A book scheduling method, system, device and medium of a self-service library

Abstract

The application relates to a book scheduling method, system, device and medium of a self-service library. The method comprises the following steps: preprocessing a book RFID reader / writer flow, a mobile self-service machine state flow, a user terminal event flow and a real-time GPS flow of a logistics vehicle to obtain a standardized event flow; based on a self-service system master table, performing position-level inventory estimation on each mobile self-service machine according to the standardized event flow to obtain a position-level inventory snapshot table; distributing the predicted demand of each mobile self-service machine in a preset length future time window according to book clusters to obtain a demand prediction result and an uncertainty index; generating a replenishment strategy of each mobile self-service machine based on the demand prediction result, book metadata and the position-level inventory snapshot table; and based on the load of a logistics vehicle, the capacity of a mobile self-service machine and time window constraints, obtaining a scheduling plan according to the replenishment strategy, the self-service system master table, book storage points and a position road network by minimizing a preset objective function. The method can improve service timeliness and resource utilization.

Description

Technical Field

[0001] This invention belongs to the field of book scheduling technology, and in particular relates to a book scheduling method, system, equipment and medium for a self-service library. Background Technology

[0002] With the development of digitalization and self-service technologies in libraries, RFID (Radio Frequency Identification) technology, fixed self-service borrowing and returning equipment, and traditional logistics scheduling solutions are gradually being applied to the field of library management.

[0003] In traditional technologies, RFID technology is mainly used for rapid book inventory, self-service borrowing and returning confirmation, and return sorting. By assigning a unique identifier to each book, batch reading and information verification are achieved at fixed reader nodes. Self-service relies on fixed-location self-service borrowing and returning machines and smart lockers to provide offline return and user self-pickup services, meeting basic self-service needs. Book scheduling mostly adopts static rules or offline batch scheduling modes, carrying out book replenishment and allocation work according to preset routes and fixed frequencies. In some scenarios, simple path planning methods are introduced to optimize transportation efficiency.

[0004] However, the above methods have limited application scenarios for RFID technology, which are limited to static inventory and borrowing / returning confirmation at fixed nodes. They do not incorporate the real-time location information and circulation trajectory of books into scheduling decisions, and cannot achieve dynamic perception of the book circulation process. Self-service facilities have fixed layouts and static capacity. In the face of time-based and temporary fluctuations in demand, it is difficult to flexibly expand capacity or adjust location, resulting in frequent shortages in hot areas and long-term idle equipment at some remote sites. Summary of the Invention

[0005] Therefore, it is necessary to provide a book scheduling method, system, equipment, and medium for self-service libraries that can improve book retrieval efficiency and reduce equipment idleness and operating costs, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a book dispatching method for a self-service library, including:

[0007] The system collects data from RFID readers, mobile self-service machines, user events, and real-time GPS data from logistics vehicles. It also preprocesses these data to obtain a standardized event stream.

[0008] Based on the self-service system master table, location-level inventory estimation is performed on each mobile self-service machine according to the standardized event flow to obtain a location-level inventory snapshot table; the self-service system master table includes the identifier, latitude and longitude, cabinet capacity, reader topology and equipment status of each mobile self-service machine, as well as the identifier, capacity, load limit and speed parameters of each logistics vehicle.

[0009] Based on location-level inventory snapshots, historical borrowing and returning records, user reservation information, and external environmental characteristics, the demand distribution of each mobile self-service machine within a preset future time window is predicted by book category, and the demand prediction results and their uncertainty indices are obtained.

[0010] Replenishment strategies for each mobile self-service machine are generated based on demand forecasting results, book metadata, and location-level inventory snapshot tables; the replenishment strategies include target replenishment quantity and replenishment priority for each book category.

[0011] Based on constraints such as logistics vehicle load, mobile self-service machine capacity, and time window, a scheduling plan is obtained by minimizing a preset objective function according to the replenishment strategy, the self-service system master table, book storage points, and location road network. The objective function includes total delivery delay, mobile self-service machine idle cost, logistics vehicle transportation cost, and uncertainty penalty term. The scheduling plan includes the logistics vehicle transportation task plan and the mobile self-service machine movement plan.

[0012] In one embodiment, the RFID reader stream, mobile self-service machine status stream, user event stream, and real-time GPS stream of logistics vehicles are preprocessed to obtain a standardized event stream, including:

[0013] The RFID reader stream, mobile self-service machine status stream, user terminal event stream, and real-time GPS stream of logistics vehicles are all given a unified timestamp, and the event streams with the unified timestamp are stored in the event database.

[0014] A positioning fusion algorithm based on reader topology and signal strength is executed on the book RFID reader stream in the event database to obtain the rough location of the book and its corresponding confidence level.

[0015] Deduplication is performed on the book RFID reader stream, mobile self-service machine status stream and user event stream in the event library, and the deduplicated event streams are aggregated by a sliding time window to obtain aggregated events.

[0016] The event stream, book rough position and its corresponding confidence and aggregated events in the event library are standardized by field standardization to obtain the standardized event stream.

[0017] In one embodiment, based on location-level inventory snapshots, historical borrowing and returning records, user reservation information, and external environmental characteristics, the demand distribution of each mobile self-service machine within a preset future time window is predicted by book category, yielding demand prediction results and their uncertainty indices, including:

[0018] A spatiotemporal graph is constructed based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information, and external environmental characteristics. The spatiotemporal graph includes mobile self-service machine nodes, geographical distance edges between mobile self-service machine nodes, and node attributes. Node attributes include historical borrowing and returning records, user reservation information, external environmental characteristics, and the location-level inventory snapshot table.

[0019] The spatiotemporal graph is input into the pre-trained joint model to obtain the demand prediction results. The joint model consists of a graph neural network with a spatial branch and a Transformer time series with a temporal branch. The joint model predicts the distribution of demand book clusters for each mobile self-service machine node within a predefined time box.

[0020] Uncertainty estimation is performed on the demand forecast results output by the joint model to obtain the uncertainty index corresponding to the demand forecast results.

[0021] In one embodiment, supplementary strategies for each mobile self-service machine are generated based on demand forecasting results, book metadata, and location-level inventory snapshot tables, including:

[0022] Cluster analysis is performed on the book metadata to obtain the available reserves and book reserve points for each book category; the book metadata includes, but is not limited to, ISBN, category, size, weight, and borrowing rate;

[0023] Based on demand forecasts, location-level inventory snapshots, and available reserves, the shortage of each book category at each mobile self-service machine is calculated, and the pre-allocation priority for each book category at each mobile self-service machine is calculated to obtain a replenishment strategy. The pre-allocation priority is obtained by weighted fusion of demand forecasts, alternative scores, and current stockout risks. The replenishment strategy includes the target replenishment quantity, priority, and desired completion time for each book category at each mobile self-service machine.

[0024] In one embodiment, based on logistics vehicle load, mobile self-service machine capacity, and time window constraints, a scheduling plan is obtained by minimizing a preset objective function according to a replenishment strategy, the self-service system master table, book storage points, and location road network, including:

[0025] The replenishment strategy, the self-service system master table, book storage points, location road network, and preset objective function are input into a pre-established mixed-integer programming model to obtain a scheduling plan. The scheduling plan includes the transportation task sequence of logistics vehicles, estimated arrival time, loaded book list, and the location points of the mobile self-service machines for each time period. The mixed-integer programming model solves for preset variables based on the capacity constraints and time window constraints of the mobile self-service machines. and ,in, Indicates logistics vehicles In the time window Is it a book storage point or a mobile self-service machine? Perform book loading docking. Indicates mobile self-service machine Location points on the location group network with location point The order relationship;

[0026] In response to operational emergencies, auction tasks are generated based on the corresponding scheduling plan. These tasks are then redistributed through a bidding auction mechanism using mobile self-service machines and logistics vehicles as autonomous agents, resulting in an updated scheduling plan. The bidding auction mechanism calculates the bid price for each task based on the current location, capacity, and expected profit of the mobile self-service machine, or the current location, load, and expected transportation cost of the logistics vehicle. Operational emergencies include, but are not limited to, mobile self-service machine delays, logistics vehicle delays, reader malfunctions, and sudden peak traffic.

