Electric heavy truck leasing risk assessment method based on big data analysis and reinforcement learning

Abstract

The application discloses an electric heavy truck leasing risk assessment method based on big data analysis and reinforcement learning, comprising the following steps: obtaining leasing operation data to generate a risk basis record table; generating a performance, vehicle asset and energy supply clearing account book; establishing three types of potential slots, writing unclosed, closed, over-limit and counter-evidence into the risk position to form a three-domain clearing potential state; enumerating candidate risk control actions to form a shadow potential state; rewriting the three types of occupation positions by an improved Dreamer model to generate a shadow leasing track; screening target risk control action chains according to a clearing violation penalty and outputting a risk level; writing back feedback data to refresh the three-domain clearing potential state of the next cycle. The application relates to the technical field of electric heavy truck leasing risk assessment, and improves the electric heavy truck leasing risk assessment accuracy and risk control closed loop efficiency.

Description

Technical Field

[0001] This invention relates to the field of risk assessment technology for electric heavy-duty truck leasing, and in particular to a risk assessment method for electric heavy-duty truck leasing based on big data analysis and reinforcement learning. Background Technology

[0002] Electric heavy-duty truck leasing involves lessee performance, vehicle asset occupancy, power battery status, energy replenishment resource allocation, and vehicle recycling management. Existing leasing risk assessment methods mostly rely on contract payment records, vehicle location records, maintenance records, energy replenishment records, and manual verification results as evaluation criteria, and form risk levels through fixed rules, static scoring tables, or manual review.

[0003] Existing methods, when processing multi-source leasing operation data, focus on judging individual indicators and lack a domain-specific clearing structure that addresses lessee performance, vehicle asset occupancy, and energy replenishment resource occupancy. Risk states such as incomplete payment, vehicle overuse, abnormal energy replenishment, and feedback with reverse proof are easily recorded in a scattered manner, making it difficult to form a unified expression of closed, incomplete, overuse, and reverse proof states within the same leasing period.

[0004] Current methods for generating risk control measures often directly output suggestions for collection, load limiting, maintenance, or energy replenishment based on the current risk level, lacking a shadowing process before and after the intervention of candidate risk control actions. This makes it difficult to determine the impact of candidate risk control actions on the subsequent evolution of leasing risks. Furthermore, the lack of comparable evidence based on penalties for breaches of debt clearing among different candidate risk control actions easily leads to invalid actions, duplicate actions, and actions with unclosed risks.

[0005] Existing methods lack completeness in writing back leasing feedback data and updating risk status for the next leasing cycle after risk control actions are executed, making it difficult to achieve closed-loop processing between leasing risk assessment, risk control action generation, execution feedback, and status updates. Therefore, providing a risk assessment method for electric heavy-duty truck leasing based on big data analysis and reinforcement learning 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 risk assessment method for electric heavy-duty truck leasing based on big data analysis and reinforcement learning. This invention utilizes leasing operation data, potential states of three-domain clearing, and an improved Dreamer model to complete risk extrapolation, risk control action screening, and feedback closed-loop updates, possessing the advantages of accurate assessment, effective actions, and strong closed-loop performance.

[0007] The electric heavy-duty truck leasing risk assessment method based on big data analysis and reinforcement learning according to embodiments of the present invention includes the following steps:

[0008] Acquire electric heavy-duty truck leasing operation data according to the leasing period and generate a basic record table of leasing risks;

[0009] Based on the lease risk basic record form, generate performance settlement ledgers, vehicle asset settlement ledgers, and energy supply settlement ledgers;

[0010] Establish potential performance slots, potential asset slots, and potential energy slots according to the clearing item number. Write unclosed items in the three types of ledgers into the risk retention position, closed items into the risk release position, and over-limit items and counter-evidence items into the risk enhancement position to form a three-domain clearing potential status.

[0011] Read the risk retention bit and risk enhancement bit, enumerate candidate risk control actions, and copy a set of three-domain clearing potential states for each candidate risk control action to form a shadow potential state.

[0012] In the improved Dreamer model, the payment occupancy, asset occupancy, and energy replenishment occupancy in the shadow potential state are rewritten according to the candidate risk control actions, and the shadow leasing trajectory is obtained by rolling.

[0013] By comparing the risk retention position, risk release position and risk enhancement position in the shadow leasing trajectory, for the cumulative clearing violation penalty of unclosed, over-limit and counter-evidence, the clearing violation penalty before the candidate risk control action is executed is used as the benchmark, the candidate risk control action with increased penalty is blocked, the candidate risk control action with reduced penalty is retained, the target risk control action chain is generated and the comprehensive leasing risk level is output.

[0014] After executing the target risk control action chain, read the leasing feedback data, write the leasing feedback data back to the three types of ledgers, and refresh the potential status of the three-domain clearing in the next leasing cycle.

[0015] Optionally, the operation of generating the basic record table of lease risks includes:

[0016] Read leasing business data, vehicle operation data, energy replenishment data, and leasing feedback data according to the leasing period;

[0017] Using lessee, vehicle, battery pack, charging station, and lease period as the aggregation index, data belonging to the same aggregation index are merged into a basic record of lease risk;

[0018] Configure record numbers according to the generation order of the lease risk basic records to form a lease risk basic record table.

[0019] Optionally, the operations for generating the performance settlement ledger, vehicle asset settlement ledger, and energy supply settlement ledger include:

[0020] Read the leasing business field, vehicle operation field, energy supply field, and leasing feedback field from the basic leasing risk record table by record number;

[0021] Register the data corresponding to the lessee's performance in the lease business field and lease feedback field as performance settlement items to form a performance settlement ledger;

[0022] Register the data corresponding to vehicle asset occupancy in the vehicle operation field and leasing feedback field as vehicle asset clearing items to form a vehicle asset clearing ledger.

[0023] The data corresponding to the energy replenishment resource occupation in the energy replenishment field and the leasing feedback field are registered as energy replenishment clearing items, forming an energy replenishment clearing ledger.

