Power distribution network supply chain dynamic scheduling method considering internal and external lead period time lag

Abstract

The invention provides a power distribution network supply chain dynamic scheduling method considering internal and external lead period time lag, and relates to the technical field of supply chain management. The method comprises the following steps: according to a preset lead feature library, obtaining lead feature data of various electric power material supply links in a supply chain; according to the conduction characteristic data, predicting conduction probability distribution of each type of electric power material supply links, and according to the conduction probability distribution, generating multiple groups of future conduction scenes of each type of electric power materials through a Monte Carlo method; establishing a day-ahead scheduling model and solving the day-ahead scheduling model according to multiple groups of future lead scenes of each type of electric power materials to obtain a day-ahead scheduling scheme of each scheduling period of the next day; and based on the day-ahead scheduling scheme, in each scheduling period of the next day, establishing an intra-day scheduling model considering the demand scene priority and solving the intra-day scheduling model to obtain a period-level dynamic scheduling scheme. According to the invention, accurate real-time matching between the material supply plan and the power grid demand can be realized, and the problem of plan failure caused by conduction uncertainty and time-varying characteristics is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of supply chain management, and in particular to a power distribution network supply chain dynamic scheduling method considering internal and external lead time lags. BACKGROUND

[0002] The stable operation of a power distribution network depends on efficient and accurate material supply. The supply chain of power materials (such as transformers, cables, circuit breakers, etc.) is long and involves multiple stages, including factory production and cross-regional transportation. The completion time of each stage (i.e., lead time) is affected by factors such as production capacity, logistics, and environment, and is subject to significant uncertainty and time variation (i.e., internal and external lead time lags). This time lag causes material plans based on fixed cycles and average lead times to easily deviate from project needs during actual execution, leading to two problems: first, material shortages due to delayed deliveries, affecting power grid project progress; second, excessive inventory due to overly advanced plans, occupying a large amount of capital and storage resources.

[0003] In existing technologies, supply chain planning often uses material requirement planning or regular replenishment strategies. These methods are essentially static and cannot perceive or respond to dynamic changes in lead times at various stages of the supply chain. Although some solutions incorporate optimization algorithms, they are still based on deterministic assumptions and lack effective mechanisms to address the core uncertainty of time lags, making it difficult to achieve fast and flexible management goals. SUMMARY

[0004] Embodiments of the present application provide a power distribution network supply chain dynamic scheduling method considering internal and external lead time lags to achieve accurate real-time matching between material supply plans and power grid needs, solving the problem of plan failure caused by lead time uncertainty and time variation.

[0005] In a first aspect, embodiments of the present application provide a power distribution network supply chain dynamic scheduling method considering internal and external lead time lags, comprising: According to a predetermined lead time characteristic library, obtain lead time characteristic data of various power material supply stages in the supply chain; According to the lead time characteristic data, predict the lead time probability distribution of each type of power material supply stage, and according to the lead time probability distribution, generate multiple future lead time scenarios for each type of power material by the Monte Carlo method; According to the multiple future lead time scenarios of each type of power material, establish a day-ahead scheduling model and solve it to obtain a day-ahead scheduling scheme for each scheduling period the next day; Based on the day-ahead scheduling scheme, in each scheduling period the next day, establish an intra-day scheduling model considering demand scenario priority and solve it to obtain a period-level dynamic scheduling scheme.

[0006] In a possible implementation, the obtaining of the lead-time characteristic data of each type of power material supply link in the supply chain comprises: obtaining lead-time characteristics of a procurement link, lead-time characteristics of a production link and lead-time characteristics of a transportation link of each type of power material in the supply chain; The lead-time characteristics of the procurement link comprise historical order production cycle, raw material supply stability data and capacity utilization rate trend data. The lead-time characteristics of the production link comprise process standard man-hours, device historical operation state sequence and quality inspection pass rate statistical information. The lead-time characteristics of the transportation link comprise passing time distribution of typical transportation routes in different seasons, influence record of weather conditions on transportation efficiency and in-transit material trajectory historical data.

[0007] In a possible implementation, the predicting of the lead-time probability distribution of each type of power material supply link according to the lead-time characteristic data comprises: According to the lead-time characteristic data, the influence law and fluctuation characteristics of the lead-time in each type of power material supply link are identified through a time series analysis method. According to the influence law and fluctuation characteristics of the lead-time, the lead-time probability distribution of each type of power material supply link is predicted through a pre-trained short-term prediction model.

[0008] In a possible implementation, the generating of multiple groups of future lead-time scenarios of each type of power material according to the lead-time probability distribution through a Monte Carlo method comprises: According to the lead-time probability distribution, multiple groups of initial future lead-time scenarios are obtained through random sampling of the Monte Carlo method. The initial future lead-time scenarios are clustered through a K-medoids algorithm, high-probability future lead-time scenarios with cumulative probability reaching a preset threshold are selected, and each high-probability future lead-time scenario is assigned a post-clustering probability, to obtain multiple groups of future lead-time scenarios of each type of power material.

[0009] In a possible implementation, the establishing of a day-ahead scheduling model and the solving of the day-ahead scheduling model according to the multiple groups of future lead-time scenarios of each type of power material to obtain a day-ahead scheduling scheme of each scheduling period of the next day comprises: According to low-probability future lead-time scenarios other than the high-probability future lead-time scenarios in the initial future lead-time scenarios, a compensation term is generated. According to the multiple groups of future lead-time scenarios of each type of power material, a day-ahead scheduling objective function is established, with the sum of reserve scheduling cost, response timeliness cost and the compensation term being minimized as the target. The constraint condition of the day-ahead scheduling objective function is established, the day-ahead scheduling objective function is solved based on the constraint condition, and a day-ahead scheduling scheme of each scheduling period of the next day is obtained.

