A leasing platform scheduling method based on big data processing

Abstract

The application discloses a kind of based on big data processing's leasing platform scheduling method, including the following steps: S1, intercepts monomer equipment flow segment from electronic product leasing platform data and engages order to be dispatched, forms equipment scheduling chain;S2, split digital release segment, with networking review state calibration is released state column in disconnected network clearance state;S3, the CfC unit of release state column is accessed to improved closed continuous time neural network model;S4, using CUSUM algorithm accumulates reverse deviation, forms CUSUM algorithm recursion quantity;S5, CUSUM algorithm recursion quantity is embedded time gate, forms monomer equipment digital release hidden state;S6, through readout layer opening or intercept order to stop order to accept position;S7, gather order to accept position and monomer equipment flow segment, form equipment scheduling table.The application relates to the technical field of big data processing and intelligent scheduling, realizes digital release risk identification and reliable scheduling.

Description

Technical Field

[0001] This invention relates to the field of big data processing and intelligent scheduling technology, and in particular to a scheduling method for a rental platform based on big data processing. Background Technology

[0002] Existing leasing platform scheduling methods typically revolve around order management, inventory registration, and equipment status recording. After electronic products are returned, the platform usually determines whether the equipment can be re-entered into the rental queue based on whether the equipment has been returned to the warehouse, whether its appearance is abnormal, whether the rental period has ended, and whether the inventory is available. It then combines this with the time requirements of pending orders and equipment matching relationships to complete the next round of allocation. This type of method can meet the inventory turnover needs of general leasing businesses, but it lacks sufficient ability to identify the digital release status of returned electronic products.

[0003] With the development of electronic product rental services such as mobile phones, tablets, and laptops, the rentability status of equipment no longer depends solely on the physical return to storage. Even after the previous rental period ends, electronic products may still have issues such as account binding, incomplete data clearing, network verification backflow, and residual system configurations. Existing scheduling methods typically directly associate the return status with the rentability status, making it difficult to distinguish between devices that have completed physical return but whose digital release has not been completed.

[0004] In addition, existing platforms mostly rely on fixed rules or manual review to determine whether equipment can be rented again. It is difficult to continuously recursively analyze the reverse deviation between the offline clearing state and the online review state. This results in equipment with the risk of digital occupancy backflow being pre-assigned to pending orders, affecting the reliability of electronic product rental delivery.

[0005] Therefore, how to provide a rental platform scheduling method based on big data processing is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a rental platform scheduling method based on big data processing. This invention employs a method combining digital release state recognition with an improved closed-loop continuous-time neural network model. By constructing a device scheduling chain, the disconnected clearing state and the connected verification state after the electronic product is returned are calibrated into a release state column. A CUSUM algorithm recursive quantity is introduced into the CfC unit to continuously accumulate the reverse deviation between the two states. This allows the CUSUM algorithm recursive quantity to be embedded in time gating and guide the current hidden state to complete the closed state update, forming the digital release hidden state of a single device. Based on this, the order acceptance position is rewritten by the readout layer to prevent devices that are physically returned but whose digital release is not closed from being pre-allocated. This method has the advantages of high scheduling reliability, strong ability to identify digital occupancy backflow risks, and applicability to continuous rental scheduling of electronic products.

[0007] A rental platform scheduling method based on big data processing according to an embodiment of the present invention includes the following steps: S1. Extract individual equipment flow segments from the data of the electronic product rental platform and combine them with the orders to be scheduled to form an equipment scheduling chain; S2. Separate digital release segments within the device scheduling chain and calibrate the disconnection clear state and the network verification state to the same release state column; S3. The release state column is connected to the CfC unit of the improved closed continuous-time neural network model. The CfC unit calculates the first state candidate value and the second state candidate value respectively in the closed state update, and retains the time gating from the previous hidden state to the current hidden state. S4. The CUSUM algorithm is used to accumulate the reverse deviation between the disconnection clear state and the network verification state along the release state column in the CfC unit to form the CUSUM algorithm recursive quantity. Positions that do not generate reverse deviation continue the previous hidden state. S5 and CfC units embed the CUSUM algorithm recursion into time gating, guiding the current hidden state to perform closed state updates between the first state candidate value and the second state candidate value, forming a single device digital release hidden state. S6. The readout layer reads the digital release hidden state of the single device. When the current hidden state is biased towards the first state candidate value, the order acceptance position in the device scheduling chain is opened. When the current hidden state is biased towards the second state candidate value, the order acceptance position in the device scheduling chain is stopped. S7. Collect the order acceptance locations and individual equipment flow segments after the platform is opened to form the equipment scheduling table of the electronic product rental platform.

[0008] Optionally, S1 specifically includes: S11. Extract the equipment records that have been transferred and the order records that are waiting to be accepted from the data of the electronic product rental platform. Group the equipment records that have been transferred according to the individual equipment identity to obtain the individual equipment record group. S12. In the individual device record group, the time sequence of the device recycling event and the platform registration event is checked, and the continuous flow segments belonging to the same individual device are extracted to form individual device flow segments. S13. Extract order segments that have a connection with the individual equipment flow segments from the order records waiting to be accepted, and calibrate the order segments and individual equipment flow segments side by side according to the event time to form candidate connection segments. S14. Eliminate candidate junction segments with inconsistent individual device identities or reversed event times, and connect the remaining candidate junction segments in series according to individual device identities to form a device scheduling chain.

[0009] Optionally, S2 specifically includes: S21. Check the event boundary between the individual equipment flow segment and the order to be scheduled along the equipment scheduling chain, and classify the release event that is located after the equipment return position and affects the opening of the order acceptance position into the digital release segment. S22. Within the digital release segment, distinguish between the network disconnection clearing event and the network verification event, organize the release result generated by the network disconnection clearing event into the network disconnection clearing state, and organize the verification result generated by the network verification event into the network verification state. S23. Using the same individual device identity and the same release location as the calibration benchmark, register the network disconnection clear state and the network verification state in parallel at the same release location to form a release status column; S24. For release positions that lack a network disconnection clear state or a network verification state, retain a blank mark. For release positions with inverted event times, remove the digital release segment and output the release status column that corresponds to the network disconnection clear state and the network verification state.

[0010] Optionally, S3 specifically includes: S31, CfC unit locates the consecutive release positions of the same single device along the release status column, and writes the event time difference between adjacent consecutive release positions into the corresponding status update position; S32. Read the current release state at the state update position, and establish a closed state update relationship for the current hidden state by combining it with the previous hidden state. S33. Calculate the first state candidate value based on the closing trend of the current release state, calculate the second state candidate value based on the reinjection trend of the current release state, and keep the first state candidate value and the second state candidate value side by side in the same state update position. S34. Calculate time gating from event time difference. Time gating limits the proportion of the first state candidate value and the second state candidate value participating in the current hidden state update, forming the state update object corresponding to the release state column.