[0024] Optionally, the operation of forming the three-domain clearing potential state includes:

[0025] Using the settlement item numbers of performance settlement items, vehicle asset settlement items, and energy supply settlement items as slot indexes, potential performance slots, potential asset slots, and potential energy slots are established respectively;

[0026] Within each potential slot, set up risk retention positions, risk release positions, and risk enhancement positions;

[0027] Data for which no corresponding lease feedback field has been obtained or where the corresponding lease feedback field is incomplete is marked as an unclosed item and written to the risk reserved bit;

[0028] Data that meets the settlement agreement in the corresponding lease feedback field is marked as a closed item and written into the risk release position; data whose collected value exceeds the settlement constraint is marked as an over-limit item; data whose lease feedback field is contrary to the settlement agreement is marked as a counter-evidence item; over-limit items and counter-evidence items are written into the risk enhancement position.

[0029] The risk retention position, risk release position, and risk enhancement position are arranged in the domain order of performance potential slots, asset potential slots, and energy potential slots to form a three-domain clearing potential state.

[0030] Optionally, the operation of forming the shadow latent state includes:

[0031] In the potential states of the three-domain clearing, potential slots for writing unclosed items, over-limit items, or counter-evidence items are selected as action trigger slots.

[0032] Based on the performance potential slot, asset potential slot, or energy potential slot to which the action trigger slot belongs, generate corresponding performance-related candidate risk control actions, asset-related candidate risk control actions, or energy replenishment-related candidate risk control actions.

[0033] Configure candidate risk control action numbers for candidate risk control actions, and bind the candidate risk control action numbers to the clearing item numbers of the action trigger slots;

[0034] Copy the three domain clearing potential states according to the candidate risk control action number to obtain the shadow potential state corresponding to the candidate risk control action number;

[0035] The correspondence between performance potential slots, asset potential slots, energy potential slots, risk retention slots, risk release slots, and risk enhancement slots is preserved in the shadow potential state.

[0036] Optionally, the operation of scrolling to obtain the shadow rental trajectory includes:

[0037] Input the shadow potential state corresponding to the candidate risk control action number into the improved Dreamer model;

[0038] In the shadow potential state, set the payment occupancy position corresponding to the performance potential slot, the asset occupancy position corresponding to the asset potential slot, and the energy replenishment occupancy position corresponding to the energy potential slot;

[0039] According to the clearing item number bound to the candidate risk control action number, write the performance-related candidate risk control action into the payment occupancy position, write the asset-related candidate risk control action into the asset occupancy position, and write the energy replenishment-related candidate risk control action into the energy replenishment occupancy position.

[0040] The improved Dreamer model performs continuous state transitions on the shadow potential state after writing the occupied bit according to the lease cycle node order, and obtains the shadow trajectory state corresponding to each lease cycle node.

[0041] By connecting the shadow trajectory status according to the candidate risk control action number, a shadow leasing trajectory is formed.

[0042] Optionally, the cumulative clearing violation penalty operation includes:

[0043] Establish a clearing violation matrix based on candidate risk control action numbers and lease cycle nodes;

[0044] Based on the closed term of the risk release position in the shadow trajectory state, generate the closed coverage relationship of the corresponding clearing item number in the clearing violation matrix;

[0045] Compare the unclosed items in the risk reservation position with the closed coverage relationship, and identify the unclosed items that are not covered by the closed coverage relationship as unclosed violation items;

[0046] The out-of-limit items in the risk-enhancing positions are identified as out-of-limit violations, and the counter-evidence items in the risk-enhancing positions are identified as counter-evidence violations.

[0047] Preset penalty weights are matched for unclosed violations, excessive violations, and violations that are proven by contradiction, and node clearing violation penalties are generated for each lease cycle node.

[0048] Accumulate the clearing violation penalties for the same candidate risk control action number according to the order of the lease cycle nodes to obtain the clearing violation penalty corresponding to the candidate risk control action number.

[0049] Optionally, the operation of generating the target risk control action chain includes:

[0050] The shadow potential state where the payment occupancy bit, asset occupancy bit, and energy replenishment occupancy bit are not written into the candidate risk control action number is determined as the baseline shadow potential state.

[0051] The benchmark shadow lease trajectory is obtained by performing continuous state transitions on the benchmark shadow potential state in the order of lease cycle nodes using the improved Dreamer model.

[0052] Accumulated liquidation violations are applied to the risk release, risk retention, and risk enhancement positions in the benchmark shadow leasing trajectory to obtain benchmark liquidation violation penalties.

[0053] Compare the penalty for each candidate risk control action number corresponding to the clearing violation with the benchmark clearing violation penalty;

[0054] Candidate risk control actions with a comparison result greater than or equal to the benchmark clearing violation penalty are marked as blocked actions, while candidate risk control actions with a comparison result less than the benchmark clearing violation penalty are marked as reserved actions.

[0055] Arrange the retained actions according to the order of the lease cycle nodes and the order of the clearing item numbers corresponding to the retained actions, and generate the target risk control action chain.

[0056] Optionally, the operation of outputting the comprehensive lease risk level includes:

[0057] Load the reserved actions in the target risk control action chain into the baseline shadow potential state according to the order of the chain;

[0058] According to the candidate risk control actions of performance, asset and energy replenishment categories to which the reserved action belongs, update the payment occupancy bit, asset occupancy bit and energy replenishment occupancy bit in the benchmark shadow potential state respectively.

[0059] The target chain shadow leasing trajectory is obtained by performing continuous state transitions on the baseline shadow latent state after updating the occupied bits using the improved Dreamer model.

[0060] Accumulated liquidation violations are performed on the risk release, risk retention, and risk enhancement positions in the target chain shadow leasing trajectory to obtain the target chain liquidation violation penalty.

[0061] The penalty for breach of target chain clearing is compared with the preset risk level threshold table to determine the overall leasing risk level;

[0062] The risk level of the comprehensive lease is associated with the target risk control action chain, the penalty for violation of the target chain clearing, and the corresponding clearing item number.

[0063] Optionally, the operation of refreshing the three-domain clearing potential status for the next lease cycle includes:

[0064] Execute the target risk control action chain and receive leasing feedback data;

[0065] Based on the candidate risk control action number and the clearing item number, the leasing feedback data will be written back to the corresponding clearing items in the performance clearing ledger, vehicle asset clearing ledger and energy supply clearing ledger respectively;

[0066] Based on the rewritten performance settlement ledger, vehicle asset settlement ledger and energy supply settlement ledger, the unclosed items, closed items, over-limit items and counter-evidence items were re-evaluated;

[0067] The risk retention position, risk release position, and risk enhancement position are updated according to the reassessment results to form the potential three-domain liquidation status for the next lease cycle.