[0027] In one embodiment, the method further includes:

[0028] Obtain batch reading data of the loaded book list from the on-board RFID reader of each logistics vehicle, and obtain loading confirmation events;

[0029] Based on the loading confirmation event, update the corresponding ISBN in the book metadata to the loading status and record the corresponding logistics vehicle identifier;

[0030] Acquire the RFID reader data of each mobile self-service machine, obtain delivery confirmation events, and update the location-level inventory snapshot table;

[0031] Based on the loading confirmation event and the delivery confirmation event, obtain the abnormal work order for which delivery was not confirmed, and trigger a manual processing request for the abnormal work order.

[0032] In response to a delivery confirmation event, a pickup notification and a one-time pickup code are pushed to the corresponding user's device.

[0033] In one embodiment, the method further includes:

[0034] Obtain actual operational data; actual operational data includes actual delivery time, user pickup rate, the difference between demand forecasts and actual demand, mobile self-service machine fault logs, and operating costs.

[0035] The parameters of the joint model are fine-tuned based on actual operational data to obtain updated model parameters and estimated prediction confidence.

[0036] The weighted weights corresponding to the pre-allocation priority and the weights corresponding to the objective function are calibrated based on actual operational data to obtain updated weight coefficients.

[0037] Secondly, this application also provides a self-service library book dispatching system, including:

[0038] The data module is used to stream data from RFID readers, mobile self-service machines, user events, and real-time GPS data from logistics vehicles. It also preprocesses these data to obtain standardized event streams.

[0039] The mobile self-service machine inventory module is used to perform location-level inventory estimation for each mobile self-service machine based on the self-service system master table and according to the standardized event flow, and obtain a location-level inventory snapshot table. The self-service system master table includes the identifier, latitude and longitude, cabinet capacity, reader topology and equipment status of each mobile self-service machine, as well as the identifier, capacity, load limit and speed parameters of each logistics vehicle.

[0040] The demand forecasting module is used to forecast the demand distribution of each mobile self-service machine within a preset future time window based on location-level inventory snapshot tables, historical borrowing and returning records, user reservation information, and external environmental characteristics, according to book categories, and to obtain the demand forecasting results and their uncertainty indicators.

[0041] The book replenishment module is used to generate replenishment strategies for each mobile self-service machine based on demand forecasting results, book metadata, and location-level inventory snapshot tables. The replenishment strategies include the target replenishment quantity and replenishment priority for each book category.

[0042] The scheduling module is used to obtain a scheduling plan based on the load of logistics vehicles, the capacity of mobile self-service machines, and time window constraints, according to the replenishment strategy, the self-service system master table, the book reserve points, and the location road network, by minimizing a preset objective function. The objective function includes total delivery delay, idle cost of mobile self-service machines, transportation cost of logistics vehicles, and uncertainty penalty terms. The scheduling plan includes the transportation task plan of logistics vehicles and the movement plan of mobile self-service machines.

[0043] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above-described self-service library book scheduling methods.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described self-service library book scheduling methods.

[0045] The aforementioned book scheduling methods, systems, equipment, and media for self-service libraries preprocess and generate standardized event streams from RFID reader streams, mobile self-service machine status streams, user-end event streams, and logistics vehicle GPS streams. This unifies raw observations from different sources and with varying formats into standardized data that can be directly consumed by upper-level modules, eliminating data isolation and facilitating collaborative work and audit traceability among modules in the chain. Using the self-service system's main table as the semantic background and the location-level inventory snapshot table as the dynamic cross-section, a static resource attribute and real-time status linkage view is formed. This ensures that scheduling and forecasting decisions are based on physical constraints while reflecting real-time availability and uncertainty. By predicting demand distribution by book category within future time windows, accompanied by uncertainty indicators, replenishment targets can be set before demand arrives, reducing the lag in temporary replenishment and transitioning from responsive operation and maintenance to forward-looking replenishment. Using book categories as replenishment and loading units helps alleviate the problem of sparse data per book, facilitating improved loading efficiency and transportation utilization while ensuring service quality. It also allows for a balance between substitutability and priority during replenishment, improving resource matching efficiency. During the scheduling phase, practical constraints such as vehicle load, mobile self-service machine capacity, and time window are incorporated and comprehensively represented through an objective function to achieve a balanced optimization of timeliness, cost, and risk, thereby obtaining an executable and goal-oriented transportation and mobility plan. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating the book scheduling method for the self-service library of the present invention.

[0048] Figure 2 This is a flowchart illustrating the steps of step S103.

[0049] Figure 3 This is a flowchart illustrating the steps of step S104.

[0050] Figure 4 This is a structural diagram of the book dispatching device for the self-service library of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0052] In one embodiment, such as Figure 1 As shown, a book scheduling method for a self-service library is provided. This embodiment illustrates the method applied to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0053] S101: Stream the data from the book RFID reader / writer stream, the mobile self-service machine status stream, the user terminal event stream, and the real-time GPS stream of the logistics vehicle. Then, preprocess the book RFID reader / writer stream, the mobile self-service machine status stream, the user terminal event stream, and the real-time GPS stream of the logistics vehicle to obtain a standardized event stream.

[0054] Indicatively, the book RFID reader stream refers to the continuous sequence of read and write events of RFID devices deployed in self-service machines, vehicles, or fixed locations for book tags, recording original fields such as the identifier mapped to the book ID or ISBN, reader identifier, timestamp, and signal strength; the mobile self-service machine status stream includes the online heartbeat, door lock / slot status, remaining available slots, and equipment fault or alarm information for each mobile self-service machine; the user-end event stream includes interactive events such as reservations, reservation cancellations, borrowing or returning requests initiated by readers on mobile devices or self-service terminals; and the real-time GPS stream of logistics vehicles records the vehicle's location information, speed, load status, etc.

[0055] The aforementioned heterogeneous event streams are preprocessed to convert them into standardized event streams in a unified format. This preprocessing includes time synchronization, field mapping and formatting, noise filtering and deduplication, as well as preliminary labeling of reading confidence. The output standardized event stream contains at least the following fields for each event: unified timestamp, source device identifier, target book identifier or event type, location or coarse location information, and confidence estimate.

[0056] S102. Based on the self-service system master table, perform location-level inventory estimation for each mobile self-service machine according to the standardized event flow to obtain a location-level inventory snapshot table. The self-service system master table includes the identifier, latitude and longitude, cabinet capacity, reader topology and equipment status of each mobile self-service machine, as well as the identifier, capacity, load limit and speed parameters of each logistics vehicle.

[0057] Indicatively, the self-service system's master table records the identifier, geographical location, cabinet capacity, reader / writer topology, and current status of each mobile self-service machine. It also records the identifier, capacity parameters, load limits, and speed parameters of each delivery vehicle. Specifically, location-level inventory estimation combines information from the master table with real-time observations from a standardized event stream at a given point in time to perform a cross-sectional estimate of the inventory of each mobile self-service machine, generating a location-level inventory snapshot table. This table not only records the known book list on each machine based on the most recent reliable read / write confirmation but also includes the confidence level and last observation time for each record to indicate data freshness and reliability. Reader access to books, replenishment priorities, and scheduling decisions are all based on the real-time inventory at each location. For example, if a mobile self-service machine is reserved for popular books by multiple readers within a short period, and the location-level inventory snapshot shows that the machine is currently experiencing reader / writer failures resulting in several books being in an unknown state, the scheduling unit should consider this uncertainty and prioritize replenishment or verification to avoid unfulfilled reservations.