[0011] Optionally, S4 specifically includes: S41 and CfC units enter the state update position one by one in the order of the release state column. At each state update position, the current release state, the previous hidden state, and the CUSUM algorithm recursion value retained by the previous state update position are read simultaneously. S42. When the network disconnection clearing state and the network verification state in the current release state fall into the same state update position, and the network disconnection clearing state points to digital release closure and the network verification state turns to digital occupation backflow, the state difference of the same state update position is determined as the reverse deviation. When the network disconnection clearing state and the network verification state keep in the same direction of closure, the current state update position is merged into the reserved path of the previous hidden state. S43. The CUSUM algorithm recursive quantity is updated within the closed state update position of the CfC unit. When encountering a reverse deviation, it is incremented along the CUSUM algorithm recursive quantity retained in the previous state update position. When encountering a same-direction closure, it is reduced along the CUSUM algorithm recursive quantity retained in the previous state update position. When the reduction result crosses the zero value boundary, it is accepted as zero value. S44. When the current release state lacks the same result of the disconnection clearing state and the network verification state, the current state update position continues along the previous hidden state, and the CUSUM algorithm recursive quantity retained in the previous state update position is used as the CUSUM algorithm recursive quantity of the current state update position. S45. After the release state column has been processed position by position, the CUSUM algorithm recursive quantities retained at each state update position are arranged in the order of the release state column to form CUSUM algorithm recursive quantities that correspond one-to-one with the closed state update positions of the CfC unit.

[0012] Optionally, S5 specifically includes: S51, CfC unit enters the current state update position along the release state column. The time gating first determines the basic interpolation weight for the transition from the previous hidden state to the current hidden state based on the event time interval. S52. When obtaining the CUSUM algorithm recursive quantity at the current state update position, convert the CUSUM algorithm recursive quantity into the reverse traction share, and deduct the reverse traction share from the basic interpolation weight corresponding to the first state candidate value. S53. The remaining share obtained after deduction is allocated to the first state candidate value, and the reverse traction share is allocated to the second state candidate value. When the reverse traction share exceeds the basic interpolation weight, it is deducted according to the boundary value of the basic interpolation weight. S54. The current hidden state first retains the continuous items of the previous hidden state along the event time interval, then absorbs the first state candidate value according to the remaining share after allocation, and absorbs the second state candidate value according to the reverse traction share, thus completing the closed state update of the current state update position. S55. When the current state update position has not obtained the CUSUM algorithm recursion, the time gating retains the basic interpolation weight, and the current hidden state completes the closed state update along the previous hidden state and the first state candidate value. S56 and CfC units continuously update the closed state at each state update position along the release state column, and connect the current hidden states formed by each state update position in the order of the release state column to form the digital release hidden state of a single device.

[0013] Optionally, S6 specifically includes: S61. The readout layer locates the order acceptance position corresponding to the hidden state of the digital release of a single device along the device scheduling chain, and reads the current hidden state formed by the corresponding state update position. S62. Extract the first state candidate value absorption share and the second state candidate value absorption share from the current hidden state. The first state candidate value absorption share takes over the digital release closing direction, and the second state candidate value absorption share takes over the digital occupation backflow direction. S63. When the first state candidate value absorbs the dominant share of the current hidden state, the order acceptance position is rewritten to the open state, and the acceptance relationship between the order acceptance position and the individual equipment flow segment is retained. S64. When the absorption share of the second state candidate value occupies the dominant share of the current hidden state, the order acceptance position is rewritten to the stop state, and the original connection relationship between the order acceptance position and the single device flow segment is maintained. S65. When the current hidden state cannot extract the first state candidate value absorption share or the second state candidate value absorption share, the order acceptance position continues along the state corresponding to the previous hidden state, and no new open state is added. S66. Complete the status rewriting of the order acceptance position according to the equipment scheduling chain sequence to form the scheduling judgment result of the order acceptance position.

[0014] Optionally, S7 specifically includes: S71. Read the scheduling judgment result of the order acceptance position along the equipment scheduling chain, keep the order acceptance position corresponding to the open state connected with the single equipment flow segment, and isolate the order acceptance position corresponding to the stopped state from the equipment scheduling chain. S72. For orders that maintain connection, the receiving position is assigned to the corresponding individual equipment flow segment according to the individual equipment identity, forming an individual equipment receiving record; S73. When the same single equipment circulation segment corresponds to multiple order acceptance positions, retain the order acceptance position that is ranked first in the equipment scheduling chain, and mark the remaining order acceptance positions as unaccepted positions. S74. Arrange the individual equipment acceptance records in the order of the equipment scheduling chain to form the equipment scheduling table of the electronic product rental platform.

[0015] The beneficial effects of this invention are: This invention, by combining big data processing from an electronic product rental platform with an improved closed-loop continuous-time neural network model, can distinguish between the physical return status and digital release status of electronic products before equipment scheduling. By constructing an equipment scheduling chain and calibrating the offline clearing state and the online verification state into the same release status column, this invention can identify situations where there is still a risk of digital occupancy backflow after equipment is returned, avoiding the misscheduling problems caused by traditional rental platforms directly allocating equipment based solely on available inventory or physical return results.

[0016] Furthermore, this invention introduces a CUSUM algorithm recursive quantity within the CfC unit, enabling the reverse deviation between the offline clearing state and the online verification state to be continuously accumulated during the closed-loop state update process. This is further reinforced by time gating, guiding the current hidden state to complete the closed-loop state update between the first and second state candidate values. Through this processing, the model can continuously identify digital occupancy risks such as account binding residue, online verification backfeedback, and unreleased system configurations, rather than relying on fixed rules or manual verification for a one-time judgment.

[0017] This invention also transforms the hidden digital release state of individual devices into the open or closed result of the order acceptance position through the readout layer, preventing electronic products that are physically returned to the warehouse but whose digital release is not closed from prematurely entering the schedulable connection. This improves the accuracy of determining the rentability status of electronic products, enhances the rental platform's ability to intercept abnormal digital occupancy in continuous rental scenarios, and improves the reliability of equipment allocation and delivery stability. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a rental platform scheduling method based on big data processing proposed in this invention; Figure 2 This is a schematic diagram of the structure of the improved closed-loop continuous-time neural network model in the rental platform scheduling method based on big data processing proposed in this invention; Figure 3 This is a schematic diagram of the closed-loop state update of the CUSUM algorithm recursive quantity embedded with time gating in a rental platform scheduling method based on big data processing proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-3 A rental platform scheduling method based on big data processing includes the following steps: S1. Extract individual equipment flow segments from the data of the electronic product rental platform and combine them with the orders to be scheduled to form an equipment scheduling chain; S2. Separate digital release segments within the device scheduling chain and calibrate the disconnection clear state and the network verification state to the same release state column; S3. The release state column is connected to the CfC unit of the improved closed continuous-time neural network model. The CfC unit calculates the first state candidate value and the second state candidate value respectively in the closed state update, and retains the time gating from the previous hidden state to the current hidden state. S4. The CUSUM algorithm is used to accumulate the reverse deviation between the disconnection clear state and the network verification state along the release state column in the CfC unit to form the CUSUM algorithm recursive quantity. Positions that do not generate reverse deviation continue the previous hidden state. S5 and CfC units embed the CUSUM algorithm recursion into time gating, guiding the current hidden state to perform closed state updates between the first state candidate value and the second state candidate value, forming a single device digital release hidden state. S6. The readout layer reads the digital release hidden state of the single device. When the current hidden state is biased towards the first state candidate value, the order acceptance position in the device scheduling chain is opened. When the current hidden state is biased towards the second state candidate value, the order acceptance position in the device scheduling chain is stopped. S7. Collect the order acceptance locations and individual equipment flow segments after the platform is opened to form the equipment scheduling table of the electronic product rental platform.