[0068] The beneficial effects of this invention are:

[0069] (1) This invention sorts out the lessee’s payments, vehicle occupancy, power batteries and energy replenishment resources by using performance settlement ledger, vehicle asset settlement ledger and energy replenishment ledger. It writes unclosed items, closed items, over-limit items and counter-evidence items into the corresponding risk positions, so that the risk status of electric heavy truck leasing is expressed in a unified manner, and improves the completeness of risk identification of multi-source leasing operation data.

[0070] (2) This invention uses the three-domain clearing potential state, shadow potential state and improved Dreamer model to perform rolling deduction of the shadow leasing trajectory after the intervention of candidate risk control actions, and selects the target risk control action chain based on the clearing violation penalty, thereby reducing invalid risk control actions and repeated risk control actions and improving the matching degree between risk control actions and leasing risk status.

[0071] (3) After the target risk control action chain is executed, the present invention writes the lease feedback data back to the performance settlement ledger, vehicle asset settlement ledger, and energy supply settlement ledger, and updates the risk retention position, risk release position, and risk enhancement position to form the three-domain settlement potential status for the next lease cycle, thereby realizing a closed-loop process of risk assessment, risk control action execution, feedback writing back, and status refresh. Attached Figure Description

[0072] 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:

[0073] Figure 1 The flowchart shows the electric heavy-duty truck leasing risk assessment method based on big data analysis and reinforcement learning proposed in this invention.

[0074] Figure 2 This is a three-domain clearing potential state structure diagram of the electric heavy truck leasing risk assessment method based on big data analysis and reinforcement learning proposed in this invention.

[0075] Figure 3 This is a flowchart illustrating the shadow deduction and feedback refresh process of the electric heavy-duty truck leasing risk assessment method based on big data analysis and reinforcement learning proposed in this invention. Detailed Implementation

[0076] 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.

[0077] refer to Figures 1-3 The risk assessment method for electric heavy-duty truck leasing based on big data analysis and reinforcement learning includes the following steps:

[0078] Acquire electric heavy-duty truck leasing operation data according to the leasing period and generate a basic record table of leasing risks;

[0079] Based on the lease risk basic record form, generate performance settlement ledgers, vehicle asset settlement ledgers, and energy supply settlement ledgers;

[0080] Establish potential performance slots, potential asset slots, and potential energy slots according to the clearing item number. Write unclosed items in the three types of ledgers into the risk retention position, closed items into the risk release position, and over-limit items and counter-evidence items into the risk enhancement position to form a three-domain clearing potential status.

[0081] Read the risk retention bit and risk enhancement bit, enumerate candidate risk control actions, and copy a set of three-domain clearing potential states for each candidate risk control action to form a shadow potential state.

[0082] In the improved Dreamer model, the payment occupancy, asset occupancy, and energy replenishment occupancy in the shadow potential state are rewritten according to the candidate risk control actions, and the shadow leasing trajectory is obtained by rolling.

[0083] By comparing the risk retention position, risk release position and risk enhancement position in the shadow leasing trajectory, for the cumulative clearing violation penalty of unclosed, over-limit and counter-evidence, the clearing violation penalty before the candidate risk control action is executed is used as the benchmark, the candidate risk control action with increased penalty is blocked, the candidate risk control action with reduced penalty is retained, the target risk control action chain is generated and the comprehensive leasing risk level is output.

[0084] After executing the target risk control action chain, read the leasing feedback data, write the leasing feedback data back to the three types of ledgers, and refresh the potential status of the three-domain clearing in the next leasing cycle.

[0085] In this embodiment, the operation of generating the basic record table of lease risks includes:

[0086] Read leasing business data, vehicle operation data, energy replenishment data, and leasing feedback data according to the leasing period;

[0087] Using lessee, vehicle, battery pack, charging station, and lease period as the aggregation index, data belonging to the same aggregation index are merged into a basic record of lease risk;

[0088] Configure record numbers according to the generation order of the lease risk basic records to form a lease risk basic record table.

[0089] The lease period refers to the data time period corresponding to the risk assessment of the same round of electric heavy-duty truck leasing. The lease period can be divided according to the settlement cycle of the lease contract or according to the risk assessment cycle. Each lease period is assigned a lease period number, and data falling within the start and end time of the same lease period is written under the same lease period number.

[0090] The leasing business data includes lessee identification, contract number, leased vehicle identification, lease start time, lease end time, rent payable, rent actually paid, deposit amount, guarantee identification, payment node, operating area constraints, load constraints, energy replenishment constraints, and agreed vehicle return time.

[0091] Vehicle operation data includes vehicle identification, vehicle location, mileage, load value, work area, work start time, work end time, number of times unmanned takeover was performed, number of times the route deviated, number of times the vehicle made an emergency stop, vehicle fault identification, maintenance identification, and vehicle recovery location.

[0092] Energy replenishment data includes vehicle identification, battery pack identification, replenishment station identification, charging start time, charging end time, battery swapping time, battery swapping identification, replenishment waiting time, photovoltaic energy supply, remaining energy storage capacity, energy storage discharge, and replenishment completion status.

[0093] The rental feedback data includes contract number, lessee identifier, vehicle identifier, payment feedback identifier, operation restriction execution identifier, load adjustment execution identifier, maintenance execution feedback identifier, charging execution feedback identifier, battery status feedback identifier, and vehicle recycling feedback identifier.

[0094] When reading leasing business data, vehicle operation data, energy replenishment data, and leasing feedback data by leasing period, the leasing start time, leasing end time, and payment node in the leasing business data are mapped to the leasing period number; the operation start time and operation end time in the vehicle operation data are mapped to the leasing period number; the charging start time, charging end time, and battery swapping time in the energy replenishment data are mapped to the leasing period number; and the feedback generation time in the leasing feedback data is mapped to the leasing period number.

[0095] When using lessee, vehicle, battery pack, charging station, and lease period as the aggregation index, the lessee identifier in the leasing business data is organized into lessee number, the leased vehicle identifier and the vehicle identifier in the vehicle operation data are organized into vehicle number, the battery pack identifier and the battery swapping battery identifier are organized into battery pack number, the charging station identifier is organized into charging station number, and the lease period number is used as the period index field.

