Methods for regional virtual inventory pools and cross-site transfer processes in power plant inventory sharing mechanisms

By setting up sensing devices and edge computing gateways in power plants, a regional virtual inventory pool is formed. Combined with dynamic safety stock thresholds and digital twin models, the deep coupling problem of cross-site inventory transfer in power plants is solved, realizing dynamic management of power plant inventory and real-time optimization of transfer schemes, thereby improving the feasibility and security of power plant inventory transfer.

CN123117980APending Publication Date: 2026-11-17THREE GORGES NEW ENERGY SIZIWANG BANNER CO LTD +2
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611190078.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-06
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies have failed to effectively achieve deep coupling of power plant inventory transfer across different sites, have not utilized digital twins for prediction and evaluation of multiple candidate transfer schemes, and lack a dynamic adjustment mechanism during execution, resulting in transfer schemes becoming infeasible or lacking in security.

Method used

By setting up sensing devices and edge computing gateways in power plants, a regional virtual inventory pool is formed. Combining dynamic safety stock threshold calculation and machine learning models, a digital twin model is used to pre-execute candidate solutions and trigger rolling re-optimization when execution deviations occur, ensuring the feasibility and security of allocation solutions.

Benefits of technology

It achieves continuous feasibility and security in the process of inter-site material transfer in power plants, reduces the probability of secondary shortages at material supply sites, improves the dynamic adaptability and execution efficiency of the transfer scheme, and reduces the computational scale of re-optimization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117980A_ABST
    Figure CN123117980A_ABST
Patent Text Reader

Abstract

This invention discloses a method for establishing a regional virtual inventory pool and cross-site transfer process within a power plant inventory sharing mechanism. It collects inventory, material health, equipment, environmental, vehicle, and transit status data from multiple power plants via an edge computing gateway to construct a regional virtual inventory pool. Candidate transfer schemes are generated based on a dynamic safety stock threshold, and machine learning and digital twin models are used to predict arrival time, material health status, and transfer risks. Multi-objective optimization is then used to determine the transfer scheme. During execution, the deviation between the actual and predicted states is calculated in real time. When the deviation exceeds a threshold, unfinished tasks are re-optimized, and the inventory status and transfer records are updated. This invention improves the efficiency of regional inventory sharing, transfer security, and dynamic response capabilities under abnormal conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power plant inventory management technology, specifically to a method for a regional virtual inventory pool and cross-site transfer process for a power plant inventory sharing mechanism. Background Technology

[0002] During operation and maintenance, power plants typically need to stockpile battery modules, critical spare parts, tools, and emergency supplies. For regional power facilities consisting of multiple stations such as battery swapping stations, energy storage stations, wind farms, photovoltaic power plants, or hydropower stations, each station differs in geographical location, equipment type, material consumption patterns, failure risks, and external replenishment cycles. If each station sets up and manages its inventory independently, some stations may experience long-term idle inventory, while other stations may face material shortages due to equipment failures, load changes, or temporary maintenance. Therefore, how to achieve regional inventory resource sharing and cross-station allocation while ensuring the operation and maintenance needs of each station is a problem that power plant inventory collaborative management needs to solve. Although existing systems can collect and display inventory data from multiple stations or initiate allocations when inventory falls below the safety stock level, their allocation decisions are usually still mainly based on the current inventory quantity, transportation distance, or allocation cost.

[0003] Chinese patent application CN119515270A discloses a method for managing the storage of power materials based on a distributed warehouse. This method obtains the actual inventory levels of each sub-warehouse within the distributed warehouse system and determines whether the actual inventory level is lower than the safety stock level. When a target sub-warehouse has insufficient inventory, a material allocation request is generated and sent to the coordination center. The coordination center then allocates power materials to the target sub-warehouse according to emergency allocation rules or preventative allocation rules, based on the threshold range of the inventory level, thereby reducing inventory costs caused by improper allocation.

[0004] Chinese patent application CN114997802B discloses a logistics distribution and scheduling system based on digital twins. This system, designed for the logistics distribution of communication equipment, includes an order processing module, a logistics resource library module, a digital model library module, a digital simulation module, a scheduling module, and a monitoring module. It establishes digital models for the equipment to be delivered, transport vehicles, loading equipment, and transport routes, and performs loading and transportation simulations according to the delivery plan. During the delivery process, it monitors vehicles and en route communication equipment in real time, feeding the monitoring results back to the digital simulation module to optimize subsequent delivery plans and improve transportation reliability.