[0058] S103. Based on location-level inventory snapshot tables, historical borrowing and returning records, user reservation information, and external environmental characteristics, predict the demand distribution of each mobile self-service machine within a preset future time window according to book category, and obtain the demand prediction results and their uncertainty index.

[0059] Indicatively, historical borrowing and returning records and user reservation information represent the long-term accumulated time series of borrowing behavior and the reservation demands that have been generated but not yet fulfilled, respectively, reflecting users' existing preferences and short-term demand signals. External environmental characteristics include seasonal factors, holiday arrangements, weather, and public transportation conditions that affect users' travel and borrowing tendencies, which help correct for temporal fluctuations in demand. Book clusters are semantic groupings of book sets. Based on themes, borrowing frequency, or interchangeability, a large number of individual items are reduced to clusters that are easy to predict and replenish, thereby achieving cluster-based resource allocation and priority decisions at the macro level. Specifically, fusing historical and real-time signals from different sources can improve the ability to identify short-term demand peaks and long-term trends. Predicting by book cluster rather than by book can improve the stability of predictions when data is sparse and facilitates integration with cluster-level replenishment requirements of replenishment strategies. The prediction output is the expected demand distribution for each self-service machine in a predefined time bin, such as the next 0–12 hours or 12–36 hours, and should also include uncertainty indicators.

[0060] S104. Generate replenishment strategies for each mobile self-service machine based on demand forecast results, book metadata, and location-level inventory snapshot tables; the replenishment strategies include the target replenishment quantity and replenishment priority by book category.

[0061] The goal of the replenishment strategy is to determine, under constraints, which book categories and quantities should be replenished to each mobile self-service machine, and to prioritize and schedule them accordingly, thereby maintaining a high service availability rate in the face of limited transportation resources and uncertain demand. Book metadata is a collection of information describing the attributes of individual books or book clusters, typically including but not limited to fields such as ISBN / ID, subject category, volume / size, weight, and borrowing rate. It can be used to measure the impact of single-item or cluster-level replenishment on vehicle capacity and slot occupancy, as well as their substitutability across different self-service machines. Demand forecasting results are used to determine the timing and scale of future shortages; location-level inventory snapshots reflect current available resources and uncertainties; and book metadata and existing available reserves are used to determine practically feasible replenishment quantities. Specifically, for each mobile self-service machine, the target replenishment quantities for several book categories, the priority of each category, and the suggested completion time (SLA) are listed. For example, when there is a large difference between the predicted demand and the current inventory for a certain book category and the uncertainty index is low, the category will be marked as high priority and added to the near-term replenishment list; conversely, when the uncertainty index is high, a conservative replenishment quantity can be set or a decision can be made after a closer observation, thereby achieving a balance between replenishment costs and failure risks.

[0062] S105. Based on the constraints of logistics vehicle load, mobile self-service machine capacity, and time window, a scheduling plan is obtained by minimizing the preset objective function according to the replenishment strategy, the self-service system master table, book storage points, and location road network. The objective function includes total delivery delay, mobile self-service machine idle cost, logistics vehicle transportation cost, and uncertainty penalty term. The scheduling plan includes the logistics vehicle transportation task plan and the mobile self-service machine movement plan.

[0063] This illustrative example illustrates the transportation task plan for logistics vehicles, specifying which book storage points or mobile self-service machines each vehicle should visit within a given time window to perform loading / delivery tasks, the estimated arrival time (ETA) for each route, and the corresponding loading list. It also outlines the movement plan for mobile self-service machines, specifying the locations where they should stop or remain during different time periods and the corresponding time window arrangements. When generating the scheduling plan, the plan considers the load and capacity limits of logistics vehicles, the cabinet capacity and slot limitations of mobile self-service machines, time window constraints (i.e., certain replenishments must be completed within a specified time limit to meet SLAs), and the time / distance characteristics of the road network. It also considers the book storage points (i.e., the source of shipments or warehouses) as the loading starting point. Specifically, the scheduling plan is generated based on a pre-defined objective function, which minimizes total delivery delay, mobile self-service machine idle cost, logistics vehicle transportation cost, and uncertainty penalty. Total delivery delay measures the cumulative lag in replenishment not being completed on time; mobile self-service machine idle cost reflects the opportunity cost of unused available counter space or service time; logistics vehicle transportation cost includes direct operating costs such as mileage, time, manpower, and energy consumption; and uncertainty penalty applies conservative costs to tasks with high predictability uncertainty to reduce stockout risk. The scheduling process seeks a feasible sequence of vehicle and mobile self-service machine actions to minimize the value of the objective function for book lending services while satisfying capacity and time window constraints.

[0064] In the aforementioned self-service library book scheduling method, the RFID reader / writer stream, mobile self-service machine status stream, user-end event stream, and vehicle GPS stream are preprocessed into a standardized event stream. This provides a unified data interface and semantics, eliminating integration bottlenecks caused by heterogeneous data formats, improving the data availability and consistency of subsequent inventory estimation, prediction, and scheduling modules, reducing decision-making errors caused by data inconsistency, supporting real-time tracking and auditing, and enabling timely identification and traceability of loading / delivery actions, thereby reducing disputes and error-finding costs. Based on the self-service system's master table and combined with the standardized event stream, location-level inventory estimation is performed on each mobile self-service machine to obtain a location-level inventory snapshot table, forming an inventory view synthesized from static attributes and real-time observations. This allows decisions to be constrained by both physical capacity and real-time availability, improving inventory visibility, reducing mismatches and duplicate deliveries caused by information blockage, providing a reliable basis for replenishment priority determination, and helping to prioritize fulfilling real and redeemable reservations and demands. Based on location-level inventory snapshots, historical borrowing and returning data, user reservations, and external environmental characteristics, future demand windows are predicted by book category clusters, and uncertainty indicators are provided. This allows for the early identification of demand hotspots in time and space, reducing reservation cancellations and reader waiting times. The uncertainty indicators support risk perception, enabling conservative / buffered strategies to be introduced in replenishment and scheduling, thereby reducing stockouts or over-transportation caused by prediction errors. Cluster-based prediction improves prediction stability in data-sparse scenarios and facilitates integration with loading units, enhancing replenishment efficiency. Cluster-level replenishment strategies are generated based on demand forecasts, book metadata, and inventory snapshots, providing clear and actionable replenishment targets. This reduces suboptimal allocations caused by unclear targets in the upper-level scheduler. Prioritization mechanisms prioritize key reservations or high-impact clusters when resources are limited, improving user experience and critical service levels. Utilizing book metadata to consider physical attributes such as volume and weight helps to more rationally utilize shelf space and vehicle load, reducing transportation and loading costs. During the scheduling phase, constraints such as logistics vehicle load, mobile self-service machine capacity, and time windows are incorporated. The objective function, which includes delivery delay, machine idle cost, transportation cost, and uncertainty penalty, is minimized to output transportation tasks and mobile machine movement plans. This generates executable plans that meet actual physical constraints, avoiding infeasibility or frequent manual intervention caused by ignoring capacity or time windows. By weighting the objective function, business-oriented trade-offs can be made between timeliness, cost, and utilization, facilitating priority adjustments based on operational goals. Incorporating uncertainty penalty into the objective function reduces blind allocation of high-risk tasks, thereby improving replenishment success rate and service stability. The output transportation task plan and movement plan can be directly issued to the execution layer for real-time monitoring, shortening the delay from decision-making to execution.