[0096] Data belonging to the same aggregation index refers to data where the lessee number, vehicle number, battery pack number, refueling station number, and lease period number are all identical. Leasing business data, vehicle operation data, energy replenishment data, and leasing feedback data belonging to the same aggregation index are written into the same record. This record includes the leasing business field, vehicle operation field, energy replenishment field, and leasing feedback field, thus obtaining the basic record for leasing risk.

[0097] When configuring record numbers according to the generation order of the lease risk basic records, a unique record number is configured after each lease risk basic record is written. The record numbers increment sequentially according to the order in which the lease risk basic records are written. Multiple lease risk basic records configured with record numbers form a lease risk basic record table.

[0098] In this embodiment, the operations for generating the performance settlement ledger, vehicle asset settlement ledger, and energy supply settlement ledger include:

[0099] Read the leasing business field, vehicle operation field, energy supply field, and leasing feedback field from the basic leasing risk record table by record number;

[0100] Register the data corresponding to the lessee's performance in the lease business field and lease feedback field as performance settlement items to form a performance settlement ledger;

[0101] Register the data corresponding to vehicle asset occupancy in the vehicle operation field and leasing feedback field as vehicle asset clearing items to form a vehicle asset clearing ledger.

[0102] The data corresponding to the energy replenishment resource occupation in the energy replenishment field and the leasing feedback field are registered as energy replenishment clearing items, forming an energy replenishment clearing ledger.

[0103] The record number is the number of each basic lease risk record in the basic lease risk record table. When reading the basic lease risk record table by record number, each basic lease risk record is read one by one according to the record number, and the corresponding lease business field, vehicle operation field, energy supply field, and lease feedback field are read under the same record number.

[0104] The rental transaction fields include lessee number, contract number, vehicle number, rental period number, rental start time, rental end time, rent payable, rent actually paid, deposit amount, guarantee mark, and agreed vehicle return time.

[0105] Vehicle operation fields include vehicle number, battery pack number, rental period number, mileage, load value, operating area, operating time, number of times unmanned takeover has occurred, number of times the route has deviated from the route, number of times the vehicle has made an emergency stop, vehicle fault identification, and maintenance identification.

[0106] The energy replenishment fields include vehicle number, battery pack number, replenishment station number, lease period number, charging time, battery swapping time, replenishment waiting time, photovoltaic energy supply, remaining energy storage capacity, and replenishment completion indicator.

[0107] The rental feedback fields include contract number, lessee number, vehicle number, rental period number, payment feedback identifier, maintenance execution feedback identifier, charging execution feedback identifier, battery status feedback identifier, and vehicle recycling feedback identifier.

[0108] When registering data corresponding to the lessee's performance in the lease business and lease feedback fields as performance settlement items, the lessee number, contract number, vehicle number, lease period number, rent payable, rent actually paid, deposit amount, guarantee identifier, vehicle return agreement time, payment feedback identifier, and vehicle return feedback identifier should be written into the performance settlement item. Each performance settlement item is configured with a performance settlement item number, record number, performance settlement item type, and a set of performance data values. Performance settlement item types include rent settlement type, deposit settlement type, guarantee settlement type, and return settlement type. Multiple performance settlement items are grouped by record number to form a performance settlement ledger.

[0109] When registering data corresponding to vehicle asset occupancy in the vehicle operation and leasing feedback fields as vehicle asset clearing items, the vehicle number, battery pack number, lease period number, mileage, load value, work area, work time, number of times autonomous driving takeovers, number of path deviations, number of abnormal emergency stops, vehicle fault identifier, maintenance identifier, maintenance execution feedback identifier, and battery status feedback identifier are written into the vehicle asset clearing item. Each vehicle asset clearing item is configured with a vehicle asset clearing item number, record number, vehicle asset clearing item type, and a set of vehicle asset occupancy values. Vehicle asset clearing item types include mileage clearing type, load clearing type, work area clearing type, autonomous driving anomaly clearing type, battery status clearing type, vehicle fault clearing type, and maintenance clearing type. Multiple vehicle asset clearing items are grouped by record number to form a vehicle asset clearing ledger.

[0110] When registering data corresponding to energy replenishment resource occupancy in the energy replenishment and leasing feedback fields as energy replenishment settlement items, the vehicle number, battery pack number, energy replenishment station number, leasing period number, charging time, battery swapping time, energy replenishment waiting time, photovoltaic energy supply, remaining energy storage capacity, energy replenishment completion identifier, and energy replenishment execution feedback identifier are written into the energy replenishment settlement item. Each energy replenishment settlement item is configured with an energy replenishment settlement item number, record number, energy replenishment settlement item type, and a set of energy replenishment resource occupancy values. Energy replenishment settlement item types include charging settlement type, battery swapping settlement type, waiting settlement type, photovoltaic energy supply settlement type, energy storage support settlement type, and energy replenishment feedback settlement type. Multiple energy replenishment settlement items are grouped by record number to form an energy replenishment settlement ledger.

[0111] In this embodiment, the operations that form the potential state of three-domain clearing include:

[0112] Using the settlement item numbers of performance settlement items, vehicle asset settlement items, and energy supply settlement items as slot indexes, potential performance slots, potential asset slots, and potential energy slots are established respectively;

[0113] Within each potential slot, set up risk retention positions, risk release positions, and risk enhancement positions;

[0114] Data for which no corresponding lease feedback field has been obtained or where the corresponding lease feedback field is incomplete is marked as an unclosed item and written to the risk reserved bit;

[0115] Data that meets the settlement agreement in the corresponding lease feedback field is marked as a closed item and written into the risk release position; data whose collected value exceeds the settlement constraint is marked as an over-limit item; data whose lease feedback field is contrary to the settlement agreement is marked as a counter-evidence item; over-limit items and counter-evidence items are written into the risk enhancement position.

[0116] The risk retention position, risk release position, and risk enhancement position are arranged in the domain order of performance potential slots, asset potential slots, and energy potential slots to form a three-domain clearing potential state.