[0021] In this embodiment, S1 specifically includes: The electronic product leasing platform retrieves data from the order database, equipment asset ledger, and warehouse registration records. The unique equipment code in the equipment asset ledger serves as the identity of each individual device. Completed transfer records refer to continuous business records left after the same individual device completes its outbound, lease, return, and return-to-warehouse registration. Waiting-to-be-accepted order records refer to leasing orders that have not yet been assigned to specific individual devices. The system first groups completed transfer records using unique equipment codes; records under the same unique equipment code are grouped into the same individual device record group. Records lacking a unique equipment code, with a unique equipment code inconsistent with the equipment asset ledger, or duplicate transfer records occurring within the same time period are written to the conflict record area and not included in the equipment scheduling chain construction.

[0022] After a single-unit device record group is formed, the system verifies the temporal relationship between device recycling events and platform registration events within the record group. A device recycling event is the event where the platform confirms the return of the electronic product to the warehouse node, and a platform registration event is the event where the warehouse system confirms the completion of the electronic product's inbound registration. When the device recycling event time is earlier than or equal to the platform registration event time, the system extracts continuous records from the end position of the previous order to the platform registration position, forming a single-unit device flow segment. When the platform registration event time is earlier than the device recycling event time, it is determined as an event time reversal, the corresponding record is written to the abnormal record area, and the connection between the corresponding record and the single-unit device record group is disconnected. The determination of continuous ownership of the same single-unit device is based on consistent unique device codes, non-reversed flow times, and the ability of the device flow status to transition from the end of the previous order to platform registration.

[0023] Once a single-device workflow segment is formed, the system extracts order segments that can be combined with it from the pending order records. An order segment must at least include the required equipment type, rental start time, and pending allocation status. If the equipment type matches the equipment type corresponding to the single-device workflow segment, and the rental start time is no earlier than the platform registration time, the order segment has a binding relationship. The system aligns the order segments with binding relationships with the single-device workflow segments according to event time, forming candidate binding segments. Candidate binding segments with inconsistent equipment types, rental start times earlier than the platform registration time, order statuses that have already been allocated, or single-device workflow segments lacking platform registration events are directly eliminated.

[0024] After candidate junction segments are screened out, the system concatenates the remaining candidate junction segments according to the individual device identity. During concatenation, the individual device flow segment is at the front end, and the order segment waiting to be accepted is at the back end, preserving the event time sequence and acceptance relationship between the two. When the same individual device flow segment corresponds to multiple order segments, the system arranges the multiple order segments according to the rental start time and retains multiple candidate acceptance relationships; when multiple individual device flow segments correspond to the same order segment, the system arranges multiple candidate junction segments according to the proximity of the platform registration time and rental start time, without deleting candidate junction relationships during the device scheduling chain construction stage. After the above processing, the individual device flow segment and the order to be scheduled form a continuous and traceable device scheduling chain, which serves as the processing object for the separation of digital release segments.

[0025] In this embodiment, S2 specifically includes: The equipment scheduling chain, as the processing object, already includes individual equipment flow segments and orders awaiting scheduling. The system first reads the event interval between the equipment return location and the order acceptance location. The equipment return location is the equipment scheduling chain position after registration on the platform, and the order acceptance location is the position where the order awaiting scheduling is added to the equipment scheduling chain. Digital release events located within the event interval between the equipment return location and the order acceptance location that affect whether the order acceptance location is open are classified into digital release segments. Digital release events originate from equipment clearing, equipment verification, equipment management status release, and system configuration recovery records. When the event occurs before the equipment return location, the system retains the event within the individual equipment flow segment; when the event occurs after the order acceptance location, the system retains the event on the side of the order awaiting scheduling and does not enter the digital release segment.

[0026] After the digital release fragment is formed, the system distinguishes between network disconnection clearing events and network verification events based on the event source and execution environment. Network disconnection clearing events refer to the results of data clearing, account logout, and configuration cleanup performed when the device is not connected to an external network; network verification events refer to the results obtained after the device is connected to the platform for network verification and the clearing results are re-verified. Each result is uniformly converted into a release state: a closed release is recorded as 1, an open release as 0, and an event parsing failure as a blank marker. The converted state of a network disconnection clearing event is written as the network disconnection clearing state, and the converted state of a network verification event is written as the network verification state. The original event record is not changed after the state conversion; the original event record is retained in the digital release fragment for verification.

[0027] Release positions are used to carry paired states of the same individual device on the same digital release object. The system establishes a calibration benchmark using the individual device identity and the digital release object. The identity of the same individual device comes from the unique device code in the device scheduling chain, and the digital release object comes from the processing object name in the digital release segment. When the network disconnection clearing state and the network verification state belong to the same individual device identity and have the same digital release object, they are registered to the same release position. The event times of the network disconnection clearing state and the network verification state are converted to a unified time format. The time comparison accuracy adopts the recording accuracy of the platform business system. When the recording accuracy is inconsistent, the lower accuracy is used as the comparison accuracy. After registration is completed, each release position saves a set of network disconnection clearing states and network verification states, and arranges them in order of event time to form a release status column.

[0028] Release positions lacking either the network disconnection clearing state or the network verification state are not deleted; instead, a void marker is written to the missing side, which serves as the default processing object in state updates. If the event time of the network verification state is earlier than that of the corresponding network disconnection clearing state, it is considered an event time reversal, and the corresponding release position is removed from the digital release segment and written to the exception record area; the original device scheduling chain connection remains unchanged. After the above processing, the network disconnection clearing state and the network verification state form a corresponding release state column, which serves as the state update object for the improved closed continuous-time neural network model.

[0029] In this embodiment, S3 specifically includes: The standard closed-loop continuous-time neural network model is used to process sequential data with irregular time intervals, and its core structure is the CfC unit. At each time point, the CfC unit reads the current input, the previous hidden state, and the adjacent time interval. Through input mapping, hidden-state mapping, state candidate value calculation, time gating calculation, and closed-loop state update, the current hidden state is obtained. Input mapping transforms the current input to the hidden-state dimension, hidden-state mapping transforms the previous hidden state to the same dimension, state candidate value calculation determines the update direction of the current state, time gating determines the transition ratio from the previous hidden state to the current hidden state, and closed-loop state update completes the hidden state recursion without calling the numerical differential equation solver. While the standard CfC unit is suitable for describing continuous processes, in electronic product rental scheduling, the disconnected clearing state and the connected verification state may have opposite release directions. When the standard CfC unit recurses along a single state candidate value, it easily treats the connected verification feedback as ordinary time fluctuations, making it difficult to distinguish between the digital release closing direction and the digital occupancy feedback direction.