[0005] In summary, existing technologies have proposed corresponding solutions for distributed power material inventory allocation and digital twin flow simulation, but a deep coupling mechanism for cross-site inventory allocation in power plants is still lacking between the two types of technologies. On the one hand, existing inventory allocation schemes typically do not utilize digital twins to predict arrival times, health status of arriving materials, and allocation risks for multiple candidate allocation schemes item by item. On the other hand, existing digital twin flow schemes typically do not directly use simulation outputs for dynamic safety stock constraints and cross-site allocation optimization, nor do they form a closed loop for resolving remaining tasks based on real-time execution deviations. Therefore, it is necessary to provide a method and system suitable for power plant regional inventory sharing and cross-site allocation. This method should comprehensively consider future station demand, equipment failure risks, material health status, in-transit inventory, vehicle status, and route risks to predict and evaluate multiple candidate allocation schemes. Furthermore, when the allocation execution status changes, it should promptly adjust unfinished allocation tasks based on the current real-time situation to improve the continuous feasibility and security of cross-site allocation schemes during dynamic execution. Summary of the Invention

[0006] The purpose of this invention is to address the dynamic changes in inventory levels across multiple sites, battery health status, future site demand, and transportation risks. This involves preventing the initial optimal allocation plan from becoming infeasible during actual execution and rapidly generating new allocation plans that meet safety stock and material health constraints when anomalies occur. To solve these problems, this invention provides a method for establishing a regional-level virtual inventory pool and a cross-site allocation process within a power plant inventory sharing mechanism, comprising the following steps: S1. Sensing devices and edge computing gateways are installed in the warehouses, equipment areas, and transport vehicles of each power station. The edge computing gateways collect data on material identification and quantity, material health status, equipment operating status, storage environment, vehicle location, vehicle operating status, and the status of materials in transit. The collected data undergoes time alignment, format conversion, outlier filtering, and missing value processing to form a regional inventory status dataset indexed by "station-material-timestamp-status version." Materials remain physically stored at each station, while the central platform performs logical centralized management of the regional inventory quantity, status, availability, and in-transit status, forming a regional virtual inventory pool.

[0007] For battery modules, material health status includes state of charge (SOC), state of health (SOH), voltage, current, temperature, and consistency parameters; for mechanical or electrical spare parts, material health status includes storage duration, vibration, humidity exposure, inspection status, and remaining usable life.

[0008] S2. Calculate the dynamic safety stock threshold; For material m at station i, calculate the dynamic safety stock threshold S at time t. i , m(t). In one implementation, the dynamic safety stock threshold is expressed as: ; in, Forecast demand within the forecast time window or replenishment lead time L; R i , m (t) represents the additional demand calculated based on the failure probability of station equipment, equipment importance level, and maintenance material list; σ i , m (t) represents the demand forecasting error; Z represents the service level coefficient; Q i , m in (t) represents the confirmed inventory level that is en route and can arrive within the stated time window. Further, neighboring site support time, external supply reliability, and site emergency response level can be converted into additional safety margins.

[0009] The station j plans to transfer a quantity q to the demanding station. j , m At that time, station j will be retained as a candidate feeding station only if the following formula is satisfied: ; Among them, I j , m (t) represents the available inventory at station j at time t. This constraint helps prevent secondary stockouts at the supplying station caused by cross-station transfers.

[0010] S3. When the target material inventory at the demand site is lower than the dynamic safety stock threshold, the system generates a transfer request. The system filters supply sites based on material model compatibility, health status, safety margin of candidate supply sites, vehicle load capacity, and hazardous materials transportation conditions, and combines supply sites, transfer quantities, transport vehicles, and transport routes to generate multiple candidate transfer plans.

[0011] A state vector is established for the candidate scheme. The state vector includes the safety margin after the material supply station is dispatched, the expected shortage of the demand station, the current health status of the materials, the vehicle load margin, the vehicle fault status, the route distance, real-time traffic, weather conditions, the expected arrival time, and the expected health status of the arriving materials.

[0012] S4. This invention does not directly use machine learning models to abstractly adjust multi-objective weights, but instead utilizes historical allocation data and real-time status data to predict technical variables that characterize the physical process of allocation. The inputs to the machine learning model may include allocation distance, route level, historical congestion, weather, vehicle type, driving conditions, initial material temperature, SOC, SOH, loading method, and terminal loading / unloading efficiency; the outputs may include actual transport time, transport delay probability, temperature upon arrival, SOC or SOH upon arrival, probability of abnormal status, and probability of allocation failure.

[0013] The machine learning output and the state vectors of candidate allocation schemes are input into the digital twin model. The digital twin model pre-executes each candidate scheme individually, rather than only displaying the final scheme. The digital twin model updates vehicle location, transportation time, material temperature, vibration, health status, and inventory changes at supply and demand depots according to a preset simulation step size, and outputs the predicted arrival time T for each candidate scheme. k Predicting the health status of arriving supplies H k and the risk value P of the transfer k .

[0014] In one implementation, the risk value P is allocated. k It can be represented as: ; Where, p k d Let p be the probability of transportation delay. k r p represents the probability of route interruption. k v p represents the probability of vehicle malfunction. k h p represents the probability of the health status of supplies exceeding the limit. k s The probability of stockout after material is transferred from the supply depot is denoted as α1; α1 to α5 are normalization coefficients. The risk indicators can also be fused using Bayesian networks, fuzzy comprehensive evaluation, or probabilistic graphical models.