[0065] In one embodiment, the RFID reader stream, mobile self-service machine status stream, user event stream, and real-time GPS stream of logistics vehicles are preprocessed to obtain a standardized event stream, including:

[0066] S11. Unify the timestamps of the RFID reader stream, mobile self-service machine status stream, user terminal event stream, and real-time GPS stream of logistics vehicles, and store the event streams with unified timestamps into the event database.

[0067] This example illustrates how time-standardization is performed on each raw event from the RFID reader / writer stream, mobile self-service machine status stream, user-end event stream, and real-time GPS stream of logistics vehicles. This eliminates the impact of clock skew and network latency between various data acquisition devices on event sorting and correlation. For instance, clock calibration is performed for each acquisition end based on the Network Time Protocol (NTP), and the server aligns and converts the received timestamp with the device's original timestamp. Furthermore, time-synchronized events are mapped and standardized using a unified event field format, while numerical fields undergo unit standardization and type validation. After field standardization, the standardized events are written to the event database. The event database supports streaming writing and retrieval. Optionally, the database entry process protects against anomalies such as entry failure, duplicate writing, or excessively large data volumes, such as temporary buffering, rate limiting, or recording abnormal events in an error log table and triggering alarms.

[0068] S12. Perform a positioning fusion algorithm based on reader topology and signal strength on the book RFID reader stream in the event database to obtain the rough location of the book and its corresponding confidence level.

[0069] Furthermore, using records of RFID reader streams for books from the event database as input, and combining reader deployment topology information (i.e., the geometric location and coverage relationship of self-service machines / vehicle-mounted / ground-based readers) with the signal strength index (RSSI) information carried in the readings, location fusion is performed to infer the coarse location of the book tag and estimate the corresponding confidence level, determining which location unit the tag is most likely to be located in, such as inside a mobile self-service machine, in a vehicle, or outside the system coverage area. Specifically, for the same tag_id readings from different readers within the same time window, a score is calculated for each possible location based on the relative relationship between each reader and the location unit, RSSI strength, the reader's historical reliability, and access duration. This score is then normalized to a probability distribution, thus outputting the most likely location and confidence value.

[0070] S13. Deduplicate the event streams of the book RFID reader / writer, the mobile self-service machine, and the user terminal, and then aggregate the deduplicated event streams using a sliding time window to obtain aggregated events.

[0071] Furthermore, frequent, redundant, or nearly duplicate readings from different sources in the event database are deduplicated and aggregated within time windows to reduce noise, lower the upstream processing load, and create stable observations usable for inventory and forecasting. Optionally, deduplication is performed based on tag_id and a time difference threshold: for records with the same tag_id, the same event type, and reported multiple times within a short period, duplicate entries are removed by comparing fields such as timestamps, retaining only the most reliable one or merging its attributes, such as retaining the maximum RSSI or the earliest / latest timestamp. Deduplication rules can be configured to retain entries based on device priority, such as prioritizing entries reported by vehicle-mounted readers if that reader has historically been more reliable, or retaining entries based on a confidence threshold.

[0072] Furthermore, events are aggregated using a sliding time window to mitigate the impact of instantaneous fluctuations. The sliding window is typically set to several seconds to tens of seconds. Within each time window, statistical characteristics are calculated for multiple readings of the same tag_id or the same location unit, such as the number of occurrences, maximum / average RSSI, first / last time, and a list of different readers involved. The statistical results are then encapsulated as aggregated events. The output of aggregated events retains temporal resolution while reducing data volume, enabling inventory estimation and location fusion to operate with lower noise levels.

[0073] Optionally, when location fusion returns a low confidence level, the aggregation module can trigger an active rereading strategy and record the request in the aggregation event for use by operations and maintenance or automatic retry mechanisms.

[0074] S14. Standardize the event stream, book rough position and its corresponding confidence and aggregated events in the event library to obtain the standardized event stream.

[0075] The raw events, aggregated events, and location fusion outputs from the event database are unified and formatted at the field level to form standardized event stream records. Field standardization specifically includes unifying event semantics, such as mapping inbound and outbound reads from different sources to consistent event type enumeration values, determining the location identifier used in the location field, and merging the coarse location and confidence scores obtained from location fusion into two standard fields: inferred_location and confidence. Each record in the standardized event stream includes timestamp, source_type, source_id, tag_id, event_type, inferred_location, confidence, gps_pos_if_any, aggregated_metrics, raw_ref, etc., along with metadata indicating the record's generation path.

[0076] In one embodiment, based on location-level inventory snapshots, historical borrowing and returning records, user reservation information, and external environmental characteristics, the demand distribution of each mobile self-service machine within a preset future time window is predicted by book category, yielding demand prediction results and their uncertainty indices, including:

[0077] S201. Construct a spatiotemporal graph based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information, and external environmental characteristics. The spatiotemporal graph includes mobile self-service machine nodes, geographical distance edges between mobile self-service machine nodes, and node attributes. Node attributes include historical borrowing and returning records, user reservation information, external environmental characteristics, and the location-level inventory snapshot table.

[0078] Schematic, the graph's node set consists of all mobile self-service machines, each uniquely identified. The edge set is weighted based on geographical distance between nodes or historical user flow probability. Geographic edge weights can be inversely mapped to the Euclidean distance or road network travel time between two nodes, while user flow probability can be obtained from historical borrowing and returning records and statistics on users' cross-point borrowing and returning behavior. These two can be linearly or non-linearly combined according to business needs to generate the final edge weights, thus reflecting the strength of mutual influence between nodes. Furthermore, time-series node attributes are constructed for each node, including at least aggregated features of historical borrowing and returning records, such as hourly or per-period borrowing / return counts, arrival and cancellation counts, as well as the current inventory quantity and confidence level of each book cluster in the location-level inventory snapshot table, recent equipment status (available slots and reader / writer fault indicators), and external environmental features, such as weather codes, holiday status, and nearby traffic disturbance indicators. For example, numerical features are standardized, categorical features are mapped using one-hot encoding or embedding, and time features are periodically encoded to help the model capture periodic patterns. Optionally, to facilitate prediction by book cluster, the inventory and historical borrowing / returning counts in the node attributes should be aggregated according to the predefined book cluster dimension to obtain the observation sequence of each node for each cluster at each time step.

[0079] S202. Input the spatiotemporal graph into the pre-trained joint model to obtain the demand prediction results. The joint model consists of a graph neural network with a spatial branch and a Transformer time series with a temporal branch. The joint model predicts the distribution of demand book clusters for each mobile self-service machine node within a predefined time box.

[0080] This example illustrates how, using a spatiotemporal graph and node temporal observation sequence features as input, a joint model generates a predicted distribution of book category cluster demand for each mobile self-service machine within a predefined time bin. The joint model consists of two collaborative modules: a spatial branch employs a graph neural network (GNN) to capture topological coupling and neighborhood influence between nodes, and a temporal branch uses a Transformer structure to capture long-term dependencies and temporal patterns within historical sequences of nodes. For instance, during the training phase, the prediction target and label generation method are clearly defined: for each training sample node i at time t, the borrowing volume or reservation fulfillment demand aggregated by time bin within the future time window is calculated as the target count by book category. Based on the target distribution characteristics, if it is sparse counting, a Poisson or negative binomial distribution is used; if it is dense numerical data, a mean squared error loss can be used, and a loss function L is set. During training, the input node feature sequence (Length is the historical window T), adjacency matrix A, and corresponding label Y are used. Mini-batch training is employed to optimize the joint model parameters. Specifically, the spatial branch can use Graph Convolutional Network (GCN), Graph Attention (GAT), or a dynamic graph network based on time-varying edge weights to spatially aggregate the node features at each time step, outputting the spatial embedding for each step. The temporal branch uses this temporal embedding as input to the Transformer, which, after passing through multi-head self-attention and stacked feedforward layers, outputs a predicted representation for several future time bins. The output head then maps the representation to the predicted distribution of each book category cluster. Training can employ early stopping, learning rate scheduling, and regularization to prevent overfitting. To address data imbalance, cluster-level weighting or resampling strategies can be introduced into the loss function.