[0030] The improved closed-loop continuous-time neural network model consists of CfC units and a readout layer. The CfC units handle the continuous recursion of the release state sequence, while the readout layer handles the conversion of the hidden digital release state of a single device into the result of opening or closing the order acceptance position. The improvement is located inside the CfC units, retaining the previous hidden state, time gating, and closed-loop state update structure of the ordinary CfC unit. The calculation of a single state candidate value is changed to the parallel calculation of the first and second state candidate values. The first state candidate value carries the digital release closing direction, and the second state candidate value carries the digital occupancy backflow direction. The same state update position simultaneously retains the first state candidate value, the second state candidate value, and the time gating, providing an internal calculation location for the CUSUM algorithm's recursive quantity to participate in the closed-loop state update.

[0031] The improved closed-loop continuous-time neural network model was trained and calibrated using historical return verification samples before deployment. The training samples consisted of release state columns, event time differences, manual verification confirmation results, and order acceptance results. Learnable parameters included first input weights, second input weights, first hidden weights, second hidden weights, first bias, second bias, gating mapping weights, gating bias, readout layer weights, and readout layer biases, all updated via backpropagation. Engineering decision thresholds included release closure state transition thresholds, time difference upper limits, gap mask retention rules, and abnormal record writing rules. These engineering decision thresholds were calibrated based on statistical analysis of historical verification records on the platform and were not considered learnable parameters. Training stopped when the validation set loss decreased below a convergence threshold across consecutive rounds. The convergence threshold was determined by the training sample size and validation set fluctuation range. Before deployment, manual verification samples were used to calibrate the time difference upper limit and release closure state transition thresholds.

[0032] When the release status column is connected to the CfC unit, each release position is converted into a fixed-length state vector. The length of the state vector is consistent with the number of digital release objects that the platform needs to verify. Each dimension corresponds to one digital release object. The offline clear state and the online verification state are written to two state components of the same dimension, respectively. A closed release is written to the positive component, an open release is written to the reverse component, and a parsing failure or missing record is written to zero along with a missing mask. The event time difference between adjacent release positions is uniformly converted to minutes. When the time difference is less than zero, the corresponding release position is written to the abnormal record area. When the time difference is missing, it is supplemented by the median value of the time difference between adjacent valid release positions of the same single device. The missing mask is retained in the state update object.

[0033] The CfC unit reads the current release state, the previous hidden state, and the event time difference at each state update position. The current release state is first multiplied by the first input weight, and the previous hidden state is multiplied by the first hidden weight. The two products are accumulated dimension-wise and then superimposed with the first bias. This is then subjected to hyperbolic tangent compression and limited to between -1 and +1 to obtain the first state candidate value. The current release state is then multiplied by the second input weight, and the previous hidden state is multiplied by the second hidden weight. The two products are accumulated dimension-wise and then superimposed with the second bias. This is then subjected to hyperbolic tangent compression and limited to between -1 and +1 to obtain the second state candidate value. The dimension of the current release state is the number of released objects. The dimensions of the previous hidden state, the first state candidate value, and the second state candidate value are all hidden state dimensions. The number of rows in the first and second input weights corresponds to the hidden state dimension, and the number of columns corresponds to the current release state dimension. The rows and columns of the first and second hidden weights also correspond to the hidden state dimensions. Dimensions where a gap mask is hit are not written into new candidate changes; the corresponding dimension is retained along the previous hidden state.

[0034] Temporal gating within the CfC unit is calculated using the event time difference and the previous hidden state. The event time difference is first normalized by dividing by the median event time difference in the training samples. If the normalized result is less than zero, it is truncated to zero; if the result is greater than the upper limit of the time difference, it is retained according to the upper limit. The normalized time difference is concatenated with the previous hidden state and then entered into a gating map. The gating map result is superimposed with a gating bias and then compressed using a sigmoid function to obtain the temporal gating. The temporal gating value is limited to between zero and one. When the temporal gating is close to zero, the current hidden state retains more of the previous hidden state; when the temporal gating is close to one, the current hidden state absorbs more state candidate values. After the above processing, each release position forms a state update object containing the current release state, the event time difference, the previous hidden state, the first state candidate value, the second state candidate value, and the temporal gating.

[0035] Compared to the ordinary closed-loop continuous-time neural network model, the improvement of this implementation lies in the change of the calculation path of the state candidate value within the CfC unit. An ordinary CfC unit only forms a single state candidate value, and the hidden state can only perform closed-loop state updates along one candidate direction. In the improved implementation, the CfC unit forms the first and second state candidate values ​​in parallel at the same state update position, separating the digital release closure direction from the digital occupancy backfeed direction within the model. With this processing, the release state sequence no longer enters the single-path recursion as ordinary input, but instead forms a state update object that can be continuously guided by the CUSUM algorithm's recursion. This facilitates the time gating in distinguishing between normal release changes and network verification backfeed changes during closed-loop state updates, reducing the possibility of digital occupancy backfeed being treated as ordinary time fluctuations.

[0036] In this embodiment, S4 specifically includes: The CUSUM algorithm recursive value is used within the CfC unit as a recursive value synchronously stored with the closed-state update position. In this implementation, the CUSUM algorithm is used to continuously accumulate the reverse deviation between the disconnection clearing state and the network verification state. The recursive process does not output independent anomaly labels, but instead inherits the CUSUM algorithm recursive value from the previous state update position at each state update position, and performs incrementing, waning, and limiting processing according to the current reverse deviation. After the release state column enters the CfC unit, each state update position corresponds to a release position of the same single device. The current release state carries the corresponding results of the disconnection clearing state and the network verification state. The disconnection clearing state is the clearing result obtained when the device is not connected to an external network, and the network verification state is the verification result obtained after the device is connected to the platform for network verification. Digital release closure is marked as 1, digital release not closed is marked as 0, and missing results are marked as empty. The previous hidden state is the hidden vector retained after the previous state update position completes the closed-state update, and the CUSUM algorithm recursive value retained at the previous state update position serves as the recursive starting point for the current state update position.

[0037] The reverse deviation in the current release state is calculated position by position within the CfC unit. When the network disconnection clear state is 1 and the network verification state is 0, the device displays as released during network disconnection clearing, but digital occupancy backflow occurs after network verification, and the reverse deviation is written to the current state update position. When both the network disconnection clearing state and the network verification state are 1, the clearing result and the verification result are closed in the same direction, and no new reverse deviation is written to the current state update position; it continues along the retained path of the previous hidden state. When both the network disconnection clearing state and the network verification state are 0, the digital release is not closed, and the current state update position is retained as an unclosed release state, not treated as a backflow deviation. When the network disconnection clearing state is 0 and the network verification state is 1, the network verification corrects the network disconnection clearing result, and the current state update position is processed along the closed path in the same direction.