[0015] S5. Use the predicted arrival time, predicted health status of arriving materials, and allocation risk value output by the digital twin as target parameters or feasibility constraints for the allocation optimization model. The decision variables of the allocation optimization model include the selection variables of the supply yard, the allocation quantity, the vehicle selection variables, and the route selection variables.

[0016] In one implementation, the allocation optimization objective is: ; Where C represents allocation and transportation costs, T represents predicted arrival time, P represents allocation risk, L represents the expected stockout loss at the demand site before the arrival of materials, and w1 to w4 are normalized weights. The optimization model must at least satisfy the following constraints: material supply and demand balance, dynamic safety stock at supply sites, model compatibility, vehicle load capacity, hazardous materials transportation, arrival time, predicted arrival health status, and regional service level.

[0017] Through the above processing, digital twin simulation is no longer a display tool after the allocation is completed, but a technical parameter used to generate and correct the optimization model; when the expected health status of a candidate solution is lower than the preset threshold, or the risk value exceeds the allowable range, the candidate solution is directly removed from the feasible domain of the optimization model.

[0018] S6. After the initial allocation plan is executed, the edge computing gateway continuously uploads real-time data on vehicle location, vehicle condition, and the temperature, vibration, SOC, and SOH of materials en route, and updates the corresponding digital twin. The system compares the real-time status with the predicted status of the digital twin to identify execution deviations.

[0019] In one implementation, the overall execution deviation E k Represented as: ; Wherein, ΔETA is the deviation between the actual estimated arrival time and the predicted arrival time, ΔTemp is the deviation between the actual temperature of the materials and the predicted temperature, ΔH is the deviation between the actual health status of the materials and the predicted health status, ΔV is the deviation between the actual status of the vehicles and the predicted status, and ΔI is the deviation between the actual inventory at the depot and the predicted inventory; θ1 to θ5 are the corresponding deviation thresholds, and β1 to β5 are the weights. The system can trigger re-optimization based on the overall execution deviation, or it can trigger it when any single key deviation exceeds the mandatory threshold.

[0020] After triggering rolling re-optimization, the system does not simply switch to the pre-stored backup routes, but instead uses the current real-time state as the new initial state. The system sets completed material outbound, transportation segments, and delivery steps as frozen variables, and sets currently unreleaseable vehicles and loaded hazardous materials that cannot be transferred en route as immutable constraints; it only regenerates candidate solutions for unfulfilled demand, adjustable vehicles, and unexecuted routes, and then performs digital twin simulation and risk constraint optimization again.

[0021] S7. To balance real-time performance and traceability, this invention does not require all high-frequency raw sensor data to be written to the blockchain. Instead, it writes the summary value of the regional inventory status snapshot, the allocation scheme identifier, the digital twin model version, the execution deviation triggering condition, the correlation between the update scheme and the initial scheme, and the inbound and outbound handover results to the consortium blockchain. The high-frequency raw data is stored in an off-chain time-series database, and consistency verification is achieved through the status snapshot summary.

[0022] When communication is interrupted, the edge computing gateway saves the most recent valid state and data to be uploaded, and performs anomaly judgment on the material status according to local security rules; for batteries or hazardous materials exceeding the mandatory safety threshold, it generates instructions to suspend transportation or handle them nearby. After the network is restored, data is retransmitted in timestamp order, and the retransmitted data is used to correct the digital twin state.

[0023] Based on the above method, a regional virtual inventory pool and cross-site allocation system for power plant inventory sharing mechanism are provided, including: (1) Perception and Edge Layer: This includes material identification and status acquisition equipment, equipment status interface, warehouse environment sensors, vehicle positioning terminal, vehicle-mounted material monitoring terminal, and edge computing gateway. The edge computing gateway completes data preprocessing, timestamp correction, local caching, and mandatory security threshold judgment.

[0024] (2) Regional Data Layer: This includes a unified data model, a regional virtual inventory pool, a time-series database, status version management, and a trusted record interface. The unified data model should at least include station identifier, material identifier, inventory status, health status, transferable status, in-transit status, and data quality identifier.

[0025] (3) Model and simulation layer: including demand forecasting model, equipment failure risk model, dynamic safety stock calculation model, machine learning variable prediction model and digital twin model. The digital twin model simulates each candidate solution and performs state correction based on real-time feedback.

[0026] (4) Decision and control layer: including candidate solution generation module, risk-constrained multi-objective optimization module, execution deviation calculation module, and rolling re-optimization module. The rolling re-optimization module can freeze executed or irreversible variables and re-solve the remaining tasks.

