historical dataset, calculate the time series average and change rate of each indicator. Specifically, use the publicly available on-time delivery rate and order backlog rate to construct a regression model to predict inventory risk: Among them, is the predicted inventory risk index, and α, β1, and β2 are regression coefficients; Step 1-4: Deploy a case server at the data collection layer. The case server obtains the occurrence times, influence scope, and severity of supply chain breakage and logistics anomalies of all suppliers in history from public accident case reports and industry research reports; For each supplier, calculate the number of accident occurrences and the impact coefficient of each accident, and construct an accident risk scoring model. The specific formula is as follows: Among them, D i is the number of occurrences of the i-th accident; S i is the impact coefficient of the i-th accident, and the value of S i is in the range of [1, 5], and T is the statistical time window; Step 1-5: Align each data source according to the dimensions of suppliers, country identifiers, and time. Through data merging operations, obtain a unified data set. Step 1-6: Calculate the comprehensive supplier stability index SSI, output the SSI score for each supplier, sort by the score, obtain a list of low-risk suppliers, and generate a list of alternative suppliers.

3. The supply chain resilience optimization method based on multi-objective optimization according to claim 2, wherein: When performing Step 2, establish a simulation server at the analysis and decision-making level. The simulation server uses the discrete event simulation DES method to simulate the operation status of the supply chain under partial interruption scenarios and evaluate the effectiveness of recovery strategies. The specific steps are as follows: Step 2-1: The simulation server conducts target and problem definition. By defining supply chain nodes and state variables, construct a supply chain simulation scenario affected by external risk disturbances. The supply chain simulation scenario includes three types of interruption events: complete chain breakage, gray failure, and logistics delay. Supply chain nodes include suppliers, logistics links, and inventory nodes. The state variables include the supplier inventory level I, the order demand quantity D, the on-time delivery rate R, where R = R on-time and the comprehensive supplier stability index SSI; Gray failure specifically refers to the decline in the supply capacity of suppliers. Step 2-2: The simulation server conducts model design. Specifically, use the discrete event simulation method to establish a dynamic model, including an inventory dynamic model, an event trigger model, and a key performance indicator model: Inventory dynamic model: I i (t + 1) = I i (t) + S i (t) - D i (t); Among them, I i (t) is the inventory level of the i-th supplier at time t; S i (t) is the supply volume of the i-th supplier at time t: S i (t) = S i,0 × f(SSI i , E(t)); Among them, S i,0 is the normal supply volume, f(·) is the function affected by risk disturbances, and E(t) represents the event factor (such as broken chain or gray failure); D i (t) is the order volume to be satisfied. If there is a logistics delay, the order fulfillment rate will decrease: D i (t) = D0(t) × [1 - δ(t)]; Among them, D0(t) is the theoretical order volume, and δ(t) is the order loss ratio caused by logistics delay. The event trigger model specifically uses the Poisson distribution to simulate event triggers, including chain breakage events, gray failures, and logistics delay events: Disconnection event: P(disconnection att) = p 断链 ; Grey fault: P(Grey fault att) = p 灰 ; If triggered, the supply capacity becomes λ × S i,0 , where λ is the fault attenuation coefficient; Logistics delay event: P(delay_att) = p 延迟 ; If triggered, the delay time Δt is determined according to a preset distribution (such as a normal distribution); Key performance indicator model: Among them, D 实际 (t) represents the number of orders actually fulfilled at time t; D0(t) represents the total order demand at time t, and T represents the total statistical time period; Record the time from the event trigger to when the inventory or order fulfillment rate resumes to the preset threshold to obtain the recovery time T 恢复 ; Calculate the additional costs generated by chain breakage or logistics delay to obtain the recovery cost RecoveryCost. Specifically, calculate the emergency transportation cost or the additional procurement cost of alternative suppliers. Step 2-3: The simulation server uses a reinforcement learning model for optimization, specifically including: Define the state s(t): s(t) = {I i (t), D0(t), SSI i , L(t), E(t)}; Define the action space a(t), which specifically includes switching to alternative suppliers, adjusting the inventory scheduling strategy, and expediting transportation logistics. Optimize through the reward function R(s, a): R(s, a) = ServiceLevel - λ4 × RecoveryCost; Among them, λ4 is the cost penalty coefficient. Step 2-4: The simulation server inputs the comprehensive supplier stability index SSI, delivery on-time rate R on-time , order backlog rate Rb acklog , predicted inventory risk indicators , the probabilities or statistical distribution parameters of chain breakage, grey failures, and logistics delays, runs the model in Step 2-2 for simulation, generates a simulation result report of the supply chain under various shock scenarios, and generates a supplier switching strategy based on the simulation result report, specifically including: The recovery capabilities of each supplier and the overall service level of the supply chain; Emergency strategy recommendations based on simulation and reinforcement learning, including alternative supplier switching and inventory scheduling strategies; Performance indicators, namely order fulfillment rate, recovery time, and recovery cost.

4. The supply chain resilience optimization method based on multi-objective optimization according to claim 3, characterized in that: When performing Step 3, deploy a strategy server at the analysis and decision-making level. The strategy server constructs a model with multi-objective functions. The specific steps are as follows: Step 3-1: Construct a recovery cost function, including logistics expediting costs, inventory shortage or surplus costs, and alternative supplier startup costs. The specific formula is as follows: Among them, L(x, t) represents the logistics cost in the t-th period, S(x, t) represents the supplier switching cost, Cost(x) represents the recovery cost, c logistics , c inventory and c switch are all cost coefficients, x is the supplier switching strategy, T is the time window; I(t) is the current inventory, I target (t) is the target inventory level; Step 3-2: Construct a service level function ServiceLevel(x) to measure the order fulfillment rate and delivery timeliness. The specific formula is as follows: Among them, D fulfilled (x, t) represents the actual order volume satisfied at the t-th period under the strategy x; D0(t) represents the theoretical order demand; Step 3-3: Construct a supply chain resilience index model Resilience(x), which is specifically defined as the time required to recover to normal operation after an impact or the degree of improvement in the system state. The formula is as follows: Among them, T recovery (x) represents the time required to restore to the preset service level under the measurement of x. The shorter the recovery time, the higher the resilience index; Step 3-4: Construct a model with multi-objective functions. The specific formula is as follows: min f(x) = w7 × Cost(x) + w8 × ServiceLevel(x) + w9 × Resilience(x); where w7, w8, and w9 are all weights; Step 3 - 5: Solve according to the multi - objective function model to obtain the optimal decision x and generate a list of recommended alternative suppliers.

Images