[0038] The CUSUM algorithm recursive value is updated synchronously with the current reverse deviation within the closed state update position. The dimension of the CUSUM algorithm recursive value is consistent with the dimension of the digital release object in the current release state, and each dimension stores the cumulative deviation value corresponding to a digital release object. When there is a reverse deviation at the current state update position, the CUSUM algorithm recursive value retained at the previous state update position is added to the current reverse deviation, and then the tolerance drift amount is subtracted to obtain the recursive candidate value for the current state update position; if the recursive candidate value is lower than 0, it is retained as 0, and if it is higher than the recursive value upper limit, it is retained as the recursive value upper limit. When the current state update position is closed in the same direction, the CUSUM algorithm recursive value retained at the previous state update position is subtracted from the fallback step size, and if the value is lower than 0 after subtraction, it is retained as 0. The tolerance drift amount, fallback step size, and recursive value upper limit are engineering judgment thresholds, which are statistically calibrated by the platform's historical review records; in one implementation, the recursive value upper limit is 1, the tolerance drift amount is 0.05, and the fallback step size is 0.10, and the platform can recalibrate the values ​​using manual review samples.

[0039] When the current release state lacks a corresponding result from a disconnected clearing state or a connected verification state, the CfC unit does not write a new reverse bias at the current state update position. The dimension corresponding to the missing result is retained along the previous hidden state, and the CUSUM algorithm recursive value retained at the previous state update position is used as the CUSUM algorithm recursive value at the current state update position. If the event time is reversed at the same release position, and the reversed release position has already been moved out during the release state column formation stage, the CfC unit only receives the release state column that has completed calibration; if the missing markers appear consecutively, the CUSUM algorithm recursive value maintains the value of the previous valid state update position and enters the current state update position along with the previous hidden state until a new corresponding result appears, after which the incrementing or decrementing calculation resumes.

[0040] The recursive record of the CUSUM algorithm recursive quantity is stored corresponding to the state update position of the CfC unit. Each state update position retains the current release state, the reference to the previous hidden state, the current reverse bias, and the current CUSUM algorithm recursive quantity. After the current state update position completes its recursion, the current CUSUM algorithm recursive quantity becomes the previous recursive quantity for the next state update position, and the current hidden state continues to participate in the closed-loop state update of the next position. After all the release state sequences have been processed, the CUSUM algorithm recursive quantities retained at each state update position are arranged in the order of the release state sequences, forming the CUSUM algorithm recursive quantities corresponding to the closed-loop state update positions of the CfC unit, and are then incorporated into the time-gated rewriting process.

[0041] Compared to the ordinary closed-loop continuous-time neural network model, the improvement of this implementation lies in the addition of a CUSUM algorithm recursive quantity within the CfC unit, which is synchronously recursively applied to the closed-loop state update position. Ordinary CfC units primarily rely on event time intervals and the previous hidden state to control the continuous change of the hidden state, failing to continuously retain the reverse change between the disconnection / clearing state and the network verification state within the model. In the improved implementation, the reverse deviation is not output as an external abnormal result, but rather accumulates, falls back, is limited, and saved within each state update position of the CfC unit, allowing digital occupancy backfeeding to form a continuous traction quantity during the hidden state recursion. Through this processing, the time gating can read the accumulated degree of the reverse deviation during closed-loop state updates, preventing network verification backfeeding from being overwritten by ordinary time fluctuations.

[0042] In this embodiment, S5 specifically includes: Within the CfC unit, time gating controls the transition ratio from the previous hidden state to the current hidden state. When the current state update position enters a closed-loop state update, the event time interval is first normalized by the time scale. The normalized result is concatenated with the previous hidden state and then entered into the gating mapping. The gating mapping is obtained by multiplying the gating weights with the concatenated vector, then adding the gating bias, and finally compressing it with Sigmoid to obtain the basic interpolation weights. The basic interpolation weights are hidden state dimension vectors, with values ​​limited to 0 to 1. Components exceeding the boundary are retained as 0 or 1. The previous hidden state, the first state candidate value, and the second state candidate value are all hidden state dimension vectors, with values ​​ranging from -1 to +1 after hyperbolic tangent compression. Components exceeding the boundary are retained as -1 or +1.

[0043] When embedding the CUSUM algorithm recursive quantity into the time-gated system, the CUSUM algorithm recursive quantity corresponding to the release state column is first read from the current state update position. The dimension of the CUSUM algorithm recursive quantity is consistent with the number of digital release objects. If the number of digital release objects is different from the hidden state dimension, it is first transformed to the hidden state dimension through traction mapping. The traction mapping is obtained by multiplying the traction weight and the CUSUM algorithm recursive quantity, then adding the traction bias, and finally compressing it with Sigmoid to obtain the reverse traction share. The traction weight and traction bias are learnable parameters, obtained through training with historical return verification samples. Each dimension of the reverse traction share corresponds to the reinjection traction strength in the same dimension of the current hidden state.

[0044] After the reverse traction share is generated, the CfC unit performs dimension-by-dimensional share deduction within the time gating. For each hidden state dimension, the basic interpolation weight is read first, followed by the reverse traction share of the same dimension. If the reverse traction share is less than or equal to the basic interpolation weight, it is directly deducted from the basic interpolation weight. If the reverse traction share is greater than the basic interpolation weight, it is only deducted up to the boundary of the basic interpolation weight to avoid negative participation shares of the first state candidate value. The share remaining after deduction is allocated to the first state candidate value, and the deducted share is allocated to the second state candidate value. The two candidate values ​​jointly participate in the closed state update of the current hidden state at the same state update position.

[0045] The closed-loop update of the current hidden state is executed after the share allocation is completed. The previous hidden state is first multiplied by the reserved share, which is obtained by subtracting the basic interpolation weight from one, and is used to accommodate continuous changes caused by the event time interval. The first state candidate value is multiplied by the remaining share after deduction, which is obtained by subtracting the reverse traction share from the basic interpolation weight, and is used to accommodate the digital release closing direction. The second state candidate value is multiplied by the reverse traction share, and is used to accommodate the digital occupation backflow direction. The three product results are accumulated dimension by dimension to form the current hidden state. When the accumulated result exceeds the range of hidden state values, it is limited by a negative one or a positive one. Through the above calculation, the CUSUM algorithm recursion is not used as an external judgment result, but rather changes the actual participation ratio of the first state candidate value and the second state candidate value within the time gating.

[0046] When the current state update position has not obtained the CUSUM algorithm recursion, the CfC unit does not generate a reverse traction share, and the time gating retains the basic interpolation weight formed by the event time interval. At this time, the second state candidate value does not obtain an enhanced participation share, and the current hidden state completes the closed state update by the previous hidden state and the first state candidate value; if the current release state has a missing mask, the missing dimension is retained along the previous hidden state, and the non-missing dimension continues to participate in the time gating update. After the current state update position is completed, the current hidden state is written to the corresponding position in the release state column as the previous hidden state for the next state update position.

[0047] After all the release status columns have been processed, the CfC unit continuously retains the current hidden states formed by each status update position in the order of the release status columns, thus obtaining the digital release hidden state of the individual device. The digital release hidden state of the individual device not only includes the continuous changes caused by the event time interval, but also the changes in the participation ratio of candidate values ​​caused by the CUSUM algorithm recursion within the time gating, which can accommodate the readout layer's determination of whether the order acceptance position is open or closed.