[0027] (5) Application terminals: including regional management terminal, station terminal, transportation terminal and industrial large screen. The regional management terminal is used to view regional inventory, approve special transfers and manage thresholds; the station terminal is used for inbound and outbound confirmation and local status handling; the transportation terminal is used for task reception, navigation, anomaly reporting and handover confirmation.

[0028] Compared with the prior art, the beneficial effects of the present invention include: (1) By taking into account future demand, equipment failure risk, replenishment cycle and in-transit inventory through dynamic safety stock threshold, the probability of secondary shortage after material supply stations are transferred out is reduced; (2) Perform digital twin pre-execution for each candidate allocation plan, and directly input the expected arrival time, health status of the arriving materials and risk value into the allocation optimization model to realize the technical coupling of simulation and decision-making; (3) Use machine learning to predict technical variables such as transportation time, changes in material status and probability of allocation failure, and avoid generating plans based solely on fixed distances or fixed weights; (4) By detecting the deviation between the real-time status and the predicted status, rolling re-optimization is triggered before the initial plan becomes infeasible, thereby improving the speed of allocation recovery in abnormal situations; (5) When re-optimizing, freeze completed or irreversible steps and only solve the remaining tasks. This avoids conflicts between the solution and the actual state and reduces the scale of recalculation. (6) By combining on-chain digests and off-chain high-frequency data, the key decision-making process can be traced, while avoiding the latency and storage pressure caused by putting all the original data on the chain. Attached Figure Description

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0030] Figure 1 This is a flowchart of the method for the regional virtual inventory pool and cross-site transfer process of the power plant inventory sharing mechanism of the present invention; Figure 2 This is a schematic diagram of the system functional architecture and data closed loop of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] Example 1 S1. Sensing devices and edge computing gateways are installed in the warehouses, equipment areas, and transport vehicles of each power station. The edge computing gateways collect data on material identification and quantity, material health status, equipment operating status, storage environment, vehicle location, vehicle operating status, and the status of materials in transit. The collected data undergoes time alignment, format conversion, outlier filtering, and missing value processing to form a regional inventory status dataset indexed by "station-material-timestamp-status version." Materials remain physically stored at each station, while the central platform performs logical centralized management of the regional inventory quantity, status, availability, and in-transit status, forming a regional virtual inventory pool.

[0033] For battery modules, material health status includes state of charge (SOC), state of health (SOH), voltage, current, temperature, and consistency parameters; for mechanical or electrical spare parts, material health status includes storage duration, vibration, humidity exposure, inspection status, and remaining usable life.

[0034] S2. Calculate the dynamic safety stock threshold; For material m at station i, calculate the dynamic safety stock threshold S at time t. i , m (t). In one implementation, the dynamic safety stock threshold is expressed as: ; in, Forecast demand within the forecast time window or replenishment lead time L; R i , m (t) represents the additional demand calculated based on the failure probability of station equipment, equipment importance level, and maintenance material list; σ i , m (t) represents the demand forecasting error; Z represents the service level coefficient; Q i , m in (t) represents the confirmed inventory level that is en route and can arrive within the stated time window. Further, neighboring site support time, external supply reliability, and site emergency response level can be converted into additional safety margins.

[0035] The station j plans to transfer a quantity q to the demanding station. j , m At that time, station j will be retained as a candidate feeding station only if the following formula is satisfied: ; Among them, I j , m (t) represents the available inventory at station j at time t. This constraint helps prevent secondary stockouts at the supplying station caused by cross-station transfers.

[0036] S3. When the target material inventory at the demand site is lower than the dynamic safety stock threshold, the system generates a transfer request. The system filters supply sites based on material model compatibility, health status, safety margin of candidate supply sites, vehicle load capacity, and hazardous materials transportation conditions, and combines supply sites, transfer quantities, transport vehicles, and transport routes to generate multiple candidate transfer plans.

[0037] A state vector is established for the candidate scheme. The state vector includes the safety margin after the material supply station is dispatched, the expected shortage of the demand station, the current health status of the materials, the vehicle load margin, the vehicle fault status, the route distance, real-time traffic, weather conditions, the expected arrival time, and the expected health status of the arriving materials.

[0038] S4. This invention does not directly use machine learning models to abstractly adjust multi-objective weights, but instead utilizes historical allocation data and real-time status data to predict technical variables that characterize the physical process of allocation. The inputs to the machine learning model may include allocation distance, route level, historical congestion, weather, vehicle type, driving conditions, initial material temperature, SOC, SOH, loading method, and terminal loading / unloading efficiency; the outputs may include actual transport time, transport delay probability, temperature upon arrival, SOC or SOH upon arrival, probability of abnormal status, and probability of allocation failure.