[0081] S203. Estimate the uncertainty of the demand forecast results output by the joint model to obtain the uncertainty index corresponding to the demand forecast results.

[0082] Furthermore, a confidence interval or variance estimate is appended to the output of the joint model to provide a risk metric for the scheduling module. Uncertainty estimation can be achieved through MC dropout, model ensemble, Bayesian approximation, or quantile regression / distributed prediction. Illustratively, MC dropout is used: the joint model is trained normally during the training phase with dropout added to certain layers; during the inference phase, dropout is enabled and multiple forward propagations are performed, collecting the mean and variance of the M prediction results. The mean is used as the final prediction point estimate, and the variance or confidence interval is used as the uncertainty metric. Optionally, the uncertainty metric can include not only the overall variance but also the uncertainty within the model to reflect the uncertainty caused by the sparsity of model parameters or training data, as well as observational uncertainty to reflect the noise or unpredictability of the input data itself.

[0083] In one embodiment, supplementary strategies for each mobile self-service machine are generated based on demand forecasting results, book metadata, and location-level inventory snapshot tables, including:

[0084] S301. Perform cluster analysis on the book metadata to obtain the available reserves and book reserve points for each book category; the book metadata includes, but is not limited to, ISBN, category, size, weight, and borrowing rate.

[0085] Intuitively, individual book information is grouped into manageable clusters (book_clusters), and the available reserves of each cluster and its corresponding book storage location are identified based on the existing inventory distribution, facilitating subsequent replenishment and allocation by cluster. Book metadata typically includes, but is not limited to, ISBN / ID, subject category tags, print volume and dimensions, weight, recent borrowing rate, and whether it is a bestseller / reservation hotspot, representing both the physical cost of the book and its circulation intensity and substitutability. Specifically, numerical attributes are normalized or standardized, and textual or categorical attributes are encoded; derived features, such as the growth rate over the past N weeks and seasonality coefficients, can be calculated for statistical quantities such as borrowing rates. To facilitate clustering and reduce dimensionality, principal component analysis (PCA) or embedding-based dimensionality reduction methods can be used to map high-dimensional features to a low-dimensional vector space. Clustering methods can include K-means, spectral clustering, or hierarchical clustering. The goal of clustering is to ensure that books within a cluster have relatively homogeneous borrowing and physical characteristics, thereby facilitating the development of unified replenishment and loading rules by cluster. The clustering output consists of several cluster identifiers and a list of book IDs contained in each cluster. Cluster-level statistics are calculated for each cluster, such as average volume within the cluster, average borrowing rate, and long-tail / hotspot ratio, to measure the cluster's transportation and inventory costs. Furthermore, the data from all book storage points are combined with the cluster mapping to calculate the available inventory for each cluster at each book storage point. Available inventory refers to the number of copies in stock that can be used for replenishment, excluding those under maintenance, reserved, or allocated but not yet shipped. Book storage points include central warehouses, branch libraries, partner bookstores, or other long-term inventory locations. For example, after clustering, a fiction cluster (book_cluster_F) is obtained, with a volume of 0.002 cubic meters and an average borrowing rate of 15 books / week. There are 120 available books in the central warehouse, 30 in branch library A, and 5 in branch library B.

[0086] S302. Based on the demand forecast results, location-level inventory snapshot table, and available reserves, calculate the shortage quantity of each book category at each mobile self-service machine, and calculate the pre-allocation priority for each book category for each mobile self-service machine to obtain the replenishment strategy; the pre-allocation priority is obtained by weighted fusion of demand forecast results, alternative score, and current stockout risk; the replenishment strategy includes the target replenishment quantity, priority, and desired completion time for each book category by each mobile self-service machine.

[0087] Indicative of each mobile self-service machine and each cluster In the preset future time window Calculate demand forecast And compare the known available quantity I(i,c) and the quantity already in transit for cluster c at the location-level inventory snapshot table with the location-level inventory snapshot table. In comparison, according to the formula The basic gap quantity is obtained, which can be divided into emergency gap and regular gap, and their corresponding uncertainty indices are recorded. After calculating the gap, a pre-allocation priority assessment is performed. The pre-allocation priority is calculated using a weighted fusion method. .in, The value is normalized to eliminate scale differences in the predicted demand; (Alternative Score) measures the existence of alternative resources in the spatial neighborhood or within an acceptable time range. It can be calculated by the available reserves of the cluster from neighboring nodes or reserve points and the arrival time. The current stockout risk is determined by the current inventory coverage days, forecast uncertainty, and expected replenishment delays. , , The weights are configurable, and the initial values ​​can be set by regression analysis of historical operating data or by expert experience.

[0088] After determining the Priority, the target replenishment quantity (target_qty) is determined. The basic gap can be amplified based on priority and SLA rules. For example, under high priority and low uncertainty, replenishment can be increased above the predicted average to ensure success. The target replenishment quantity is also affected by the available reserves at book storage points and allocation strategies. For instance, a mobile self-service machine M predicts a demand of D=10 for the fiction cluster in next 12 hours, current inventory I=2, and in-transit=1, resulting in a gap of 7. Nearby storage point A has 20 books with an ETA of 2 hours, and storage point B has 3 books with an ETA of 4 hours. The substitutability score is low for M, and the prediction uncertainty std=1.5. Let the initial weights be... , , A higher priority was calculated. Based on the SLA that needs to be completed within 24 hours and the available grid limits, target_qty was set to 9, and the preferred supply point A and transportation priority were listed in the supplementary strategy for the scheduler to generate specific tasks.

[0089] Optionally, when candidate supply points lack sufficient available reserves or transportation disruptions render some supply points inaccessible, the replenishment strategy will automatically flag the problem and generate alternative suggestions, such as picking up goods from more distant points and merging multi-point transportation, notifying users, or adjusting appointments. At the same time, the replenishment strategy will be marked as requiring manual review or triggering emergency resource allocation.

[0090] In one embodiment, based on logistics vehicle load, mobile self-service machine capacity, and time window constraints, a scheduling plan is obtained by minimizing a preset objective function according to a replenishment strategy, the self-service system master table, book storage points, and location road network, including:

[0091] S21. Input the replenishment strategy, the self-service system master table, book storage points, location road network, and preset objective function into the pre-established mixed integer programming model to obtain the scheduling plan. The scheduling plan includes the transportation task sequence of logistics vehicles, the estimated arrival time, the list of books to be loaded, and the location points of the mobile self-service machines in each time period. The mixed integer programming model solves the preset variables based on the capacity constraints and time window constraints of the mobile self-service machines. and ,in, Indicates logistics vehicles In the time window Is it a book storage point or a mobile self-service machine that performs book loading and docking? Indicates mobile self-service machine Location points on the location group network The order relationship with position point j.

[0092] Indicatively, the input information is mapped to the model's parameters and constraint matrices. The supplementary strategy is a set of several target supplementary items, each specifying the target supplement quantity, priority, and expected completion time (SLA). The self-service system master table provides the static attributes of mobile self-service machines and vehicles, including location, capacity, speed, and load limits. The book storage points and location road network provide feasible transportation origins and spatiotemporal distance / time cost matrices based on the actual road network or estimated travel time.