[0048] Compared to the ordinary closed-loop continuous-time neural network model, the improvement of this implementation lies in the rewriting of the interpolation path of time gating. In ordinary CfC units, time gating mainly controls the fusion ratio between the candidate state value and the previous hidden state by the event time interval. The inverse deviation between the disconnection clearing state and the network verification state cannot directly change the fusion direction of the candidate values. This implementation converts the CUSUM algorithm recursive quantity into a reverse traction share, and within the time gating, deducts the corresponding share from the basic interpolation weight of the first state candidate value before transferring it to the second state candidate value. Through this processing, network verification backfeedback is no longer treated as a smooth absorption of ordinary time fluctuations, but instead directly tractions the current hidden state towards the digital occupancy backfeedback direction during closed-loop state updates, thereby improving the ability of a single device's digital release hidden state to distinguish backfeedback risks.

[0049] In this embodiment, S6 specifically includes: The readout layer takes over the hidden state of a single device's digital release. This hidden state originates from the last valid state update position of the CfC unit. The order acceptance position of the same single device in the device scheduling chain is determined by the single device's identity and event sequence. The readout layer first reads the current hidden state, and simultaneously reads the first and second candidate states, their participation ratios, and the share retained from the previous hidden state at the same state update position. The participation ratio of candidate states comes from the share allocation results within the time gating; the first candidate state corresponds to the remaining share after deduction, and the second candidate state corresponds to the reverse-pull share. The participation ratio of candidate states is retained by the time gating after share deduction and allocation at the same state update position. The readout layer does not regenerate the participation ratio of candidate states; it only reads the share record at the same state update position.

[0050] The absorption share of the first state candidate value is obtained in the readout layer from the components of the first state candidate value in the current hidden state. Specifically, the readout layer multiplies the first state candidate value dimensionally by the subtracted remaining share to obtain the first candidate absorption vector; the second state candidate value is multiplied dimensionally by the reverse pulling share to obtain the second candidate absorption vector. The two absorption vectors are added to the consecutive terms corresponding to the retained shares of the previous hidden state to obtain the current hidden state. Therefore, the absorption shares of the first and second state candidate values ​​can be separated from the composition of the current hidden state. Dimensions where the missing mask is hit are not included in the absorption share comparison; the missing dimension is retained along the component corresponding to the previous hidden state.

[0051] The dominant share is determined by dimension-wise accumulation within the readout layer. The readout layer first takes non-negative values ​​for the dimensions of the first candidate absorption vector that are not hit by the empty mask and accumulates them to obtain the first absorption total; then it takes non-negative values ​​for the dimensions of the second candidate absorption vector that are not hit by the empty mask and accumulates them to obtain the second absorption total. When the first absorption total is greater than the second absorption total and reaches the open threshold, the absorption share of the first state candidate value occupies the dominant share of the current hidden state; when the second absorption total is greater than the first absorption total and reaches the stop threshold, the absorption share of the second state candidate value occupies the dominant share of the current hidden state. The open threshold and stop threshold range from 0 to 1 and are statistically calibrated based on historical return review records and manual confirmation results; in one implementation, both the open threshold and the stop threshold are set to 0.5, and the platform can recalibrate them based on the manual confirmation results.

[0052] The open state of an order acceptance position is driven by the absorption share of the first candidate state value. The read-out layer locates the order acceptance position corresponding to the current individual device in the device scheduling chain. When the first absorption volume dominates, the order acceptance position is rewritten to an open state, while preserving the acceptance relationship between the order acceptance position and the individual device's flow segment. When preserving the acceptance relationship, the original individual device identity, platform registration time, and the association relationship with the orders to be scheduled in the device scheduling chain remain unchanged; the open state only changes whether the order acceptance position can participate in the device scheduling table aggregation.

[0053] The stoppage state of an order acceptance location is driven by the absorption share of the second state candidate value. When the second absorption total accounts for the dominant share, the readout layer rewrites the order acceptance location to the stoppage state while maintaining the original connection relationship between the order acceptance location and the individual device flow segment. The stoppage state does not delete the original connection relationship, which continues to be used to trace the source of digital occupancy backflow; the stoppage state only prevents the order acceptance location from entering the open order acceptance set, ensuring that the device scheduling table only aggregates available acceptance devices.

[0054] When the current hidden state cannot extract the absorption share of the first or second state candidate value, the readout layer performs default processing. When the participation ratio of candidate values ​​is missing, the dimension of the current hidden state is inconsistent with the dimension of the state candidate value, or all effective dimensions are hit by the empty mask, the order acceptance position continues from the state rewriting result of the previous order acceptance position; if the previous order acceptance position is an open state, the open state is retained; if the previous order acceptance position is a truncated state, the truncated state is retained; if the state rewriting result of the previous order acceptance position is missing, no new open state is added.

[0055] The read-out layer rewrites the status of each order acceptance location according to the device scheduling chain. Each order acceptance location retains only one of the following states: open or closed. After the status rewriting is completed, a scheduling decision result for the order acceptance location is formed. The scheduling decision result continues to retain the individual device identity, order acceptance location, status rewriting result, and original connection relationship, serving as the direct processing object for forming the device scheduling table of the electronic product rental platform.

[0056] In this embodiment, S7 specifically includes: The equipment scheduling table formation process considers the scheduling determination results of the order acceptance positions. These results are derived from the state rewriting results completed by the read layer. An open state indicates that the order acceptance position is allowed to participate in the equipment scheduling table aggregation, while a truncated state indicates that the order acceptance position retains a traceable connection but does not enter a schedulable connection. The system reads each order acceptance position sequentially along the equipment scheduling chain. When encountering an open state, it maintains the connection between the order acceptance position and the individual equipment flow segment. When encountering a truncated state, it isolates the order acceptance position from the schedulable connection. The isolated order acceptance position is not deleted but retained in the truncated record of the equipment scheduling chain for tracing the reasons for incomplete digital release closures.

[0057] Individual device acceptance records are formed by assigning the order acceptance position, which maintains the connection, to the corresponding individual device flow segment. The assignment rule uses the consistency of the individual device's identity as the criterion; the individual device's identity comes from the unique device code in the device scheduling chain. When the unique device code carried by the order acceptance position matches the unique device code in the individual device flow segment, the order acceptance position is written into the corresponding individual device acceptance record; if the unique device codes do not match, the order acceptance position is not written into the individual device acceptance record and is transferred to an unaccepted position. Unaccepted positions retain the order source, original connection relationship, and reason for not accepting the order from the order acceptance position to prevent the order acceptance position from being lost when the device scheduling table is formed.

[0058] When multiple order acceptance positions correspond to the same single device's workflow segment, the system handles conflicts based on the order of the order acceptance position in the device scheduling chain. The order of the order acceptance position in the device scheduling chain is determined by its position within the chain; order acceptance positions that appear earlier are retained first. Order acceptance positions that appear earlier are written into the single device's acceptance record, while those that appear later are marked as unaccepted positions. When multiple order acceptance positions have the same position, the system distinguishes them by the registration time of the order entering the device scheduling chain; order acceptance positions with earlier registration times are retained, while those with missing registration times are placed as unaccepted positions.