[0039] The machine learning output and the state vectors of candidate allocation schemes are input into the digital twin model. The digital twin model pre-executes each candidate scheme individually, rather than only displaying the final scheme. The digital twin model updates vehicle location, transportation time, material temperature, vibration, health status, and inventory changes at supply and demand depots according to a preset simulation step size, and outputs the predicted arrival time T for each candidate scheme. k Predicting the health status of arriving supplies H k and the risk value P of the transfer k .

[0040] In one implementation, the risk value P is allocated. k It can be represented as: ; Where, p k d Let p be the probability of transportation delay. k r p represents the probability of route interruption. k v p represents the probability of vehicle malfunction. k h p represents the probability of the health status of supplies exceeding the limit. k s The probability of stockout after material is transferred from the supply depot is denoted as α1; α1 to α5 are normalization coefficients. The risk indicators can also be fused using Bayesian networks, fuzzy comprehensive evaluation, or probabilistic graphical models.

[0041] S5. Use the predicted arrival time, predicted health status of arriving materials, and allocation risk value output by the digital twin as target parameters or feasibility constraints for the allocation optimization model. The decision variables of the allocation optimization model include the selection variables of the supply yard, the allocation quantity, the vehicle selection variables, and the route selection variables.

[0042] In one implementation, the allocation optimization objective is: ; Where C represents allocation and transportation costs, T represents predicted arrival time, P represents allocation risk, L represents the expected stockout loss at the demand site before the arrival of materials, and w1 to w4 are normalized weights. The optimization model must at least satisfy the following constraints: material supply and demand balance, dynamic safety stock at supply sites, model compatibility, vehicle load capacity, hazardous materials transportation, arrival time, predicted arrival health status, and regional service level.

[0043] Through the above processing, digital twin simulation is no longer a display tool after the allocation is completed, but a technical parameter used to generate and correct the optimization model; when the expected health status of a candidate solution is lower than the preset threshold, or the risk value exceeds the allowable range, the candidate solution is directly removed from the feasible domain of the optimization model.

[0044] S6. After the initial allocation plan is executed, the edge computing gateway continuously uploads real-time data on vehicle location, vehicle condition, and the temperature, vibration, SOC, and SOH of materials en route, and updates the corresponding digital twin. The system compares the real-time status with the predicted status of the digital twin to identify execution deviations.

[0045] In one implementation, the overall execution deviation E k Represented as: ; Wherein, ΔETA is the deviation between the actual estimated arrival time and the predicted arrival time, ΔTemp is the deviation between the actual temperature of the materials and the predicted temperature, ΔH is the deviation between the actual health status of the materials and the predicted health status, ΔV is the deviation between the actual status of the vehicles and the predicted status, and ΔI is the deviation between the actual inventory at the depot and the predicted inventory; θ1 to θ5 are the corresponding deviation thresholds, and β1 to β5 are the weights. The system can trigger re-optimization based on the overall execution deviation, or it can trigger it when any single key deviation exceeds the mandatory threshold.

[0046] After triggering rolling re-optimization, the system does not simply switch to the pre-stored backup routes, but instead uses the current real-time state as the new initial state. The system sets completed material outbound, transportation segments, and delivery steps as frozen variables, and sets currently unreleaseable vehicles and loaded hazardous materials that cannot be transferred en route as immutable constraints; it only regenerates candidate solutions for unfulfilled demand, adjustable vehicles, and unexecuted routes, and then performs digital twin simulation and risk constraint optimization again.

[0047] S7. To balance real-time performance and traceability, this invention does not require all high-frequency raw sensor data to be written to the blockchain. Instead, it writes the summary value of the regional inventory status snapshot, the allocation scheme identifier, the digital twin model version, the execution deviation triggering condition, the correlation between the update scheme and the initial scheme, and the inbound and outbound handover results to the consortium blockchain. The high-frequency raw data is stored in an off-chain time-series database, and consistency verification is achieved through the status snapshot summary.

[0048] When communication is interrupted, the edge computing gateway saves the most recent valid state and data to be uploaded, and performs anomaly judgment on the material status according to local security rules; for batteries or hazardous materials exceeding the mandatory safety threshold, it generates instructions to suspend transportation or handle them nearby. After the network is restored, data is retransmitted in timestamp order, and the retransmitted data is used to correct the digital twin state.

[0049] Example 2 like Figure 2 As shown in Embodiment 1, a system for a regional virtual inventory pool and cross-site transfer process for a power plant inventory sharing mechanism is provided, specifically including: (1) Perception and Edge Layer: This includes material identification and status acquisition equipment, equipment status interface, warehouse environment sensors, vehicle positioning terminal, vehicle-mounted material monitoring terminal, and edge computing gateway. The edge computing gateway completes data preprocessing, timestamp correction, local caching, and mandatory security threshold judgment.

[0050] (2) Regional Data Layer: This includes a unified data model, a regional virtual inventory pool, a time-series database, status version management, and a trusted record interface. The unified data model should at least include station identifier, material identifier, inventory status, health status, transferable status, in-transit status, and data quality identifier.