[0010] In a possible implementation, the day-ahead scheduling scheme is used to establish and solve an intra-day scheduling model considering demand scenario priority in each scheduling period of the next day, so as to obtain a period-level dynamic scheduling scheme, including: In each scheduling period of the next day, a day-ahead scheduling objective function is established, with the sum of reserve scheduling cost, response timeliness cost, delay amount penalty cost, scheduling penalty cost, maximum delay penalty cost and the compensation term being minimized as the target; The resource reservation constraint and the adjustment amplitude deviation constraint are established according to the day-ahead scheduling scheme; The demand time limit constraint is established according to the demand scenario priority; The day-ahead scheduling objective function is solved based on the resource reservation constraint, the adjustment amplitude deviation constraint and the demand time limit constraint, and a period-level dynamic scheduling scheme is obtained.

[0011] In a possible implementation, the constraint condition of the day-ahead scheduling objective function includes: The material type matching constraint: for any demand scenario, the type of the scheduled power material is consistent with the type of the power material required by the demand scenario; The inventory capacity constraint: at any time, the inventory of each type of power material in any warehouse does not exceed the maximum storage capacity of the corresponding warehouse; The safety inventory constraint: at any time, the inventory of each type of power material in any warehouse is greater than the corresponding minimum safety inventory; The inventory update constraint: the inventory of each type of power material satisfies the timing balance.

[0012] In a possible implementation, the demand time limit constraint is established according to the demand scenario priority, including: Obtain evaluation index data of each demand scenario; wherein, the evaluation indexes include project emergency degree, planned construction time, progress completion rate and demand reporting time; Determine the weight of each evaluation index by the analytic hierarchy process; According to the evaluation index data of each demand scenario and the weight of each evaluation index, calculate the priority score of each demand scenario; Based on the priority score of each demand scenario, establish the demand time limit constraint.

[0013] In a possible implementation, the day-ahead scheduling objective function includes:

[0014] in, The number of future guidance scenarios; For future guidance scenarios s The probability of; These are the weighting coefficients; Cost of power material reserves; Costs related to power material dispatching; To address the time and cost of response; As compensation, For the compensation term coefficient, Compensation value.

[0015] In one possible implementation, the intraday scheduling objective function includes:

[0016] in, For a specific moment; Cost of power material reserves; Penalty cost for delay; Costs related to power material dispatching; To adjust the cost of penalties; To address the time and cost of response; The maximum delay penalty cost; As compensation, For the compensation term coefficient, Compensation value.

[0017] In a second aspect, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0018] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: In this embodiment of the invention, the probability distribution of the lead time for various power supply links in the supply chain is predicted by using lead time characteristic data. Multiple future lead time scenarios for each type of power supply are generated using the Monte Carlo method, simulating various possible lead time fluctuation scenarios and their probabilities of occurrence. This provides more comprehensive and realistic data support for subsequent optimization decisions. Furthermore, a two-layer optimization architecture of "day-ahead coarse scheduling - intraday fine-tuning" is established. Day-ahead, a day-ahead scheduling model is established and solved based on multiple future lead time scenarios for each type of power supply to obtain the day-ahead scheduling scheme for each scheduling period of the next day. Intraday, based on the day-ahead scheduling scheme, an intraday scheduling model considering the priority of demand scenarios is established and solved for each scheduling period to obtain a time-period-level dynamic scheduling scheme. This invention achieves an organic combination of global optimization and local real-time response, enabling the system to maintain long-term low cost while quickly responding to random demand matching situations, solving the problem of traditional methods struggling to balance low cost and timeliness. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the implementation of a dynamic scheduling method for the distribution network supply chain that considers the time delay of internal and external conduction periods, provided by an embodiment of the present invention. Figure 2 This is a flowchart illustrating the implementation of a dynamic scheduling method for the distribution network supply chain that considers the time delay of internal and external conduction periods, provided by another embodiment of the present invention. Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] In the supply chain management of power distribution networks, a crucial issue is how to achieve precise and real-time matching between supply plans and dynamically changing power grid operation and maintenance and engineering construction needs. This is crucial to address the frequent plan failures caused by the uncertainty and time-varying nature of procurement, production, and transportation. The challenges in solving this technical problem include: 1. Failure of rigid planning: When power grid engineering tasks generate demand for power materials, static plans based on fixed cycles cannot respond quickly, resulting in the failure of key materials to arrive in a timely manner, affecting the progress of the project.

[0022] 2. Inadequate handling of uncertainties: Frequent events such as supplier production delays and transportation disruptions have resulted in a lack of quantitative analysis and effective response mechanisms for these uncertainties using traditional methods.

[0023] 3. Difficulty in multi-stage coordination: There are complex time correlations among the procurement, production, and transportation stages. Fluctuations in the lead time of one stage can have a chain reaction. Existing systems lack the ability to coordinate and optimize such spatiotemporal correlations.

[0024] Figure 1 A schematic diagram illustrating the implementation process of a dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays, as provided in an embodiment of the present invention, is described in detail below: Step 101: Obtain the timing characteristic data of various power supply links in the supply chain based on the preset timing characteristic library.

[0025] This embodiment decomposes the internal and external conduction periods of power materials in the procurement, production, and transportation stages, and constructs a conduction period feature library through multi-source data fusion technology.

[0026] Procurement process lead time feature collection: Integrate supplier historical cooperation data to establish supplier profiles that include historical order scheduling cycles, raw material supply stability data, and capacity utilization trend data.

[0027] Production process characteristic collection: This is achieved by collecting historical operation records of the production line, including standard working hours for each process, historical operating status sequences of equipment, and statistical information on quality inspection pass rates.