[0059] The equipment scheduling table is formed by arranging individual equipment acceptance records in the order of the equipment scheduling chain. Each individual equipment acceptance record stores the individual equipment identity, the individual equipment flow segment, and the order acceptance position in the open state. Order acceptance positions in the stopped state are retained in the stopped list, and unaccepted positions are retained in the pending reallocation list. The stopped list and the pending reallocation list are stored as traceability fields in the equipment scheduling table, not as schedulable connections. When the equipment scheduling table is updated, order acceptance positions in the open state are added to the schedulable list, order acceptance positions in the stopped state are used to trace the reasons for the incomplete digital release, and unaccepted positions are used to rematch pending orders. After processing, the equipment scheduling table retains executable scheduling relationships and provides a clear processing object for the platform to reallocate pending orders.

[0060] Example 1: To verify the feasibility of this invention in practice, it was applied to the return verification and order scheduling process of an electronic product rental platform, which primarily rents out mobile phones, tablets, and laptops. The platform's original scheduling method relied on equipment return to the warehouse, visual inspection, and inventory availability as the main criteria. Once equipment completed its return registration, it would quickly enter the rental queue. However, in actual operation, it was found that some equipment, even after being physically returned to the warehouse, still encountered problems such as account binding prompts, application authorization conflicts, system configuration reverting, or unreleased equipment management status after online verification. The original system required manual verification again, which could easily lead to equipment being prematurely allocated to the next order, affecting the stability of delivery preparation and rental delivery.

[0061] When applying this invention, the platform first extracts individual device flow segments from the rental data and combines them with pending orders to form a device scheduling chain. After the device is returned, the system does not directly mark the device as rentable. Instead, it organizes the digital release process after return into a release status column, placing the disconnection clearing state and the network verification state in the same release position for calibration. After the release status column is connected to an improved closed-loop continuous-time neural network model, the CfC unit calculates the first and second state candidate values ​​in the closed-loop state update, and accumulates the reverse deviation between the disconnection clearing state and the network verification state through the CUSUM algorithm recursion, so that the current hidden state can reflect whether there is a trend of digital occupancy backflow of the device.

[0062] During the scheduling process, the read layer reads the hidden state of digital release for individual devices and rewrites the order acceptance position in the device scheduling chain to an open or closed state. For devices with closed digital releases, the system retains the corresponding order acceptance position and enters it into the device scheduling table; for devices with a risk of backflow, the system closes the order acceptance position to prevent devices that are physically returned to the warehouse but whose digital releases are not closed from being pre-allocated. After internal platform review, this invention can reduce manual secondary confirmation, improve the stability of judging the rentability status of electronic products, and reduce delivery anomalies caused by unreleased digital occupancy.

[0063] To ensure data authenticity, 1800 electronic products that completed return verification within the platform's continuous operation cycle were selected as test samples. Among them, 236 devices were manually verified to have issues with digital occupancy backflow or unclosed digital release, while 1564 devices had no digital release anomalies. The method of this invention was compared with the platform's original physical return scheduling method and fixed-rule digital release verification method. The statistical results are shown in Table 1 below: Table 1. Comparison of Digital Release Scheduling Effects of Electronic Product Leasing Platforms

[0064] As can be seen from the data in Table 1 above, the method of the present invention is significantly superior to the original physical warehouse return scheduling method and the fixed-rule digital release verification method in terms of indicators such as digital release anomaly identification rate, false release rate, average scheduling confirmation time, manual review ratio, and order delivery anomaly rate. The original physical warehouse return scheduling method mainly allocates based on whether the equipment has returned to the warehouse and whether the inventory is available. It is insufficient in identifying issues such as account binding, configuration backfilling, and incomplete digital release closure after the return of electronic products. The digital release anomaly identification rate is only 71.6%, the false release rate reaches 8.9%, and the order delivery anomaly rate is 6.7%, indicating that some equipment with digital occupancy risks is still pre-allocated. The fixed-rule digital release verification method improves the anomaly identification capability through rule interception, increasing the digital release anomaly identification rate to 84.3% and reducing the false release rate to 5.1%. However, it still relies on fixed judgment conditions and is insufficient in identifying the continuous reverse change between the offline clearing state and the online review state. The digital release anomaly identification rate of the method of the present invention reaches 94.9%, the false release rate is reduced to 1.8%, the average scheduling confirmation time is shortened to 24.6s, the proportion of manual review is reduced to 16.9%, and the order delivery anomaly rate is reduced to 1.5%. This shows that the present invention can more accurately distinguish equipment that has completed physical return to the warehouse but whose digital release has not been closed, and reduce manual review and abnormal delivery.

[0065] This embodiment constructs a device scheduling chain to associate individual electronic product device flow segments with pending scheduling orders, transforming the scheduling object from ordinary inventory devices into traceable individual device receiving locations. After the device is returned, the offline clearing state and the online verification state are calibrated to the same release state column, enabling continuous processing of the device digital release process. An improved closed-loop continuous-time neural network model calculates the first and second state candidate values ​​within the CfC unit and uses the CUSUM algorithm recursive quantity to accumulate the inverse deviation between the offline clearing state and the online verification state, allowing issues such as account binding backflow, configuration residue, and repeated changes in release results to enter the closed-loop state update process. By embedding the CUSUM algorithm recursive quantity into time gating, the current hidden state can be dynamically updated between the digital release closing direction and the digital occupancy backflow direction. The readout layer then transforms the individual device digital release hidden state into the open or closed result of the order receiving location. The overall solution avoids the problem of judging the rentable status solely based on physical return to the warehouse, improving the scheduling reliability and delivery stability of the electronic product rental platform in continuous rental scenarios.

[0066] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A scheduling method for a rental platform based on big data processing, characterized in that, Includes the following steps: S1. Extract individual equipment flow segments from the data of the electronic product rental platform and combine them with the orders to be scheduled to form an equipment scheduling chain; S2. Separate digital release segments within the device scheduling chain and calibrate the disconnection clear state and the network verification state to the same release state column; S3. The release state column is connected to the CfC unit of the improved closed continuous-time neural network model. The CfC unit calculates the first state candidate value and the second state candidate value respectively in the closed state update, and retains the time gating from the previous hidden state to the current hidden state. S4. The CUSUM algorithm is used to accumulate the reverse deviation between the disconnection clear state and the network verification state along the release state column in the CfC unit to form the CUSUM algorithm recursive quantity. Positions that do not generate reverse deviation continue the previous hidden state. S5 and CfC units embed the CUSUM algorithm recursion into time gating, guiding the current hidden state to perform closed state updates between the first state candidate value and the second state candidate value, forming a single device digital release hidden state. S6. The readout layer reads the digital release hidden state of the single device. When the current hidden state is biased towards the first state candidate value, the order acceptance position in the device scheduling chain is opened. When the current hidden state is biased towards the second state candidate value, the order acceptance position in the device scheduling chain is stopped. S7. Collect the order acceptance locations and individual equipment flow segments after the platform is opened to form the equipment scheduling table of the electronic product rental platform.