[0051] (3) Model and simulation layer: including demand forecasting model, equipment failure risk model, dynamic safety stock calculation model, machine learning variable prediction model and digital twin model. The digital twin model simulates each candidate solution and performs state correction based on real-time feedback.

[0052] (4) Decision and control layer: including candidate solution generation module, risk-constrained multi-objective optimization module, execution deviation calculation module, and rolling re-optimization module. The rolling re-optimization module can freeze executed or irreversible variables and re-solve the remaining tasks.

[0053] (5) Application terminals: including regional management terminal, station terminal, transportation terminal and industrial large screen. The regional management terminal is used to view regional inventory, approve special transfers and manage thresholds; the station terminal is used for inbound and outbound confirmation and local status handling; the transportation terminal is used for task reception, navigation, anomaly reporting and handover confirmation.

[0054] Example 3 This embodiment sets up five battery swapping stations (A, B, C, D, and E) within the area to manage M1 type battery modules. Each station's edge computing gateway collects inventory quantity, SOC, SOH, and temperature every 30 seconds; the vehicle end collects location, speed, vehicle fault codes, and onboard environment data every 10 seconds. The central platform updates demand forecasts every 15 minutes, recalculating dynamic safety stock whenever there are significant changes in inventory, equipment failure, or allocation status. See Table 1 for details.

[0055] Table 1. Inventory Table for Each Station

[0056] Station B's current available inventory is lower than the dynamic safety stock, creating an urgent need to allocate 10 sets of M1 type battery modules. Based on the constraint that the inventory after allocation cannot fall below the dynamic safety stock, Station A can allocate a maximum of 18 sets, Station C a maximum of 11 sets, Station D a maximum of 3 sets, and Station E a maximum of 17 sets. The system generates candidate solutions for Station A, Station C, and a "Station C + Station D joint supply" solution, considering battery health status and distance. Although Station E has higher inventory, its average SOH is low and it is located far away, so it is only considered a secondary candidate. See Table 2.

[0057] Table 2 Candidate Solution Table

[0058] The digital twin model pre-executed the above candidate schemes. Although scheme K2 has lower transportation costs, its average SOH is lower when the batteries arrive at station C; scheme K1 has a longer arrival time; scheme K3 meets the constraints of dynamic safety stock, arrival health status, and vehicle load, and has the lowest overall risk. Therefore, the optimization model selected scheme K3 as the initial allocation scheme.

[0059] Twenty minutes after the allocation was executed, the vehicles at Station C were affected by a road accident, resulting in an actual ETA (Electronic Time To Arrival) 18 minutes longer than predicted. Simultaneously, the onboard temperature was 4.5°C higher than the twin-battery prediction temperature, causing the overall execution deviation to exceed the threshold. The system froze the completed battery shipments and routes already traveled at Station C, and re-simulated using the current vehicle location and battery temperature as initial conditions. The re-optimization results maintained that the vehicles at Station C continued to travel to Station B. Simultaneously, the three battery sets originally handled by Station D were replaced by three sets dispatched from Station A, to be handled by backup vehicles closer to the current demand station. The updated plan ensured that the estimated arrival time was controlled before a substantial shortage occurred at Station B.

[0060] After the supplies arrive, Bilibili scans the code to confirm their entry into the warehouse. The system updates the inventory quantity and battery status of stations A, B, and C, and writes the status snapshot summary, initial plan K3, deviation trigger record, update plan, and handover results into the consortium blockchain.

[0061] Example 3 This embodiment is used for critical spare parts such as power modules and main bearing sensors in wind farm converters. The equipment condition monitoring model estimates future maintenance needs based on fault characteristics and remaining lifespan. The additional demand in the dynamic safety stock is calculated based on the correspondence between equipment failure probability and the bill of materials. The health status of spare parts includes storage time, inspection validity period, packaging humidity, vibration and shock records, and compatible models.

[0062] When the inventory of power modules at a wind farm falls below the dynamic safety stock, the system generates candidate solutions for multiple supply depots and transportation routes. A digital twin model synchronously simulates the maintenance support level of the supply depot after it is reassigned, the expected downtime losses at the demanding depot, and the vibration risk of spare parts during transportation. If a solution, although geographically short, results in the supply depot being unable to cover high-probability failure demands within the predicted time window after reassignment, that solution is eliminated due to dynamic safety stock constraints.

[0063] During transportation, if the vehicle vibration amplitude continuously exceeds the predicted range, the edge gateway first executes a local alarm and deceleration command; if the cumulative vibration dose reaches the mandatory threshold, the system marks the spare part as pending re-inspection and reassigns a qualified spare part from another station in the rolling re-optimization to prevent spare parts with uncertain status from being directly put into maintenance.