[0028] Transportation process characteristic collection: Collect historical data of logistics transportation, covering the travel time distribution of typical transportation routes in different seasons, the impact of weather conditions on transportation efficiency, and historical data of the trajectory of goods in transit.

[0029] Step 102: Based on the guidance period characteristic data, predict the guidance period probability distribution of various power supply links, and generate multiple sets of future guidance period scenarios for each type of power supply using the Monte Carlo method based on the guidance period probability distribution.

[0030] In this embodiment, based on collected multi-dimensional historical data, the influence patterns and fluctuation characteristics of each stage's lead time are first identified using time series analysis. Then, based on these patterns and characteristics, a pre-constructed short-term prediction model is used to output the probability distribution of the lead time. Here, the short-term prediction model can be any deep learning model, and this embodiment does not limit it.

[0031] Then, using the Monte Carlo method, random sampling is performed based on the probability distribution to generate multiple sets of possible future scenarios. The occurrence probability of each scenario is quantified by the sampling frequency, providing input basis that covers uncertainty for subsequent optimization decisions.

[0032] In one possible implementation, there are many probability values ​​for the scenarios output by Inmont Carlo. In order to optimize the accuracy of scheduling, scenario reduction technology can be used to extract core scenarios and their probabilities as probabilistic inputs for subsequent optimization decisions: future guidance scenarios are clustered by K-medoids algorithm, high-probability future guidance scenarios with accumulated probabilities reaching a preset threshold are selected, and clustered probabilities are assigned to each high-probability future guidance scenario to obtain multiple sets of future guidance scenarios for each type of power material.

[0033] For example, the K-medoids algorithm is first used to cluster the 500 scenes output by Monte Carlo, and typical scenes with a cumulative probability of 85% are selected. Each scene is assigned a probability weight after clustering. Then, a compensation term can be generated separately for the remaining 15% of low-probability scenes as part of the optimization objective function to enhance the robustness of the optimization scheme.

[0034] Step 103: Based on multiple future guidance scenarios for each type of power material, establish and solve the day-ahead scheduling model to obtain the day-ahead scheduling scheme for each scheduling period of the next day.

[0035] At the initial coarse scheduling layer, the model focuses on "dual-objective collaborative optimization," namely minimizing reserve scheduling costs (specifically including reserve costs and allocation costs) and maximizing the responsiveness of material matching. This layer uses a minimum scheduling period of one hour and, based on scenario feature prediction data acquired 24 hours in advance, determines the baseline inventory allocation, regular transportation routes, and capacity allocation schemes for each time period of the following day by solving the model. During the solution process, multiple constraints must be strictly met: in terms of capacity, the production / outbound volume of materials at each supply point must not exceed its maximum daily processing capacity; in terms of inventory, the inventory level of each warehouse after allocation must be maintained above a safe threshold; and in terms of transportation, the average daily capacity input for each route must not exceed the road's carrying capacity limit. The initial scheme output by this layer not only defines a rigid framework for intraday fine-tuning but also lays a feasible foundation for intraday adjustments by reserving three types of flexibility.

[0036] In one possible implementation, the day-ahead scheduling objective function is as follows:

[0037] in, The number of future guidance scenarios, For future guidance scenarios s The probability will be used to number the scene. These are weighting coefficients used to balance economy and timeliness. The larger the value, the more emphasis is placed on cost control; the smaller the value, the more emphasis is placed on timeliness. This can be dynamically adjusted according to the needs of the scenario. The objective function controls resource allocation through cost terms and enhances process efficiency and response speed through time penalty terms, ultimately achieving a low-cost, fast-response dynamic scheduling optimization goal.

[0038] For the cost of power material reserves, For the duration of this material's storage, The fee rate for holding electrical materials per unit of time.

[0039] To reduce the cost of power material dispatching, For scheduling resource types, For the dispatch area, For scheduling time window, For variables of 0-1, 1 represents the state of being in the middle of the middle. t Time window, first j Regional scope, call number i The scheduling behavior of this type of resource has occurred; 0 indicates that it has not occurred. Fixed costs for basic scheduling, A fixed fee is added for the time. A fixed fee is added to the area. Additional fixed fees are charged for different types of electrical equipment.

[0040] In order to respond to time-sensitive costs, In response to the timeliness and cost threshold, This is the timeout penalty coefficient. This is the time interval from when a request is triggered to when processing begins.

[0041] As compensation, For the compensation term coefficient, Compensation value.

[0042] In one possible implementation, the constraints of the day-ahead scheduling objective function include: (1) Material type matching constraints: At any time Targeting the demand scenarios k The type of power materials dispatched must be completely consistent with the type of demand. Because power materials are highly specialized, strict matching is required to avoid the dispatched materials being unusable. For a set of demand scenarios, for t Moment Scene k The demand for power supplies; Scheduled to the scene at time t k Power supplies, This is a type determination function.

[0043] (2) Inventory capacity constraints: Any warehouse ,time A certain type of material The inventory level must not exceed the warehouse's maximum storage capacity. For warehouse collection, A collection of material types, for t Time Warehouse Storage materials Maximum capacity.

[0044] (3) Safety stock constraint: To cope with random demand, warehouses must maintain a minimum safety stock to avoid the inability to dispatch power supplies due to zero inventory. For power materials In the warehouse The minimum safety stock level.

[0045] (4) Inventory update constraints: Inventory levels must meet time-series balance requirements. Let be the amount of goods received at time t. for t Real-time outbound volume.

[0046] Step 104: Based on the previous day's scheduling scheme, establish and solve the intraday scheduling model that takes into account the priority of demand scenarios for each scheduling period of the next day to obtain the time period-level dynamic scheduling scheme.