[0117] The clearing item numbers include performance clearing item numbers, vehicle asset clearing item numbers, and energy supply clearing item numbers. Performance clearing item numbers are derived from the performance clearing ledger, vehicle asset clearing item numbers are derived from the vehicle asset clearing ledger, and energy supply clearing item numbers are derived from the energy supply clearing ledger. Each clearing item number corresponds to a potential slot.

[0118] When using the settlement item number as the slot index, the settlement item number is written into the performance field to create a potential performance slot; the vehicle asset settlement item number is written into the asset field to create an asset potential slot; and the energy supply settlement item number is written into the energy field to create an energy potential slot. Potential performance slots, asset potential slots, and energy potential slots are stored separately, and potential slots in different fields do not overlap.

[0119] Each potential slot registers a clearing item number, record number, clearing item type, risk reservation position, risk release position, and risk enhancement position. The risk reservation position, risk release position, and risk enhancement position are all set to the same clearing item number.

[0120] When writing to a risk-reserved bit, read the clearing item number corresponding to the potential slot and query the lease feedback field corresponding to that clearing item number. If no corresponding lease feedback field is found, or if the found corresponding lease feedback field is empty, pending execution, or a processing value, mark the clearing item as an unclosed item and write the unclosed item identifier to the risk-reserved bit.

[0121] When writing to a risk release position, the clearing item agreement and corresponding lease feedback field corresponding to the potential slot are read. When the corresponding lease feedback field meets the clearing item agreement, the clearing item is marked as a closed item, and the closed item identifier is written to the risk release position.

[0122] When writing to the risk enhancement bit, the system reads the collected value, clearing item constraint, clearing item agreement, and lease feedback field corresponding to the potential slot. If the collected value exceeds the clearing item constraint, the clearing item is marked as an over-limit item, and the over-limit item identifier is written to the risk enhancement bit. If the lease feedback field contradicts the clearing item agreement, the clearing item is marked as a counter-evidence item, and the counter-evidence item identifier is written to the risk enhancement bit.

[0123] When arranging risk retention, risk release, and risk enhancement positions in domain order, first arrange the risk retention, risk release, and risk enhancement positions within the performance potential slots, then arrange the risk retention, risk release, and risk enhancement positions within the asset potential slots, and finally arrange the risk retention, risk release, and risk enhancement positions within the energy potential slots. The arranged performance potential slots, asset potential slots, and energy potential slots together constitute the three-domain liquidation potential status.

[0124] In this embodiment, the operations for forming the shadow latent state include:

[0125] In the potential states of the three-domain clearing, potential slots for writing unclosed items, over-limit items, or counter-evidence items are selected as action trigger slots.

[0126] Based on the performance potential slot, asset potential slot, or energy potential slot to which the action trigger slot belongs, generate corresponding performance-related candidate risk control actions, asset-related candidate risk control actions, or energy replenishment-related candidate risk control actions.

[0127] Configure candidate risk control action numbers for candidate risk control actions, and bind the candidate risk control action numbers to the clearing item numbers of the action trigger slots;

[0128] Copy the three domain clearing potential states according to the candidate risk control action number to obtain the shadow potential state corresponding to the candidate risk control action number;

[0129] The correspondence between performance potential slots, asset potential slots, energy potential slots, risk retention slots, risk release slots, and risk enhancement slots is preserved in the shadow potential state.

[0130] Among them, the action trigger slot is a potential slot for writing unclosed item identifiers in the risk reservation slot, or for writing over-limit item identifiers and counter-evidence item identifiers in the risk enhancement slot. Each action trigger slot registers the domain to which the slot belongs, the clearing item number, the risk slot type, and the risk item identifier.

[0131] When filtering action trigger slots, the potential status of the three domains for clearing is read in the domain order of performance potential slots, asset potential slots, and energy potential slots. If a potential slot with an unclosed item flag is read in the risk reservation field, the potential slot is written to the action trigger slot table; if a potential slot with an over-limit item flag or a counter-evidence flag is read in the risk enhancement field, the potential slot is written to the action trigger slot table.

[0132] When an action trigger slot belongs to a potential performance slot, a performance-related candidate risk control action is generated according to the performance clearing item type corresponding to that action trigger slot. Performance-related candidate risk control actions include payment collection actions, deposit adjustment actions, guarantee verification actions, and vehicle recovery verification actions.

[0133] When an action trigger slot belongs to a potential asset slot, an asset-type candidate risk control action is generated according to the vehicle asset clearing item type corresponding to that action trigger slot. Asset-type candidate risk control actions include load limit actions, operating area restriction actions, unmanned driving verification actions, battery status check actions, vehicle fault diagnosis actions, and maintenance trigger actions.

[0134] When the trigger slot for an action is a potential energy slot, candidate energy replenishment risk control actions are generated according to the energy replenishment clearing item type corresponding to that trigger slot. Candidate energy replenishment risk control actions include charging adjustment actions, battery swapping adjustment actions, energy replenishment waiting processing actions, photovoltaic energy supply allocation actions, energy storage support allocation actions, and energy replenishment feedback verification actions.

[0135] When configuring candidate risk control action numbers, the candidate risk control action numbers are configured according to the generation order of the candidate risk control actions. When a candidate risk control action number is bound to a clearing item number, the candidate risk control action number, the clearing item number of the action trigger slot, the domain to which the slot belongs, the risk position type, the risk item identifier, and the candidate risk control action type are written into the same candidate risk control action record.

[0136] When copying the three-domain clearing potential status according to the candidate risk control action number, one copying operation corresponds to one candidate risk control action number. The copying operation retains the performance potential slot, asset potential slot, and energy potential slot in the three-domain clearing potential status, and retains the risk retention slot, risk release slot, and risk enhancement slot in each potential slot.

[0137] Upon obtaining the shadow potential state, the copied three-domain clearing potential state is marked as the shadow potential state under the corresponding candidate risk control action number. Each shadow potential state registers the candidate risk control action number, the corresponding clearing item number, the performance potential slot, the asset potential slot, the energy potential slot, the risk retention slot, the risk release slot, and the risk enhancement slot.

[0138] When preserving the correspondence in the shadow potential state, the clearing item number, potential slot and risk position in the three-domain clearing potential state are mapped to the potential slot and risk position under the same clearing item number in the shadow potential state, so that each candidate risk control action number corresponds to an independent shadow potential state.