[0064] Example 4 The machine learning prediction model uses historical allocation tasks to construct training samples. The input features of each training sample include the distance between supply and demand stations, road grade, time period, weather, vehicle load, driving speed, initial temperature of the goods, initial health status of the goods, and loading and unloading time; the labels include actual transportation time, maximum temperature rise, arrival health status, and whether any abnormalities have occurred.

[0065] After model training is complete, predicted values ​​of technical variables are output each time a candidate solution is generated. The digital twin model first initializes the state evolution parameters of the candidate solutions using the predicted values; after the adjustment is executed, the actual data and the predicted data are used to form error samples. When the cumulative error samples reach a preset number or the model error continuously exceeds a threshold, the model is triggered to retrain or incrementally update.

[0066] To ensure the safety of optimization constraints, when the confidence level of the machine learning model is lower than the preset value, the system uses conservative boundary values ​​to replace the point prediction values. For example, it uses the upper limit of the transportation time prediction interval and the lower limit of the arrival health status prediction interval for optimization.

[0067] Example 5 In one verification method, the following control schemes are set up respectively: independent inventory at each station and no transfer is allowed; fixed safety stock plus manual transfer; fixed weight multi-objective optimization plus pre-stored backup routes; digital twin pre-execution is used but no rolling re-optimization is performed; and the dynamic safety stock, scheme-by-scheme twin simulation, risk constraint optimization and rolling re-optimization scheme of the present invention.

[0068] Evaluation indicators include total regional inventory, inventory capital occupancy rate, stockout rate, average response time, transportation costs, allocation failure rate, arrival material health qualification rate, number of secondary stockouts after being transferred from the supply depot, re-decision time after anomalies, and regional service level. In actual applications, specific data can be filled in based on pilot or simulation results to demonstrate the technical effects of incorporating digital twin outputs into optimization constraints and the rolling re-optimization of execution deviations.

[0069] The above embodiments are merely preferred embodiments of the present invention. Those skilled in the art can make equivalent substitutions for the prediction model, risk fusion method, optimization algorithm, sensor type, and communication method without departing from the core concept of the present invention. Any technical solution that uses digital twin simulation of each candidate allocation scheme, applies the simulation output to constrain or optimize allocation decisions, and performs rolling re-optimization of the remaining tasks after the execution deviation exceeds a threshold, should fall within the protection scope of the present invention.

Claims

1. A method for establishing a regional virtual inventory pool and cross-station transfer process for a power plant inventory sharing mechanism, characterized in that: include: S1. Construct a regional inventory status dataset. By setting up edge computing gateways at multiple power plant sites, collect material inventory data, material health status data, equipment operation status data, and storage environment data of each site. Also collect vehicle status data, location information, and in-transit material status data of dispatch vehicles. Standardize the collected data to form a regional virtual inventory pool. S2. Generate candidate allocation plans. When the target material inventory at the demand site is lower than the dynamic safety stock threshold, generate allocation requirements. Select candidate supply sites based on the available inventory, forecasted demand, material health status, and safety stock remaining after allocation. Combine candidate supply sites, allocation quantity, transport vehicles, and transport routes into multiple candidate allocation plans. S3. Perform digital twin simulation of candidate allocation schemes. Input each candidate allocation scheme and its corresponding inventory status data, material health status data, vehicle status data, route status data and environmental data into the digital twin model. Simulate the allocation process for each candidate allocation scheme to obtain the predicted arrival time, predicted arrival material health status and allocation risk value for each candidate allocation scheme. S4. Generate an initial allocation plan. Input the predicted arrival time, predicted health status of the arriving materials, and allocation risk value into the allocation optimization model. With allocation cost, predicted arrival time, allocation risk, and expected stockout loss as optimization objectives, and with the constraints that the inventory after the material supply station is transferred out is not lower than its dynamic safety stock threshold and the predicted health status of the arriving materials is not lower than the preset health threshold, the initial allocation plan is obtained by solving the problem. S5. Perform execution status deviation judgment. During the execution of the initial allocation plan, input the vehicle location, vehicle status and in-transit material status collected in real time by the edge computing gateway into the digital twin model, update the predicted status of the allocation process, and calculate the execution deviation between the real-time status and the predicted status. S6. Perform rolling re-optimization. When the execution deviation exceeds the corresponding deviation threshold, use the current inventory status, vehicle location, in-transit material status and completed transfer steps as new initial conditions to regenerate candidate transfer schemes for tasks that have not yet been completed, and repeat steps S3 and S4 to obtain an updated transfer scheme. S7. Update the regional virtual inventory pool. After the materials are shipped out, transported or stored, update the quantity and health status of the materials at the corresponding site, and record the initial allocation plan, execution deviation, updated allocation plan and allocation execution result.