[0047] In the intraday fine-tuning layer, based on real-time updated scenario information, including actual demand, forecast deviations, traffic status feedback from transportation routes, and short-term weather conditions, a fine-tuning is performed on the day-ahead coarse scheduling. This layer introduces a scenario priority evaluation mechanism, constructing a priority quantification matrix based on the scenario's impact range, demand time limit, and loss rate per unit delay time, based on real-time updated information. Power resources are allocated for high-priority scenarios. At the same time, this layer introduces minimizing delay penalty costs and coordination costs on top of the day-ahead objective function. The optimization objectives of this layer include minimizing scheduling costs, delay penalty costs, and coordination costs, and a rolling optimization mechanism is used to dynamically adjust and finely calibrate short-term scheduling plans to achieve optimal economy and timeliness of the output scheduling scheme.

[0048] In one possible implementation, the intraday scheduling objective function is as follows:

[0049] in, To incur penalties for delays, To delay delivery volume, To delay the duration, This is the penalty coefficient that increases with the duration of the delay.

[0050] To adjust the cost of punishment, To determine the extent of the planned adjustment, To adjust the frequency, To adjust the magnitude penalty coefficient, To adjust the frequency penalty coefficient.

[0051] To maximize the cost of delay penalty, This is the delay penalty coefficient. This represents the maximum delay duration.

[0052] In one possible implementation, the constraints of the intraday scheduling objective function include: (1) Resource reservation constraints based on the day-ahead scheduling scheme, i.e., the deviation between the actual daily scheduling quantity and the day-ahead planned quantity must be within the allowable range: .

[0053] (2) Adjustment deviation constraints based on the day-ahead scheduling scheme, i.e., the day-ahead plan must reserve certain resources for intraday scheduling: .

[0054] (3) Demand time constraints are established based on the priority of demand scenarios. Priorities are quantified and filtered by impact scope, demand time limit, and delay loss rate. High-value and high-priority scenarios are listed as core optimization objects, while medium and low-priority scenarios are listed as secondary objects. The demand time limit requirements of priority scenarios (such as high-priority scenarios needing to be delivered within 2 hours) are directly converted into hard constraints of the objective function. When solving, the timeliness threshold of high-priority scenarios must be met; otherwise, the solution is considered infeasible.

[0055] In this embodiment of the invention, the probability distribution of the lead time for various power supply links in the supply chain is predicted by using lead time characteristic data. Multiple future lead time scenarios for each type of power supply are generated using the Monte Carlo method, simulating various possible lead time fluctuation scenarios and their probabilities of occurrence. This provides more comprehensive and realistic data support for subsequent optimization decisions. Furthermore, a two-layer optimization architecture of "day-ahead coarse scheduling - intraday fine-tuning" is established. Day-ahead, a day-ahead scheduling model is established and solved based on multiple future lead time scenarios for each type of power supply to obtain the day-ahead scheduling scheme for each scheduling period of the next day. Intraday, based on the day-ahead scheduling scheme, an intraday scheduling model considering the priority of demand scenarios is established and solved for each scheduling period to obtain a time-period-level dynamic scheduling scheme. This invention achieves an organic combination of global optimization and local real-time response, enabling the system to maintain long-term low cost while quickly responding to random demand matching situations, solving the problem of traditional methods struggling to balance low cost and timeliness.

[0056] In some embodiments, demand time constraints are established based on demand scenario priorities, including: Obtain evaluation indicator data for various demand scenarios; among which, the evaluation indicators include the urgency of the project, the planned start time, the progress completion rate, and the time of demand submission. The weights of each evaluation indicator are determined using the analytic hierarchy process (AHP). Based on the evaluation index data and the weight of each evaluation index for each demand scenario, calculate the priority score for each demand scenario. Based on the priority scoring of each demand scenario, demand time constraints are established.

[0057] In this embodiment, the core of establishing a power material supply priority ranking mechanism lies in constructing a scientific indicator evaluation system and a dynamic adjustment algorithm. This mechanism is based on four dimensions: project urgency, planned start time, progress completion rate, and demand submission time. Through quantitative scoring, weight allocation, and comprehensive calculation, it achieves an objective ranking of all material demands. Simultaneously, a dynamic adjustment mechanism is established for non-routine demands, possessing real-time response capabilities. It can dynamically adjust the priority order based on random events, changes in project duration, and other factors, ensuring that critical projects and critical demands are always given priority.

[0058] (1) The urgency of the project is the primary basis for prioritization, and this indicator is quantitatively evaluated from two levels. First, the project type is classified, and a basic score is assigned based on its impact on power grid security and power supply reliability. Second, the scope of impact is assessed by calculating the scope of impact index by statistically analyzing factors such as the number of affected users, the proportion of power outage load, and the level of important users involved. Finally, these two sub-indicators are weighted and summed according to their respective weight ratios to obtain a comprehensive score for the urgency of the project.

[0059] Definition of project urgency level index: ; in, For project type, This is the influence range index.

[0060] ; ; in, Number of affected users; For the affected load; This represents the total load of the region.

[0061] (2) The planned start date indicator reflects the time urgency of material needs. Its core idea is that the closer to the start date, the higher the priority should be. This indicator is modeled using an exponential decay function. Specifically, the difference in the number of days between the current date and the planned start date is used as the independent variable, and the score is calculated using a negative exponential function. When a project has already started late, the indicator score is set to a maximum of 100 points, representing the highest priority. For projects that are about to start, the score decreases rapidly as the remaining days increase. This exponential decay model conforms to the scheduling principle of proximity priority in actual management. In order to more accurately reflect the actual situation, this indicator also introduces a schedule constraint correction factor. For projects with a high proportion of critical paths and high schedule pressure, a certain bonus will be given on top of the base score to reflect their importance to the overall progress.