[0139] In this embodiment, the operation of obtaining the shadow rental trajectory by scrolling includes:

[0140] Input the shadow potential state corresponding to the candidate risk control action number into the improved Dreamer model;

[0141] In the shadow potential state, set the payment occupancy position corresponding to the performance potential slot, the asset occupancy position corresponding to the asset potential slot, and the energy replenishment occupancy position corresponding to the energy potential slot;

[0142] According to the clearing item number bound to the candidate risk control action number, write the performance-related candidate risk control action into the payment occupancy position, write the asset-related candidate risk control action into the asset occupancy position, and write the energy replenishment-related candidate risk control action into the energy replenishment occupancy position.

[0143] The improved Dreamer model performs continuous state transitions on the shadow potential state after writing the occupied bit according to the lease cycle node order, and obtains the shadow trajectory state corresponding to each lease cycle node.

[0144] By connecting the shadow trajectory status according to the candidate risk control action number, a shadow leasing trajectory is formed.

[0145] Specifically, when inputting the shadow potential state corresponding to the candidate risk control action number into the improved Dreamer model, the candidate risk control action number, candidate risk control action type, bound clearing item number, performance potential slot, asset potential slot, energy potential slot, risk retention slot, risk release slot, and risk enhancement slot are written into the same model input record.

[0146] When setting payment occupancy, asset occupancy, and energy replenishment occupancy positions in the shadow potential state, a payment occupancy position is set within each performance potential slot, an asset occupancy position is set within each asset potential slot, and an energy replenishment occupancy position is set within each energy potential slot. The payment occupancy position registers the candidate risk control action number, clearing item number, and occupancy identifier corresponding to performance-related candidate risk control actions. Performance-related candidate risk control actions include those for payments, deposits, guarantees, and vehicle recovery. The asset occupancy position registers the candidate risk control action number, clearing item number, and occupancy identifier corresponding to asset-related candidate risk control actions. The energy replenishment occupancy position registers the candidate risk control action number, clearing item number, and occupancy identifier corresponding to energy replenishment-related candidate risk control actions. When the occupancy identifier is 1, a candidate risk control action has been written to the corresponding occupancy position; when the occupancy identifier is 0, a candidate risk control action has not been written to the corresponding occupancy position.

[0147] When writing a candidate risk control action for performance into the payment occupancy slot, according to the clearing item number bound to the candidate risk control action number, search for a performance potential slot with the same clearing item number in the shadow potential state, write the candidate risk control action number and the clearing item number into the payment occupancy slot of that performance potential slot, and set the occupancy flag of the payment occupancy slot to 1.

[0148] When writing candidate risk control actions for assets into the asset occupancy slot, according to the clearing item number bound to the candidate risk control action number, search for the asset potential slot with the same clearing item number in the shadow potential state, write the candidate risk control action number and the clearing item number into the asset occupancy slot of the asset potential slot, and set the occupancy flag of the asset occupancy slot to 1.

[0149] When writing a candidate risk control action for energy replenishment into the energy replenishment occupancy bit, according to the clearing item number bound to the candidate risk control action number, search for the energy potential slot with the same clearing item number in the shadow potential state, write the candidate risk control action number and the clearing item number into the energy replenishment occupancy bit of the energy potential slot, and set the occupancy flag of the energy replenishment occupancy bit to 1.

[0150] The lease cycle node sequence includes payment node, operation node, refueling node, maintenance node, and recycling node. The improved Dreamer model establishes a node state sequence according to the arrangement of payment node, operation node, refueling node, maintenance node, and recycling node.

[0151] When performing continuous state transitions, the shadow potential state written after the occupancy bit is used as the input state of the payment node; the shadow trajectory state output by the previous node is read at the current lease cycle node, and the occupancy bit corresponding to the current lease cycle node is read; the payment node and the recycling node correspond to the payment occupancy bit, the operation node and the maintenance node correspond to the asset occupancy bit, and the recharge node corresponds to the recharge occupancy bit.

[0152] If there is a record with an occupation flag of 1 in the occupied slot corresponding to the current lease cycle node, the candidate risk control action number in that occupied slot is written to the node occupation record under the same clearing item number. When the current lease cycle node is a payment node or a recovery node, the node occupation record is written to the corresponding performance potential slot; when the current lease cycle node is an operation node or a maintenance node, the node occupation record is written to the corresponding asset potential slot; when the current lease cycle node is a refueling node, the node occupation record is written to the corresponding energy potential slot.

[0153] After the node occupancy record is written at each lease cycle node, the risk retention, risk release, and risk enhancement positions of the potential performance slots, potential asset slots, and potential energy slots under that node are retained, generating the shadow trajectory status corresponding to that lease cycle node. Each shadow trajectory status registers the candidate risk control action number, lease cycle node, three types of potential slot numbers, occupancy identifiers of the three types of occupancy positions, and current identifiers of the three types of risk positions.

[0154] When connecting shadow trajectory status by candidate risk control action number, the shadow trajectory status corresponding to payment node, operation node, energy replenishment node, maintenance node and recycling node under the same candidate risk control action number are arranged in node order and written into the same shadow leasing trajectory.

[0155] In this implementation, the operations for accumulating penalties for breach of accounting regulations include:

[0156] Establish a clearing violation matrix based on candidate risk control action numbers and lease cycle nodes;

[0157] Based on the closed term of the risk release position in the shadow trajectory state, generate the closed coverage relationship of the corresponding clearing item number in the clearing violation matrix;

[0158] Compare the unclosed items in the risk reservation position with the closed coverage relationship, and identify the unclosed items that are not covered by the closed coverage relationship as unclosed violation items;

[0159] The out-of-limit items in the risk-enhancing positions are identified as out-of-limit violations, and the counter-evidence items in the risk-enhancing positions are identified as counter-evidence violations.

[0160] Preset penalty weights are matched for unclosed violations, excessive violations, and violations that are proven by contradiction, and node clearing violation penalties are generated for each lease cycle node.

[0161] Accumulate the clearing violation penalties for the same candidate risk control action number according to the order of the lease cycle nodes to obtain the clearing violation penalty corresponding to the candidate risk control action number.