[0093] Based on the above parameters, decision variables and objective function are constructed. The main decision variables include binary variables. and , Indicates logistics vehicles In the time window Go to the destination? That is, book storage points or mobile self-service machines perform loading or delivery tasks; Used to indicate a mobile self-service machine Is it in its route? Visit immediately afterwards This is used for path order and time expansion. Optionally, integer variables can also be introduced. Indicates a location A certain book category was assigned The replenishment quantity, and continuous variables Indicates logistics vehicles Expected arrival The time frame. The objective function, composed of weighted multi-objectives, is formalized as minimization. Lateness is the cumulative delay exceeding the SLA, which can be the cumulative positive difference between arrival time and target completion time. IdleCost is the estimated opportunity cost of idle or unused slots of mobile self-service machines during their service hours. TransportCost is the weighted sum of vehicle travel distance, time, and operating unit price. UncertaintyPenalty depends on the penalty term for high-risk tasks based on forecast uncertainty. If the forecast uncertainty is high, a buffer needs to be added to the allocation to reduce the probability of failure.

[0094] Among these constraints, the vehicle capacity constraint stipulates that the cumulative number of books loaded on each vehicle along its travel route must not exceed its maximum load capacity, and the time window constraint stipulates that for each replenishment task... Set an acceptable start and end time interval. Vehicles must arrive and complete loading / delivery within this interval. Flow continuity constraint requires that vehicle paths form a valid sequence. Each replenishment target quantity must be satisfied or allowed. This can be represented by an interpretable undersatisfied variable and included in the penalty of the objective function. Also include constraints on the capacity and slot occupancy of mobile self-service machines.

[0095] S22. In response to an operational emergency, an auction task is generated based on the corresponding scheduling plan. The task is then redistributed through a bidding auction mechanism using each mobile self-service machine and each logistics vehicle as autonomous agents, resulting in an updated scheduling plan. In the bidding auction mechanism, the bid for each auction task is calculated based on the current location, current capacity, and expected profit of each mobile self-service machine, or the current location, current load, and expected transportation cost of each logistics vehicle. Operational emergencies include, but are not limited to, mobile self-service machine delays, logistics vehicle delays, reader / writer malfunctions, and sudden peak periods.

[0096] When an operational emergency is detected, tasks affecting one or more originally planned tasks, such as vehicle n1's anticipated delay preventing it from loading several self-service machines on time within time window t, are marked as auction tasks. Each auction task includes information such as task type (loading or delivery), destination location, expected completion time, required replenishment clusters and quantities, and preliminary reward or cost estimates. Tasks requiring reallocation are added to the auction pool. Bidding agents are divided into logistics vehicle agents and mobile self-service machine agents. When valuing an auction task, the logistics vehicle agent assesses its current location, current load, available remaining capacity, estimated transport time to the task location, and the incremental cost of completing the task for its subsequent plans. Correspondingly, if the mobile self-service machine agent has mobility and leeway, when evaluating a mobile task, such as moving itself to approach a replenishment source or receive a loading task, it calculates its valuation based on its current location information, the number of currently available slots, and the expected revenue / service guarantee brought by the task. To standardize bidding expression, a unified bidding formula is used. ,in This is an estimate of the additional costs required for the agent to complete the task. For vehicles, this includes additional mileage, manpower, and time costs; for mobile self-service machines, it includes mobile energy consumption and service interruption costs. It is a measure of the urgency / priority of the task. The configuration parameters are used to balance cost sensitivity and timeliness.

[0097] Optionally, the auction process can employ a scalable closed or open auction mechanism. In a closed auction, agents submit a bid within a limited time window, and after collecting all bids, the winner is determined either by the highest bidder or by a cost-efficiency composite score. Open auctions allow for multiple rounds of bidding and are suitable for use when there are fewer participating agents and more refined trade-offs are required. To ensure fairness and prevent a single agent from dominating for an extended period, a resource usage discount factor or a historical participation penalty can be introduced into the auction rules. After winning the auction, the task is immediately assigned to the winning agent, and an execution instruction is issued. This instruction is simultaneously written into a standardized event stream for subsequent tracking.

[0098] In one embodiment, the method further includes:

[0099] S31. Obtain batch reading data of the loaded book list from the on-board RFID reader of each logistics vehicle, and obtain a loading confirmation event.

[0100] As an illustration, when the vehicle arrives at the book storage point and begins loading, the onboard RFID reader continuously performs batch readings of book tags within the loading area. The read tag_id list is then matched against the loading list in the scheduling plan, generating a loading confirmation event for each book.

[0101] S32. Based on the loading confirmation event, update the corresponding ISBN to the loading status in the book metadata and record the corresponding logistics vehicle identifier.

[0102] Based on the confirmed loading list in the loading confirmation event, perform a status change operation on each tag_id in the list in the book metadata table, that is, update the status field of the book to in_transit, and write the logistics vehicle identifier in the corresponding metadata record.

[0103] S33. Obtain the reading data of the RFID reader / writer of each mobile self-service machine, obtain the delivery confirmation event, and update the location-level inventory snapshot table.

[0104] When a vehicle arrives at the target mobile self-service machine and performs a delivery operation, successful delivery is determined by combining data from the vehicle-mounted reader and the ground / airborne reader of the target machine. Optionally, a two-sided confirmation mechanism is adopted: after the vehicle unloads and places books, the vehicle-mounted reader should read the departure status of the corresponding tag_id within a short time window, or the tag should no longer be detected in the vehicle-mounted reader's readings; simultaneously, the reader of the target mobile self-service machine should read the same tag_id within the same or slightly later time window and report its presence in the cabinet. If reliable readings are obtained from both sides, a delivery confirmation event is generated. After generating the delivery confirmation event, the location-level inventory snapshot table is updated, changing the status of the corresponding tag_id from in_transit to in_machine (in the mobile self-service machine), and recording the arrival event, source vehicle, etc., while also updating the available slot count and confidence level of the mobile self-service machine.

[0105] S34. Based on the loading confirmation event and the delivery confirmation event, obtain the abnormal work order for which delivery was not confirmed, and trigger a manual processing request for the abnormal work order.

[0106] Furthermore, the en route record of each tag_id listed in the loading confirmation event is matched and verified with the corresponding destination delivery confirmation event. If a tag_id exists in the loading confirmation event but does not appear in the destination delivery confirmation event within an acceptable time window, it is determined as a failure to confirm delivery and an abnormal work order is generated. Optionally, for possible reader misreading or temporary network failures, read / write retry can be automatically initiated, the vehicle / ground reader can be requested to scan again, or a remote device restart can be triggered; for confirmed cases of under-delivery or incorrect delivery, the nearest available vehicle can be dispatched to perform re-delivery or re-delivery can be prioritized in the next delivery; for suspected loss or mis-taken out of the warehouse and unable to be recovered in a short time, a manual on-site verification process can be triggered.

[0107] S35. In response to the delivery confirmation event, push a pickup notification and a one-time pickup code to the corresponding user terminal.

[0108] Upon receiving a delivery confirmation event and confirming in the location-level inventory snapshot table that the book has entered the designated mobile self-service machine's pick-up status, the system retrieves the reservation record or pending user information corresponding to the book's tag_id, generating a one-time pickup code (OTP). This pickup code includes a single-use authentication code, validity period, corresponding book information, and pickup machine location. The code is encrypted and signed to prevent tampering. A pickup notification is then sent through the user's preferred channel, containing the pickup code, pickup machine number, expiration time, and a simple operation guide.

[0109] In one embodiment, the method further includes:

[0110] S41. Obtain actual operational data; actual operational data includes actual delivery time, user pickup rate, the difference between demand forecast and actual demand, mobile self-service machine fault logs, and operating costs.

[0111] Actual delivery time is the actual start and end time and duration of each delivery task; user pickup rate is the ratio of users who arrive on time and successfully pick up their items; the difference between demand forecast and actual demand is the error sequence between the forecast value and the actual observed lending / reservation redemption amount; mobile self-service machine fault / alarm logs include events such as reader / writer failure, door lock failure, and compartment abnormality, as well as their processing time; operating costs include vehicle mileage, labor costs, energy consumption, and maintenance costs.