[0062] Time urgency function: ; in, ; The current date; The planned start date; This is a correction factor for project duration constraints. These are the initial fractional coefficients; This is the decay rate parameter; To correct the factor weight coefficients; The planned start date; This is the time sensitivity coefficient; This represents the number of days remaining until the planned start date.

[0063] (3) The progress completion rate indicator measures the project execution status and the need for expedited work. It not only calculates the basic completion rate (completed amount / planned amount) but also conducts progress deviation analysis, assigning different priority weights to projects based on whether they are ahead of schedule, on schedule, behind schedule, or severely behind schedule (lagging projects have higher weights). The indicator also calculates the urgency of expedited work (based on the target gap and the remaining time, it assesses the amount that needs to be caught up per unit of time; the higher the score, the greater the schedule pressure). The final score is derived by combining the weighted basic completion rate score (accounting for 60%) and the urgency of expedited work score (accounting for 40%), which both incentivizes progress according to plan and prioritizes lagging projects that truly need acceleration.

[0064] Basic completion rate: ; This represents the amount of work already completed. This refers to the planned workload.

[0065] Schedule deviation: ; The target completion rate.

[0066] Weighting coefficients: .

[0067] Urgency of completing the project: ; The remaining construction period.

[0068] Base score: .

[0069] Urgency score: , This is the adjustment coefficient for the urgency score, which represents the magnitude of the impact of the urgency of the task on the final score.

[0070] Final score: .

[0071] (4) The demand submission time indicator reflects the first-come, first-served principle, protecting the rights and interests of projects submitted early. It uses the earliest submission time as the benchmark, assigning points linearly according to the submission time (the earlier the submission, the higher the score). To prevent excessive favoritism towards early demands, a mild daily attenuation deduction (usually 0.5-1 points / day) is designed, along with waiting time compensation (unmet demands accumulate points based on the number of waiting days). Furthermore, projects with frequent plan changes are penalized with a deduction (5-10 points per instance) to maintain the seriousness of demand management. This indicator has a relatively small weight in the overall evaluation, but it is crucial for ensuring fairness in the supply order.

[0072] Time decay score: ; For the earliest time the request is submitted, For the project Submission time, This represents the daily attenuation coefficient.

[0073] Waiting for compensation: ; For the number of days to wait, This is the bonus coefficient for waiting.

[0074] Change of penalty: ; For the number of times the plan is changed, This is the deduction coefficient for a single change.

[0075] Final score: .

[0076] The constructed comprehensive scoring model for prioritizing power material demand uses the analytic hierarchy process (AHP) to determine the weighting system of indicators. It integrates the quantitative evaluation results of four core dimensions through a weighted summation algorithm: First, scores for each dimension are calculated based on the urgency of the project, planned start time, progress completion rate, and demand submission time. Then, weighted summation is performed at 47%, 28%, 16%, and 9% respectively to generate a comprehensive score from 0 to 100. Finally, based on the score, demand is divided into four priority levels: immediate delivery, priority production, routine supply, and deferred processing. A dynamic adjustment mechanism is established to apply a 1.5x weighting to critical projects and a 0.8x suppression to duplicate submissions, achieving precise matching and orderly supply of power materials.

[0077] In some embodiments, the model is solved by using a meta-reinforcement learning framework to generate and execute flexible scheduling schemes based on a dynamic scheduling optimization model.

[0078] In the meta-training phase, a model-agnostic meta-learning algorithm is employed to train meta-policies across task distributions encompassing various supply chain environments. Within each task's internal loop, a hierarchical attention reinforcement learning algorithm is used for rapid adaptation, incorporating a co-design of high-level and low-level policies.

[0079] The high-level strategy operates at a coarse-grained time scale, responsible for setting strategic goals and outputting planning vectors such as target inventory levels for various materials in each warehouse over a future period. The low-level strategy operates at a fine-grained time scale, generating specific procurement and allocation instructions based on the current system state and the goals of the high-level strategy. By introducing an attention mechanism, the low-level strategy can dynamically identify and prioritize the warehouses and materials to be processed, achieving adaptive allocation of computing resources.

[0080] During the meta-testing phase, when the system encounters entirely new supply chain tasks, it performs internal loop gradient updates on the meta-strategy using a small amount of interaction data to achieve rapid adaptation under small sample conditions. The final output flexible scheduling solution consists of two components: a baseline plan and a trigger-based response strategy. The baseline plan provides the order placement sequence and logistics path under normal conditions, while the trigger-based response strategy pre-sets alternative plans that the system will automatically activate when specific key events are predicted and perceived.

[0081] This implementation enhances the system's adaptability to complex supply chain environments through the synergy of a hierarchical decision-making structure and an attention mechanism, and strengthens the interpretability of the decision-making process through a clear division of labor mechanism, enabling the rapid generation of scheduling solutions that are both robust and flexible in dynamic environments.

[0082] Figure 2 This is a schematic diagram illustrating the implementation process of a dynamic scheduling method for the distribution network supply chain that considers the time delay of internal and external conduction periods, provided by another embodiment of the present invention. The details are as follows: S1. Construction of Guiding Period Feature Database and Predictive Modeling Analysis of influencing factors and data collection on the induction period: The factors of the induction period of materials in the procurement, production and transportation stages are decomposed, and a database of induction period characteristics is constructed through multi-source data fusion technology.

[0083] Procurement process lead time feature collection: Integrate supplier historical cooperation data to establish supplier performance profiles, including historical order scheduling cycles, raw material supply stability data, and capacity utilization trend data.