[0162] The clearing violation matrix uses candidate risk control action numbers as one type of index and lease cycle nodes as another type of index. Candidate risk control action numbers are derived from shadow lease trajectories, and lease cycle nodes are derived from the node sequence corresponding to the shadow trajectory status. Each candidate risk control action number and each lease cycle node corresponds to a matrix cell in the clearing violation matrix.

[0163] When establishing the clearing violation matrix, payment nodes, operation nodes, energy replenishment nodes, maintenance nodes, and recovery nodes under the same candidate risk control action number are arranged sequentially, and the shadow trajectory status corresponding to each lease cycle node is associated with the corresponding matrix unit. Each matrix unit records the candidate risk control action number, lease cycle node, clearing item number, closure coverage relationship, unclosed violation, over-limit violation, counter-evidence violation, and node clearing violation penalty.

[0164] When generating a closed coverage relationship, the risk release point is determined in the shadow trajectory state corresponding to the current lease cycle node. If a closed item exists in the risk release point, the clearing item number corresponding to the closed item is extracted, and the clearing item number, the closed item identifier, and the current lease cycle node are registered as a closed coverage relationship.

[0165] Closed coverage relationships take effect under the same candidate risk control action number. If a clearing item number has already generated a closed coverage relationship in the previous lease cycle node, that clearing item number will remain in a closed coverage state in subsequent lease cycle nodes and will not be mixed with closed coverage relationships under other candidate risk control action numbers.

[0166] When an unclosed violation is identified, a risk reservation position is determined in the shadow trajectory status corresponding to the current lease cycle node. If an unclosed violation exists in the risk reservation position, the clearing item number corresponding to the unclosed violation is extracted, and the clearing item number is compared with the closed coverage relationship already generated under the same candidate risk control action number.

[0167] If the clearing item number corresponding to an unclosed item matches a closure coverage relationship, the unclosed item is not registered as an unclosed violation. If the clearing item number corresponding to an unclosed item does not match a closure coverage relationship, the unclosed item is identified as an unclosed violation, and the clearing item number, lease cycle node, and unclosed item identifier corresponding to the unclosed item are registered in the corresponding matrix cell.

[0168] Within the same matrix unit, when the same clearing item number corresponds to both a closed item and an unclosed item, the closure coverage relationship shall prevail, and no unclosed violation item corresponding to that clearing item number shall be generated.

[0169] When identifying violations exceeding limits and violations requiring rebuttal, a risk enhancement bit is determined in the shadow trajectory state corresponding to the current lease cycle node. If a violation exceeding limits exists in the risk enhancement bit, it is identified as a violation exceeding limits, and the clearing item number, lease cycle node, and violation item identifier corresponding to the violation are recorded in the corresponding matrix cell. If a violation requiring rebuttal exists in the risk enhancement bit, it is identified as a violation requiring rebuttal, and the clearing item number, lease cycle node, and rebuttal item identifier corresponding to the violation are recorded in the corresponding matrix cell.

[0170] The preset penalty weights are configured based on the violation type and the domain to which the potential slot belongs. Violation types include unclosed violations, violations exceeding limits, and violations requiring proof of contradiction. Potential slot domains include performance potential slots, asset potential slots, and energy potential slots. Unclosed violations are matched with unclosed penalty weights, violations exceeding limits are matched with excess penalty weights, and violations requiring proof of contradiction are matched with proof of contradiction penalty weights.

[0171] When generating node clearing violation penalties, the corresponding preset penalty weights are matched for unclosed violations, excessive violations, and counter-evidence violations within the same matrix cell. The matched preset penalty weights are then accumulated within the matrix cell to obtain the node clearing violation penalty for the current lease cycle node.

[0172] When accumulating penalties for defaults in the lease cycle, select the matrix units corresponding to the same candidate risk control action number, and arrange the penalties for defaults in the order of payment node, operation node, energy replenishment node, maintenance node, and recovery node. Then, sum the penalty amounts for defaults in the order of these nodes to obtain the penalty for defaults corresponding to that candidate risk control action number.

[0173] The penalty for violating the clearing rules is recorded and saved in conjunction with the candidate risk control action number.

[0174] In this embodiment, the operations for generating the target risk control action chain include:

[0175] The shadow potential state where the payment occupancy bit, asset occupancy bit, and energy replenishment occupancy bit are not written into the candidate risk control action number is determined as the baseline shadow potential state.

[0176] The benchmark shadow lease trajectory is obtained by performing continuous state transitions on the benchmark shadow potential state in the order of lease cycle nodes using the improved Dreamer model.

[0177] Accumulated liquidation violations are applied to the risk release, risk retention, and risk enhancement positions in the benchmark shadow leasing trajectory to obtain benchmark liquidation violation penalties.

[0178] Compare the penalty for each candidate risk control action number corresponding to the clearing violation with the benchmark clearing violation penalty;

[0179] Candidate risk control actions with a comparison result greater than or equal to the benchmark clearing violation penalty are marked as blocked actions, while candidate risk control actions with a comparison result less than the benchmark clearing violation penalty are marked as reserved actions.

[0180] Arrange the retained actions according to the order of the lease cycle nodes and the order of the clearing item numbers corresponding to the retained actions, and generate the target risk control action chain.

[0181] Among them, the candidate risk control action number is not written in the payment occupancy position, asset occupancy position and energy replenishment occupancy position. This means that the candidate risk control action number in the payment occupancy position, asset occupancy position and energy replenishment occupancy position is empty, and the corresponding occupancy identifier is 0.

[0182] When determining the baseline shadow potential state, the three-domain clearing potential state is copied. The copied performance potential slots, asset potential slots, and energy potential slots retain their original slot order, while the payment occupancy slot, asset occupancy slot, and energy replenishment occupancy slot remain unoccupied. The copied state is then marked as the baseline shadow potential state. The baseline shadow potential state is not bound to performance-related candidate risk control actions, asset-related candidate risk control actions, or energy replenishment candidate risk control actions.

[0183] When performing continuous state transitions using the improved Dreamer model, the baseline shadow potential state is used as the initial input state, and baseline shadow trajectory states are generated node by node in the order of payment node, operation node, energy replenishment node, maintenance node, and recovery node. Each baseline shadow trajectory state retains the performance potential slot, asset potential slot, energy potential slot, risk retention slot, risk release slot, and risk enhancement slot under the corresponding lease cycle node.