2. The rental platform scheduling method based on big data processing according to claim 1, characterized in that, S1 specifically includes: S11. Extract the equipment records that have been transferred and the order records that are waiting to be accepted from the data of the electronic product rental platform. Group the equipment records that have been transferred according to the individual equipment identity to obtain the individual equipment record group. S12. In the individual device record group, the time sequence of the device recycling event and the platform registration event is checked, and the continuous flow segments belonging to the same individual device are extracted to form individual device flow segments. S13. Extract order segments that have a connection with the individual equipment flow segments from the order records waiting to be accepted, and calibrate the order segments and individual equipment flow segments side by side according to the event time to form candidate connection segments. S14. Eliminate candidate junction segments with inconsistent individual device identities or reversed event times, and connect the remaining candidate junction segments in series according to individual device identities to form a device scheduling chain.

3. The rental platform scheduling method based on big data processing according to claim 1, characterized in that, S2 specifically includes: S21. Check the event boundary between the individual equipment flow segment and the order to be scheduled along the equipment scheduling chain, and classify the release event that is located after the equipment return position and affects the opening of the order acceptance position into the digital release segment. S22. Within the digital release segment, distinguish between the network disconnection clearing event and the network verification event, organize the release result generated by the network disconnection clearing event into the network disconnection clearing state, and organize the verification result generated by the network verification event into the network verification state. S23. Using the same individual device identity and the same release location as the calibration benchmark, register the network disconnection clear state and the network verification state in parallel at the same release location to form a release status column; S24. For release positions that lack a network disconnection clear state or a network verification state, retain a blank mark. For release positions with inverted event times, remove the digital release segment and output the release status column that corresponds to the network disconnection clear state and the network verification state.

4. The rental platform scheduling method based on big data processing according to claim 1, characterized in that, S3 specifically includes: S31, CfC unit locates the consecutive release positions of the same single device along the release status column, and writes the event time difference between adjacent consecutive release positions into the corresponding status update position; S32. Read the current release state at the state update position, and establish a closed state update relationship for the current hidden state by combining it with the previous hidden state. S33. Calculate the first state candidate value based on the closing trend of the current release state, calculate the second state candidate value based on the reinjection trend of the current release state, and keep the first state candidate value and the second state candidate value side by side in the same state update position. S34. Calculate time gating from event time difference. Time gating limits the proportion of the first state candidate value and the second state candidate value participating in the current hidden state update, forming the state update object corresponding to the release state column.

5. The rental platform scheduling method based on big data processing according to claim 1, characterized in that, S4 specifically includes: S41 and CfC units enter the state update position one by one in the order of the release state column. At each state update position, the current release state, the previous hidden state, and the CUSUM algorithm recursion value retained by the previous state update position are read simultaneously. S42. When the network disconnection clearing state and the network verification state in the current release state fall into the same state update position, and the network disconnection clearing state points to digital release closure and the network verification state turns to digital occupation backflow, the state difference of the same state update position is determined as the reverse deviation. When the network disconnection clearing state and the network verification state keep in the same direction of closure, the current state update position is merged into the reserved path of the previous hidden state. S43. The CUSUM algorithm recursive quantity is updated within the closed state update position of the CfC unit. When encountering a reverse deviation, it is incremented along the CUSUM algorithm recursive quantity retained in the previous state update position. When encountering a same-direction closure, it is reduced along the CUSUM algorithm recursive quantity retained in the previous state update position. When the reduction result crosses the zero value boundary, it is accepted as zero value. S44. When the current release state lacks the same result of the disconnection clearing state and the network verification state, the current state update position continues along the previous hidden state, and the CUSUM algorithm recursive quantity retained in the previous state update position is used as the CUSUM algorithm recursive quantity of the current state update position. S45. After the release state column has been processed position by position, the CUSUM algorithm recursive quantities retained at each state update position are arranged in the order of the release state column to form CUSUM algorithm recursive quantities that correspond one-to-one with the closed state update positions of the CfC unit.

6. The rental platform scheduling method based on big data processing according to claim 1, characterized in that, S5 specifically includes: S51, CfC unit enters the current state update position along the release state column. The time gating first determines the basic interpolation weight for the transition from the previous hidden state to the current hidden state based on the event time interval. S52. When obtaining the CUSUM algorithm recursive quantity at the current state update position, convert the CUSUM algorithm recursive quantity into the reverse traction share, and deduct the reverse traction share from the basic interpolation weight corresponding to the first state candidate value. S53. The remaining share obtained after deduction is allocated to the first state candidate value, and the reverse traction share is allocated to the second state candidate value. When the reverse traction share exceeds the basic interpolation weight, it is deducted according to the boundary value of the basic interpolation weight. S54. The current hidden state first retains the continuous items of the previous hidden state along the event time interval, then absorbs the first state candidate value according to the remaining share after allocation, and absorbs the second state candidate value according to the reverse traction share, thus completing the closed state update of the current state update position. S55. When the current state update position has not obtained the CUSUM algorithm recursion, the time gating retains the basic interpolation weight, and the current hidden state completes the closed state update along the previous hidden state and the first state candidate value. S56 and CfC units continuously update the closed state at each state update position along the release state column, and connect the current hidden states formed by each state update position in the order of the release state column to form the digital release hidden state of a single device.

7. The rental platform scheduling method based on big data processing according to claim 1, characterized in that, S6 specifically includes: S61. The readout layer locates the order acceptance position corresponding to the hidden state of the digital release of a single device along the device scheduling chain, and reads the current hidden state formed by the corresponding state update position. S62. Extract the first state candidate value absorption share and the second state candidate value absorption share from the current hidden state. The first state candidate value absorption share takes over the digital release closing direction, and the second state candidate value absorption share takes over the digital occupation backflow direction. S63. When the first state candidate value absorbs the dominant share of the current hidden state, the order acceptance position is rewritten to the open state, and the acceptance relationship between the order acceptance position and the individual equipment flow segment is retained. S64. When the absorption share of the second state candidate value occupies the dominant share of the current hidden state, the order acceptance position is rewritten to the stop state, and the original connection relationship between the order acceptance position and the single device flow segment is maintained. S65. When the current hidden state cannot extract the first state candidate value absorption share or the second state candidate value absorption share, the order acceptance position continues along the state corresponding to the previous hidden state, and no new open state is added. S66. Complete the status rewriting of the order acceptance position according to the equipment scheduling chain sequence to form the scheduling judgment result of the order acceptance position.

8. The rental platform scheduling method based on big data processing according to claim 1, characterized in that, Specifically, S7 includes: S71. Read the scheduling judgment result of the order acceptance position along the equipment scheduling chain, keep the order acceptance position corresponding to the open state connected with the single equipment flow segment, and isolate the order acceptance position corresponding to the stopped state from the equipment scheduling chain. S72. For orders that maintain connection, the receiving position is assigned to the corresponding individual equipment flow segment according to the individual equipment identity, forming an individual equipment receiving record; S73. When the same single equipment circulation segment corresponds to multiple order acceptance positions, retain the order acceptance position that is ranked first in the equipment scheduling chain, and mark the remaining order acceptance positions as unaccepted positions. S74. Arrange the individual equipment acceptance records in the order of the equipment scheduling chain to form the equipment scheduling table of the electronic product rental platform.

Images