[0084] Production process characteristic collection: Collect historical operation records of the production line, including standard working hours of each process, historical operating status sequence of equipment, and statistical information on quality inspection pass rate.

[0085] Transportation process characteristic collection: Collect historical data of logistics transportation, including the travel time distribution of typical transportation routes in different seasons, records of the impact of weather conditions on transportation efficiency, and historical data of the trajectory of goods in transit.

[0086] Based on the collected multi-dimensional historical data, a time series analysis method is used to construct a prediction model for the lead time. Statistical learning techniques are used to identify the influence patterns and fluctuation characteristics of the lead time at each stage, providing data support for subsequent dynamic scheduling.

[0087] The Monte Carlo model is used to randomly generate possible future scenarios and their probabilities based on the predicted data. The scenario reduction technique first uses the K-medoids algorithm to cluster the 500 scenarios output by Monte Carlo, selects typical scenarios with a cumulative probability of 85%, and assigns a probability weight to each scenario after aggregation. Then, a compensation term is generated separately for the remaining 15% of low-probability scenarios as part of the optimization objective function to enhance the robustness of the optimization scheme.

[0088] S2. Construct a priority supply ranking mechanism for power materials in various scenarios. The core of establishing a priority ranking mechanism for power material supply lies in constructing a scientific indicator evaluation system and a dynamic adjustment algorithm. This mechanism is based on four dimensions: project urgency, planned start time, progress completion rate, and demand submission time. Through quantitative scoring, weight allocation, and comprehensive calculation, it achieves an objective ranking of all material demands. Simultaneously, a dynamic adjustment mechanism is established for non-routine demands, possessing real-time response capabilities. It can dynamically adjust the priority order based on random events, schedule changes, and other factors, ensuring that critical projects and urgent needs are always given priority.

[0089] S3. Construct a dynamic scheduling optimization model Construct a multi-timescale optimization model with the objective of minimizing total expected cost. The core feature of this model is that it uses the dynamic lead time data perceived in S1 as the key input parameter, rather than fixed values. It explicitly models the temporal relationships between internal and external processes; for example, how delays in the production process affect the start time of subsequent transportation processes and ultimately the final arrival time of goods. Establish a multi-timescale dynamic scheduling optimization model: the objective function is to minimize total expected cost, including conventional inventory holding costs, order delay penalty costs, transportation premium costs, and potential coordination costs for other material supply imbalances caused by this adjustment. The constraint system includes: material conservation constraints, ensuring that the inflow and outflow of materials at each node remain balanced under any scenario to meet dynamically changing service demands; capacity constraints, such as the upper limit of transport capacity for each transportation channel, the upper limit of backup supplier capacity, and warehouse turnover capacity limits; temporal relationship constraints, explicitly characterizing the sequential relationship of operations in each process to ensure logical consistency, such as production completion time must be earlier than transportation start time; and mathematical constraints such as the non-negativity and integer nature of decision variables. This model introduces random scenarios to simulate various possible future period fluctuations. At each rescheduling time, based on the latest system state and prediction information, it continuously solves to obtain a dynamic scheduling scheme that balances current optimality and future adaptability.

[0090] A multi-timescale dynamic scheduling optimization model is established, employing a two-tier optimization architecture of day-ahead coarse scheduling and intraday fine-tuning. In the day-ahead phase, based on historical data and probabilistic forecasts, various potential period fluctuation scenarios over the next few days are simulated using random scenario generation technology to generate a baseline scheduling scheme that balances economic efficiency and robustness, providing a global optimization perspective for subsequent decisions. In the intraday phase, the latest short-term forecast information is used, and a rolling optimization mechanism is employed to dynamically adjust and finely calibrate the scheduling plan for a short period (e.g., the next few hours), thereby effectively addressing the matching of power supply demand and achieving an organic combination of global optimization and local real-time response.

[0091] S4. A dynamic scheduling optimization model based on S2 is constructed, employing a meta-reinforcement learning framework to generate and execute flexible scheduling schemes. This scheme, through the combination of a hierarchical decision structure and an attention mechanism, achieves policy learning and rapid task adaptation at different time scales, and can output flexible scheduling schemes that include baseline plans and scenario response strategies.

[0092] In the meta-training phase, the meta-policy is trained on a task distribution encompassing various supply chain environments, including different supplier characteristics, transportation route configurations, and demand patterns. The inner loop of each task uses a novel hierarchical attention reinforcement learning algorithm for rapid adaptation, which incorporates the following core design: The high-level strategy operates at a coarse-grained time scale (e.g., weekly), responsible for setting strategic goals and outputting planning vectors such as target inventory levels for various materials in each warehouse over a future period. The low-level strategy, on the other hand, operates at a fine-grained time scale (e.g., daily), generating specific operational instructions for procurement and allocation based on the current system state and the goals given by the high-level strategy. In this structure, an attention mechanism is introduced, enabling the low-level strategy to dynamically identify and prioritize the warehouses and materials to be processed, thereby achieving adaptive allocation of computing resources.

[0093] During the meta-testing phase, when the system faces entirely new supply chain tasks, it can use a small amount of interactive data to perform internal loop gradient updates on the meta-strategy, quickly adapting to new business scenarios and achieving efficient self-adaptation under small sample conditions.

[0094] The final output of the flexible scheduling plan consists of two components: a baseline plan, which is the recommended order placement sequence and logistics route under normal conditions; and a trigger-based response strategy, which is a pre-set alternative plan that the system will automatically activate when a specific critical event (such as a supplier delay exceeding a threshold or a transportation route interruption) is predicted and perceived by the S1 link (such as switching to an alternative supplier or activating an emergency transportation method).