[0112] S42. Fine-tune the parameters of the joint model based on actual operational data to obtain updated model parameters and estimated prediction confidence.

[0113] S43. Based on actual operational data, calibrate the weighted weights corresponding to the pre-allocation priority and the weights corresponding to the objective function to obtain the updated weight coefficients.

[0114] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0115] Based on the same inventive concept, this application also provides a self-service library book scheduling system for implementing the book scheduling method of the self-service library described above. The solution provided by this system is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more self-service library book scheduling system embodiments provided below can be found in the limitations of the self-service library book scheduling method described above, and will not be repeated here.

[0116] In one exemplary embodiment, such as Figure 4 As shown, a self-service library book dispatching system is provided, including:

[0117] Data module 401 is used to stream the RFID reader stream, mobile self-service machine status stream, user terminal event stream and real-time GPS stream of logistics vehicles, and preprocess the RFID reader stream, mobile self-service machine status stream, user terminal event stream and real-time GPS stream of logistics vehicles to obtain a standardized event stream.

[0118] The mobile self-service machine inventory module 402 is used to perform location-level inventory estimation for each mobile self-service machine based on the self-service system master table and according to the standardized event flow, and obtain a location-level inventory snapshot table. The self-service system master table includes the identifier, latitude and longitude, cabinet capacity, reader topology and equipment status of each mobile self-service machine, as well as the identifier, capacity, load limit and speed parameters of each logistics vehicle.

[0119] The demand forecasting module 403 is used to forecast the demand distribution of each mobile self-service machine within a preset length future time window based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information and external environmental characteristics, according to book category clusters, and to obtain the demand forecasting results and their uncertainty index.

[0120] The book replenishment module 404 is used to generate replenishment strategies for each mobile self-service machine based on demand forecast results, book metadata, and location-level inventory snapshot tables; the replenishment strategies include the target replenishment quantity and replenishment priority by book category;

[0121] The scheduling module 405 is used to obtain a scheduling plan based on the load of logistics vehicles, the capacity of mobile self-service machines, and time window constraints, according to the replenishment strategy, the self-service system master table, the book reserve points, and the location road network, by minimizing a preset objective function. The objective function includes the total delivery delay, the idle cost of mobile self-service machines, the transportation cost of logistics vehicles, and an uncertainty penalty term. The scheduling plan includes the transportation task plan of logistics vehicles and the movement plan of mobile self-service machines.

[0122] In one embodiment, the data module 401 is further configured to:

[0123] The RFID reader stream, mobile self-service machine status stream, user terminal event stream, and real-time GPS stream of logistics vehicles are all given a unified timestamp, and the event streams with the unified timestamp are stored in the event database.

[0124] A positioning fusion algorithm based on reader topology and signal strength is executed on the book RFID reader stream in the event database to obtain the rough location of the book and its corresponding confidence level.

[0125] Deduplication is performed on the book RFID reader stream, mobile self-service machine status stream and user event stream in the event library, and the deduplicated event streams are aggregated by a sliding time window to obtain aggregated events.

[0126] The event stream, book rough position and its corresponding confidence and aggregated events in the event library are standardized by field standardization to obtain the standardized event stream.

[0127] In one embodiment, the demand forecasting module 403 is further configured to:

[0128] A spatiotemporal graph is constructed based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information, and external environmental characteristics. The spatiotemporal graph includes mobile self-service machine nodes, geographical distance edges between mobile self-service machine nodes, and node attributes. Node attributes include historical borrowing and returning records, user reservation information, external environmental characteristics, and the location-level inventory snapshot table.

[0129] The spatiotemporal graph is input into the pre-trained joint model to obtain the demand prediction results. The joint model consists of a graph neural network with a spatial branch and a Transformer time series with a temporal branch. The joint model predicts the distribution of demand book clusters for each mobile self-service machine node within a predefined time box.

[0130] Uncertainty estimation is performed on the demand forecast results output by the joint model to obtain the uncertainty index corresponding to the demand forecast results.

[0131] In one embodiment, the book replenishment module 404 is further configured to:

[0132] Cluster analysis is performed on the book metadata to obtain the available reserves and book reserve points for each book category; the book metadata includes, but is not limited to, ISBN, category, size, weight, and borrowing rate;

[0133] Based on demand forecasts, location-level inventory snapshots, and available reserves, the shortage of each book category at each mobile self-service machine is calculated, and the pre-allocation priority for each book category at each mobile self-service machine is calculated to obtain a replenishment strategy. The pre-allocation priority is obtained by weighted fusion of demand forecasts, alternative scores, and current stockout risks. The replenishment strategy includes the target replenishment quantity, priority, and desired completion time for each book category at each mobile self-service machine.

[0134] In one embodiment, the scheduling module 405 is further configured to:

[0135] The replenishment strategy, the self-service system master table, book storage points, location road network, and preset objective function are input into a pre-established mixed-integer programming model to obtain a scheduling plan. The scheduling plan includes the transportation task sequence of logistics vehicles, estimated arrival time, loaded book list, and the location points of the mobile self-service machines for each time period. The mixed-integer programming model solves for preset variables based on the capacity constraints and time window constraints of the mobile self-service machines. and ,in, Indicates logistics vehicles In the time window Is it a book storage point or a mobile self-service machine that performs book loading and docking? Indicates mobile self-service machine Location points on the location group network The order relationship with position point j;

[0136] In response to operational emergencies, auction tasks are generated based on the corresponding scheduling plan. These tasks are then redistributed through a bidding auction mechanism using mobile self-service machines and logistics vehicles as autonomous agents, resulting in an updated scheduling plan. The bidding auction mechanism calculates the bid price for each task based on the current location, capacity, and expected profit of the mobile self-service machine, or the current location, load, and expected transportation cost of the logistics vehicle. Operational emergencies include, but are not limited to, mobile self-service machine delays, logistics vehicle delays, reader malfunctions, and sudden peak traffic.

[0137] In one embodiment, a monitoring module is also included, for:

[0138] Obtain batch reading data of the loaded book list from the on-board RFID reader of each logistics vehicle, and obtain loading confirmation events;

[0139] Based on the loading confirmation event, update the corresponding ISBN in the book metadata to the loading status and record the corresponding logistics vehicle identifier;

[0140] Acquire the RFID reader data of each mobile self-service machine, obtain delivery confirmation events, and update the location-level inventory snapshot table;

[0141] Based on the loading confirmation event and the delivery confirmation event, obtain the abnormal work order for which delivery was not confirmed, and trigger a manual processing request for the abnormal work order.

[0142] In response to a delivery confirmation event, a pickup notification and a one-time pickup code are pushed to the corresponding user's device.

[0143] In one embodiment, a feedback module is also included for:

[0144] Obtain actual operational data; actual operational data includes actual delivery time, user pickup rate, the difference between demand forecasts and actual demand, mobile self-service machine fault logs, and operating costs.

[0145] The parameters of the joint model are fine-tuned based on actual operational data to obtain updated model parameters and estimated prediction confidence.

[0146] The weighted weights corresponding to the pre-allocation priority and the weights corresponding to the objective function are calibrated based on actual operational data to obtain updated weight coefficients.

[0147] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps in the above method embodiments.