2. The method for regional virtual inventory pools and cross-site transfer processes in the power plant inventory sharing mechanism according to claim 1, characterized in that, The dynamic safety stock threshold is calculated based on the material demand within the predicted time window, the additional demand corresponding to the risk of equipment failure at the site, the demand forecast error, the external replenishment cycle, the confirmed inventory in transit, and the estimated support time of the nearby site. Candidate supply depots are reserved only if their current inventory of the target material minus the quantity to be allocated is not lower than the dynamic safety stock threshold.

3. The method for regional virtual inventory pools and cross-site transfer processes in the power plant inventory sharing mechanism according to claim 1, characterized in that, For each candidate allocation plan, a plan state vector is established, which includes: the safety margin of inventory after the material supply station is transferred out, the expected shortage of the demand station, the current health status of the allocated materials, the load margin of the transport vehicles, the expected travel time of the route, the route risk parameters, the expected arrival time, and the expected health status of the arriving materials.

4. The method for regional virtual inventory pools and cross-site transfer processes in the power plant inventory sharing mechanism according to claim 1, characterized in that, A machine learning prediction model is trained based on historical allocation data and real-time status data. The machine learning prediction model outputs at least one of the following: actual transportation time prediction, transportation delay probability, material health status prediction upon arrival, material status abnormality probability, or allocation failure probability. The output is used to initialize or correct the digital twin model and serves as the target or constraint parameter of the allocation optimization model.

5. The method for regional virtual inventory pools and cross-site transfer processes in the power plant inventory sharing mechanism according to claim 1, characterized in that, The allocation risk value P mentioned in step S4 k It can be represented as: ; Where, p k d Let p be the probability of transportation delay. k r p represents the probability of route interruption. k v p represents the probability of vehicle malfunction. k h p represents the probability of the health status of supplies exceeding the limit. k s α1 represents the probability of stockouts after material is transferred from the supply depot; α1 to α5 are normalization coefficients.

6. The method for regional virtual inventory pools and cross-site transfer processes in the power plant inventory sharing mechanism according to claim 1, characterized in that, The decision variables of the allocation optimization model include the selection variables of the material supply site, the allocation quantity, the selection variables of the transport vehicle, and the selection variables of the transport route. The optimization model must at least satisfy constraints on material supply and demand balance, dynamic safety stock at material supply stations, model compatibility, vehicle load capacity, hazardous materials transportation, arrival time, predicted arrival health status, and regional service level.

7. The method for regional virtual inventory pools and cross-site transfer processes in the power plant inventory sharing mechanism according to claim 1, characterized in that, The execution deviations mentioned in step S6 include the deviation between the actual arrival time and the predicted arrival time, the deviation between the actual temperature and the predicted temperature of the goods in transit, the deviation between the actual health status and the predicted health status of the goods, the deviation between the actual status and the predicted vehicle status of the vehicles, and the deviation between the actual inventory and the predicted inventory at the depot; corresponding deviation thresholds are set for different deviations.

8. The method for regional virtual inventory pool and cross-site transfer process of power plant inventory sharing mechanism according to claim 1, characterized in that, When performing rolling re-optimization, materials that have been dispatched, completed transportation routes, delivered materials, and currently unreleased transportation vehicles are set as frozen variables. The solution is only recalculated for uncompleted material needs, adjustable vehicles, and unexecuted routes to prevent conflicts between the updated allocation plan and the actual execution status.

9. The method for regional virtual inventory pools and cross-site transfer processes in the power plant inventory sharing mechanism according to claim 1, characterized in that, The summary value of the regional inventory status snapshot, the allocation plan identifier, the digital twin model version, the execution deviation triggering condition, the relationship between the updated allocation plan and the initial allocation plan, and the material entry and exit handover results are written into the consortium blockchain. The raw real-time sensing data is stored in an off-chain database and its consistency is verified using the digest value.

10. A system for implementing a regional-level virtual inventory pool and cross-site transfer process for a power plant inventory sharing mechanism as described in any one of claims 1-9, characterized in that, include: The status data construction module is used to collect and standardize the status data of inventory, material health, equipment, environment, vehicles and materials in transit from multiple power plants to build a regional virtual inventory pool. The dynamic threshold and candidate scheme generation module is used to calculate the dynamic safety stock threshold and generate multiple candidate allocation schemes; The digital twin simulation module is used to pre-execute each candidate allocation plan and output the predicted arrival time, the predicted health status of the arriving materials, and the allocation risk value. The risk constraint optimization module is used to solve the initial allocation scheme based on the output; The execution deviation calculation module is used to update the digital twin model based on real-time status and calculate the execution deviation. The rolling re-optimization module is used to freeze executed or irreversible steps and re-solve the remaining allocation tasks when the execution deviation exceeds a threshold. The status update and trusted recording module is used to update the regional virtual inventory pool and record the allocation plan and its execution results.

Citation Information

Patent Citations

  • A logistics distribution and scheduling system based on digital twins

    CN114997802B

  • Electric power material storage management method, system and equipment based on distributed warehouse, and medium

    CN119515270A