[0095] This design, through the synergy of a hierarchical decision-making structure and an attention mechanism, not only enhances the system's adaptability to complex supply chain environments but also strengthens the interpretability of the decision-making process through a clear division of labor mechanism, ultimately enabling the rapid generation of scheduling solutions that are both robust and flexible in dynamic environments.

[0096] Compared to existing technologies, the innovation of this invention lies in transforming supply chain scheduling from a static open-loop process based on fixed parameters into a data-driven dynamic closed-loop intelligent system. This transformation is achieved through the following key points: (1) Innovation in data input—from static experience values ​​to dynamic matching This invention independently developed a method for constructing a leading time characteristic database based on multi-source data fusion. Through in-depth mining and time-series analysis of historical data from various stages of procurement, production, and transportation, a precise leading time prediction model was established. Compared with traditional methods that rely solely on static historical data, this invention innovatively introduces Monte Carlo scenario generation technology, which can simulate various possible future leading time fluctuation scenarios and their probabilities of occurrence, providing more comprehensive and realistic data support for subsequent optimization decisions. This innovation enables the system to anticipate supply chain risks in advance, realizing a shift from passive response to proactive early warning.

[0097] The resulting improvements include a qualitative leap in input accuracy. The model no longer provides a vague average delivery cycle based on historical data, such as "it used to take an average of 10 days to complete similar orders," but rather, "This batch of orders is expected to take 12 days due to supplier A's saturated capacity; if we switch to supplier B, it will take 8 days, but the cost will increase by 15%." This enables precise decision-making. Furthermore, it allows for forward-looking planning, buying valuable time for proactive adjustments to the schedule.

[0098] (2) Innovation of multi-timescale collaborative optimization This invention proposes a two-layer optimization architecture of day-ahead coarse scheduling and intraday fine-tuning, and establishes a dynamic scheduling optimization model considering multi-dimensional constraints. Compared with traditional optimization methods with a single clock granularity, the innovation of this invention lies in explicitly modeling the time correlation between internal and external links. Through the synergistic effect of material conservation constraints, capacity constraints, and time correlation constraints, it achieves an organic combination of global optimization and local real-time response. This design enables the system to maintain long-term economic efficiency while quickly responding to random demand matching situations, solving the problem of traditional methods struggling to balance timeliness and economy.

[0099] The resulting improvements include: enhanced planning coordination, with the model automatically calculating the global impact of delays in a single link on the entire supply chain plan, thus making systemic adjustments rather than localized remedies; and more holistic and economical decision-making. For example, the model might suggest that "even if there is a slight delay in production, the total cost of expedited transportation, even with higher fees, is still lower than the loss from project delays," something that traditional models that view each link in isolation cannot do.

[0100] (3) Innovation in intelligent decision-making and rapid adaptation This invention develops an intelligent decision-making framework based on meta-reinforcement learning, particularly by innovatively designing a hierarchical attention-based reinforcement learning algorithm. Compared to traditional optimization algorithms or ordinary reinforcement learning methods, this method's innovations are reflected in three aspects: First, through the division of labor and cooperation between high-level and low-level policies, effective separation of strategic planning and tactical execution is achieved; second, the introduction of an attention mechanism enables the system to dynamically focus on key decision-making elements, significantly improving decision-making efficiency; and finally, the adoption of a meta-learning framework gives the system the ability to quickly adapt to small-sample environments, enabling it to rapidly adapt to entirely new supply chain environments. This series of innovations allows the system to maintain optimal decision-making while possessing human-like flexibility and adaptability.

[0101] The resulting improvements: Meta-reinforcement learning possesses powerful few-sample adaptive capabilities. When new elements appear in the supply chain network (such as introducing a new supplier or adding a warehouse node), the system does not need to be trained from scratch; it can quickly adapt with only a small amount of data, significantly reducing model maintenance costs and the deployment cycle for new scenarios. Simultaneously, the system has online learning capabilities, continuously learning and evolving from constantly generated new operational data. This allows scheduling strategies to continuously optimize along with business development, becoming an intelligent system that gets smarter with use.

[0102] (4) Innovation in the generation of flexible scheduling schemes This invention designs a flexible scheduling output mechanism that combines a baseline plan with trigger-based response strategies. Unlike traditional rigid scheduling schemes, this scheme has dual safeguards: the baseline plan ensures the economy of normal operation, while the pre-set trigger-based strategy library provides a rapid response channel for changing situations. This design not only improves the robustness of the system but also enhances the interpretability of the decision-making process through a clear division of labor, enabling managers to understand and trust the system's decision-making logic.

[0103] The resulting improvements: When changing demands actually occur, the flexible architecture of the baseline plan plus trigger-based strategies ensures business continuity. Even if a supplier suddenly stops supplying or a main transportation route is disrupted, the system can automatically and seamlessly switch to the backup plan, ensuring uninterrupted supply of materials and minimizing the negative impact of random changes.

[0104] Compared with existing technologies, this invention achieves the ultimate goal of efficient supply with lower investment, reducing the risk of shortages and stockpiles, reducing costs, and improving the reliability of power grid supply.

[0105] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0106] Figure 3This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 3 As shown, the electronic device 3 in this embodiment includes a processor 30 and a memory 31. The memory 31 stores a computer program 32. When the processor 30 executes the computer program 32, it implements the steps in the various method embodiments described above. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each module in the various device embodiments described above.

[0107] For example, computer program 32 may be divided into one or more modules / units, which are stored in memory 31 and executed by processor 30 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 32 in electronic device 3.

[0108] Electronic device 3 may include, but is not limited to, processor 30 and memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 3 may also include input / output devices, network access devices, buses, etc.