[0148] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0149] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0150] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A book scheduling method for a self-service library, characterized in that, The method includes: The system collects data from the RFID reader stream, the mobile self-service machine status stream, the user terminal event stream, and the real-time GPS stream of the logistics vehicle. It also preprocesses the RFID reader stream, the mobile self-service machine status stream, the user terminal event stream, and the real-time GPS stream of the logistics vehicle to obtain a standardized event stream. Based on the self-service system master table, location-level inventory estimation is performed on each mobile self-service machine according to the standardized event flow to obtain a location-level inventory snapshot table; the self-service system master table includes the identifier, latitude and longitude, cabinet capacity, reader topology and equipment status of each mobile self-service machine, as well as the identifier, capacity, load limit and speed parameters of each logistics vehicle. Based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information and external environmental characteristics, the demand distribution of each mobile self-service machine within a preset length future time window is predicted according to book category, and the demand prediction results and their uncertainty index are obtained. Based on the demand forecast results, book metadata, and the location-level inventory snapshot table, a replenishment strategy is generated for each mobile self-service machine; the replenishment strategy includes the target replenishment quantity and replenishment priority by book category. Based on the constraints of logistics vehicle load, mobile self-service machine capacity, and time window, a scheduling plan is obtained by minimizing the preset objective function according to the replenishment strategy, the self-service system master table, book reserve points, and location road network. The objective function includes total delivery delay, mobile self-service machine idle cost, logistics vehicle transportation cost, and uncertainty penalty term. The scheduling plan includes the logistics vehicle transportation task plan and the mobile self-service machine movement plan.

2. The method according to claim 1, characterized in that, The preprocessing of the book RFID reader stream, mobile self-service machine status stream, user terminal event stream, and real-time GPS stream of logistics vehicles yields a standardized event stream, including: The RFID reader stream, mobile self-service machine status stream, user terminal event stream, and real-time GPS stream of logistics vehicles are all given a unified timestamp, and the event streams with unified timestamps are stored in the event database. A positioning fusion algorithm based on reader topology and signal strength is executed on the book RFID reader stream in the event database to obtain the rough location of the book and its corresponding confidence level. The event streams of the book RFID reader, the mobile self-service machine status stream, and the user terminal event stream in the event library are deduplicated, and the deduplicated event streams are aggregated by a sliding time window to obtain aggregated events. The standardized event stream is obtained by standardizing the event stream of the event library, the rough position of the book and its corresponding confidence, and the aggregated event.

3. The method according to claim 1, characterized in that, Based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information, and external environmental characteristics, the demand distribution of each mobile self-service machine within a preset future time window is predicted according to book category, resulting in demand prediction results and their uncertainty indices, including: A spatiotemporal graph is constructed based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information, and external environmental features. The spatiotemporal graph includes mobile self-service machine nodes, geographical distance edges between mobile self-service machine nodes, and node attributes. The node attributes include the historical borrowing and returning records, user reservation information, external environmental features, and the location-level inventory snapshot table. The spatiotemporal graph is input into a pre-trained joint model to obtain the demand prediction results; the joint model consists of a spatial branch graph neural network and a temporal branch Transformer time series; the joint model predicts the distribution of demand book clusters for each of the mobile self-service machine nodes within a predefined time box; Uncertainty estimation is performed on the demand forecast results output by the joint model to obtain the uncertainty index corresponding to the demand forecast results.

4. The method according to claim 3, characterized in that, The supplementary strategies for generating each mobile self-service machine based on the demand forecast results, book metadata, and the location-level inventory snapshot table include: Cluster analysis is performed on the book metadata to obtain the available reserves and book reserve points for each book cluster; the book metadata includes, but is not limited to, ISBN, category, size, weight, and borrowing rate; Based on the demand forecast results, the location-level inventory snapshot table, and the available reserves, the shortage quantity of each book category at each mobile self-service machine is calculated, and the pre-allocation priority for each book category at each mobile self-service machine is calculated to obtain the replenishment strategy; the pre-allocation priority is obtained by weighted fusion of demand forecast results, alternativeity score, and current stockout risk; the replenishment strategy includes the target replenishment quantity, priority, and desired completion time limit for each book category at each mobile self-service machine.

5. The method according to claim 4, characterized in that, The scheduling plan, based on constraints such as logistics vehicle load, mobile self-service machine capacity, and time window, and minimizing a preset objective function according to the replenishment strategy, the self-service system master table, book reserve points, and location road network, includes: The supplementary strategy, the self-service system master table, book storage points, location road network, and preset objective function are input into a pre-established mixed-integer programming model to obtain a scheduling plan. The scheduling plan includes the transportation task sequence of logistics vehicles, estimated arrival time, book loading list, and the location points of the mobile self-service machines for each time period. The mixed-integer programming model solves for preset variables based on the capacity constraints and time window constraints of the mobile self-service machines. and ,in, Indicates logistics vehicles In the time window Is it a book storage point or a mobile self-service machine? Perform book loading docking. Indicates mobile self-service machine Location points on the location group network with location point The order relationship; In response to an operational emergency, an auction task is generated based on the scheduling plan corresponding to the operational emergency. The task is then redistributed through a bidding mechanism using each mobile self-service machine and each logistics vehicle as autonomous agents, resulting in an updated scheduling plan. In this bidding mechanism, the bid for each auction task is calculated based on the current location, current capacity, and expected profit of each mobile self-service machine, or the current location, current load, and expected transportation cost of each logistics vehicle. The operational emergency includes, but is not limited to, mobile self-service machine delays, logistics vehicle delays, reader / writer malfunctions, and sudden peak traffic.

6. The method according to claim 5, characterized in that, The method further includes: Obtain batch reading data of the loaded book list from the on-board RFID reader of each logistics vehicle, and obtain loading confirmation events; Based on the loading confirmation event, the corresponding ISBN in the book metadata will be updated to the loading status and the corresponding logistics vehicle identifier will be recorded. Acquire the reading data from the RFID readers of each mobile self-service machine, obtain the delivery confirmation event, and update the location-level inventory snapshot table; Based on the loading confirmation event and the delivery confirmation event, an abnormal work order that failed to confirm delivery is obtained, and a manual processing request is triggered for the abnormal work order. In response to the delivery confirmation event, a pickup notification and a one-time pickup code are pushed to the corresponding user terminal.

7. The method according to claim 6, characterized in that, The method further includes: Obtain actual operational data; the actual operational data includes actual delivery time, user pickup rate, the difference between demand forecast and actual demand, mobile self-service machine fault logs, and operating costs. Based on the actual operational data, the parameters of the joint model are fine-tuned to obtain the updated model parameters and estimated prediction confidence. Based on the actual operational data, the weighted weights corresponding to the pre-allocation priority and the weights corresponding to the objective function are calibrated to obtain the updated weight coefficients.

8. A self-service library book dispatching system, characterized in that, The system includes: The data module is used to stream the data from RFID readers, mobile self-service machines, user events, and real-time GPS data from logistics vehicles, and to preprocess these data to obtain a standardized event stream. The mobile self-service machine inventory module is used to perform location-level inventory estimation for each mobile self-service machine based on the self-service system master table and the standardized event flow, to obtain a location-level inventory snapshot table; the self-service system master table includes the identifier, latitude and longitude, cabinet capacity, reader topology and equipment status of each mobile self-service machine, as well as the identifier, capacity, load limit and speed parameters of each logistics vehicle. The demand forecasting module is used to forecast the demand distribution of each mobile self-service machine within a preset length time window based on the location-level inventory snapshot table, historical borrowing and returning records, user reservation information, and external environmental characteristics, according to book category clusters, and to obtain the demand forecasting results and their uncertainty index. The book replenishment module is used to generate replenishment strategies for each mobile self-service machine based on the demand forecast results, book metadata, and the location-level inventory snapshot table; the replenishment strategies include target replenishment quantity and replenishment priority by book category; The scheduling module is used to obtain a scheduling plan based on the load of logistics vehicles, the capacity of mobile self-service machines, and time window constraints, according to the replenishment strategy, the self-service system master table, book reserve points, and location road network minimizing a preset objective function. The objective function includes total delivery delay, mobile self-service machine idle cost, logistics vehicle transportation cost, and uncertainty penalty term. The scheduling plan includes the transportation task plan of logistics vehicles and the movement plan of mobile self-service machines.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images