[0109] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0110] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0111] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A dynamic scheduling method for the distribution network supply chain considering the time delay of internal and external conduction periods, characterized in that, include: Based on the pre-set lead time feature library, obtain lead time feature data for various power material supply links in the supply chain; Based on the aforementioned guidance period characteristic data, the guidance period probability distribution of various power supply links is predicted, and based on the guidance period probability distribution, multiple sets of future guidance period scenarios for each type of power supply are generated using the Monte Carlo method. Based on multiple future guidance scenarios for each type of power material, a day-ahead scheduling model is established and solved to obtain the day-ahead scheduling scheme for each scheduling period of the next day. Based on the aforementioned daily scheduling scheme, an intraday scheduling model that considers the priority of demand scenarios is established and solved for each scheduling period of the following day to obtain a time-period-level dynamic scheduling scheme.

2. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 1, characterized in that, The acquisition of lead time characteristic data for various power supply links in the supply chain includes: Obtain the lead time characteristics of various power materials in the supply chain during the procurement, production, and transportation stages; The characteristics of the procurement process include: historical order scheduling cycle, raw material supply stability data, and capacity utilization trend data. The production process characteristics include: standard working hours for each process, historical operating status sequence of equipment, and statistical information on quality inspection pass rate; The characteristics of the transportation process include: the distribution of travel time for typical transportation routes in different seasons, records of the impact of weather conditions on transportation efficiency, and historical data on the trajectory of goods in transit.

3. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 1, characterized in that, Based on the aforementioned lead time characteristic data, predict the lead time probability distribution for various power supply links, including: Based on the aforementioned lead time characteristic data, the influence patterns and fluctuation characteristics of lead time in various power supply links are identified through time series analysis methods. Based on the influence patterns and fluctuation characteristics of the aforementioned lead time, a pre-trained short-term prediction model is used to predict the probability distribution of lead time in various power supply links.

4. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to any one of claims 1 to 3, characterized in that, The process of generating multiple future guidance period scenarios for each type of power resource using the Monte Carlo method based on the guidance period probability distribution includes: Based on the probability distribution of the future period, random sampling is performed using the Monte Carlo method to obtain multiple initial future period scenarios; The initial future guidance period scenarios are clustered using the K-medoids algorithm. High-probability future guidance period scenarios with accumulated probabilities reaching a preset threshold are selected, and clustered probabilities are assigned to each high-probability future guidance period scenario to obtain multiple sets of future guidance period scenarios for each type of power material.

5. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 4, characterized in that, The step involves establishing and solving a day-ahead scheduling model based on multiple future guidance scenarios for each type of power material, to obtain the day-ahead scheduling scheme for each scheduling period on the following day, including: Based on the initial future guidance scenario, low-probability future guidance scenarios other than the high-probability future guidance scenario are generated, and compensation terms are generated. Based on multiple future guidance scenarios for each type of power material, a day-ahead scheduling objective function is established with the goal of minimizing the sum of reserve scheduling cost, response time cost, and the compensation item. Establish the constraints of the day-ahead scheduling objective function, solve the day-ahead scheduling objective function based on the constraints, and obtain the day-ahead scheduling scheme for each scheduling period of the next day.

6. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 5, characterized in that, Based on the previous day's scheduling scheme, a daily scheduling model considering the priority of demand scenarios is established and solved for each scheduling period of the following day to obtain a time-period-level dynamic scheduling scheme, including: In each scheduling period of the following day, an intraday scheduling objective function is established with the goal of minimizing the sum of reserve scheduling cost, response time cost, delay penalty cost, scheduling penalty cost, maximum delay penalty cost, and the compensation term. Based on the aforementioned day-ahead scheduling plan, establish resource reservation constraints and adjustment range deviation constraints; Establish time constraints for requirements based on the priority of the aforementioned requirement scenarios; Based on the resource reservation constraint, the adjustment range deviation constraint, and the demand time limit constraint, the intraday scheduling objective function is solved to obtain a time-period-level dynamic scheduling scheme.

7. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 5, characterized in that, The constraints of the day-ahead scheduling objective function include: Material type matching constraint: For any demand scenario, the type of power material dispatched must be consistent with the type of power material required for that demand scenario; Inventory capacity constraint: At any time, the inventory of each type of power material in any warehouse shall not exceed the maximum storage capacity of the corresponding warehouse. Safety stock constraint: At any time in any warehouse, the inventory of each type of electrical material must be greater than the corresponding minimum safety stock. Inventory update constraint: The inventory of each type of power material must meet the time-series balance requirement.

8. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 6, characterized in that, The step of establishing time constraints for requirements based on the priority of the required scenarios includes: Obtain evaluation index data for various demand scenarios; wherein, the evaluation index includes the urgency of the project, the planned start time, the progress completion rate, and the demand submission time; The weights of each evaluation indicator are determined using the analytic hierarchy process (AHP). Based on the evaluation index data and the weight of each evaluation index for each demand scenario, calculate the priority score for each demand scenario. Based on the priority scoring of each demand scenario, demand time constraints are established.

9. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 5, characterized in that, The day-ahead scheduling objective function includes: in, The number of future guidance scenarios; For future guidance scenarios s The probability of; These are the weighting coefficients; Cost of power material reserves; Costs related to power material dispatching; To address the time and cost of response; As compensation, For the compensation term coefficient, Compensation value.

10. The dynamic scheduling method for distribution network supply chain considering internal and external conduction time delays according to claim 6, characterized in that, The intraday scheduling objective function includes: in, For a specific moment; Cost of power material reserves; Penalty cost for delay; Costs related to power material dispatching; To adjust the cost of penalties; To address the time and cost of response; The maximum delay penalty cost; As compensation, For the compensation term coefficient, Compensation value.

Images