[0184] Upon obtaining the baseline shadow leasing trajectory, the statuses of the baseline shadow trajectories corresponding to payment nodes, operation nodes, energy replenishment nodes, maintenance nodes, and recovery nodes are connected sequentially to form the baseline shadow leasing trajectory. The baseline shadow leasing trajectory serves as a comparison benchmark when no candidate risk control actions are initiated.

[0185] When performing clearing and cumulative violation calculations on the benchmark shadow lease trajectory, the risk release position, risk retention position, and risk enhancement position in the benchmark shadow trajectory status are processed in the order of lease cycle nodes. Closed items in the risk release position form a benchmark closed coverage relationship; unclosed items in the risk retention position that are not covered by the benchmark closed coverage relationship form benchmark unclosed violation items; over-limit items in the risk enhancement position form benchmark over-limit violation items; and counter-evidence items in the risk enhancement position form benchmark counter-evidence violation items.

[0186] When generating benchmark clearing violation penalties, benchmark unclosed violations, benchmark over-limit violations, and benchmark counter-evidence violations are matched with corresponding preset penalty weights. Benchmark node clearing violation penalties are generated according to the lease cycle nodes, and then benchmark node clearing violation penalties are accumulated in the order of the lease cycle nodes to obtain the benchmark clearing violation penalties.

[0187] When comparing the clearing violation penalty corresponding to each candidate risk control action number with the benchmark clearing violation penalty, the candidate risk control action number is selected one by one, the clearing violation penalty corresponding to the candidate risk control action number is retrieved, and the clearing violation penalty is compared with the benchmark clearing violation penalty.

[0188] If the penalty for a clearance violation corresponding to a candidate risk control action number is not less than the benchmark clearance violation penalty, the candidate risk control action number is marked as a blocked action, and the binding relationship between the candidate risk control action number and the corresponding clearance item number is maintained. If the penalty for a clearance violation corresponding to a candidate risk control action number is less than the benchmark clearance violation penalty, the candidate risk control action number is marked as a reserved action, and the binding relationship between the candidate risk control action number, the corresponding clearance item number, and the lease cycle node is maintained.

[0189] When arranging reserved actions according to the order of the lease cycle nodes and the order of the clearing item numbers corresponding to the reserved actions, the reserved actions are first arranged in the order of payment node, operation node, energy replenishment node, maintenance node, and recycling node; when there are multiple reserved actions under the same lease cycle node, they are arranged in the order of the clearing item numbers corresponding to the reserved actions.

[0190] When generating the target risk control action chain, the arranged reserved actions are connected sequentially, and action sequence identifiers are registered between adjacent reserved actions. The target risk control action chain includes candidate risk control action number, clearing item number, lease cycle node, reserved action identifier, and action sequence identifier.

[0191] In this implementation, the operation of outputting the comprehensive leasing risk level includes:

[0192] Load the reserved actions in the target risk control action chain into the baseline shadow potential state according to the order of the chain;

[0193] According to the candidate risk control actions of performance, asset and energy replenishment categories to which the reserved action belongs, update the payment occupancy bit, asset occupancy bit and energy replenishment occupancy bit in the benchmark shadow potential state respectively.

[0194] The target chain shadow leasing trajectory is obtained by performing continuous state transitions on the baseline shadow latent state after updating the occupied bits using the improved Dreamer model.

[0195] Accumulated liquidation violations are performed on the risk release, risk retention, and risk enhancement positions in the target chain shadow leasing trajectory to obtain the target chain liquidation violation penalty.

[0196] The penalty for breach of target chain clearing is compared with the preset risk level threshold table to determine the overall leasing risk level;

[0197] The risk level of the comprehensive lease is associated with the target risk control action chain, the penalty for violation of the target chain clearing, and the corresponding clearing item number.

[0198] The target risk control action chain consists of reserved actions arranged in the order of lease cycle nodes and clearing item numbers. Each reserved action corresponds to a candidate risk control action number, a clearing item number, a lease cycle node, and an action sequence identifier.

[0199] When loading reserved actions into the baseline shadow potential state, reserved actions in the target risk control action chain are selected sequentially according to the action sequence identifier. The candidate risk control action number, clearing item number, lease cycle node, and candidate risk control action type corresponding to the reserved action are then loaded into the baseline shadow potential state. During the loading process, the original correspondence between the performance potential slots, asset potential slots, energy potential slots, risk retention slots, risk release slots, and risk enhancement slots in the baseline shadow potential state remains unchanged.

[0200] When updating the payment occupancy slot, select the reserved action that is a candidate risk control action of the performance type, locate the performance potential slot with the same clearing item number in the baseline shadow potential state, write the candidate risk control action number corresponding to the reserved action into the payment occupancy slot, and set the occupancy flag of the payment occupancy slot to 1.

[0201] When updating the asset occupancy slot, select the reserved action that is the candidate risk control action type of asset-type candidate risk control action, locate the asset potential slot with the same clearing item number in the benchmark shadow potential state, write the candidate risk control action number corresponding to the reserved action into the asset occupancy slot, and set the occupancy flag of the asset occupancy slot to 1.

[0202] When updating the energy replenishment occupancy bit, select the reserved action of the candidate risk control action type of energy replenishment, locate the energy potential slot with the same clearing item number in the baseline shadow potential state, write the candidate risk control action number corresponding to the reserved action into the energy replenishment occupancy bit, and set the occupancy flag of the energy replenishment occupancy bit to 1.

[0203] When multiple reserved actions exist in the same risk control action chain, the payment occupancy position, asset occupancy position, and energy replenishment occupancy position are updated sequentially according to the order in the chain. After all reserved actions are loaded, the baseline shadow potential state after updating the occupancy positions is formed.

[0204] When performing continuous state transitions using the improved Dreamer model, the baseline shadow potential state after updating the occupied slots is used as the initial input state. The shadow trajectory states corresponding to each lease cycle node are generated in the order of payment node, operation node, energy replenishment node, maintenance node, and recycling node. Each lease cycle node inherits the shadow trajectory state output by the previous lease cycle node and retains the three types of potential slots, three types of occupied slots, and three types of risk slots under the current node.

[0205] Upon obtaining the target chain shadow leasing trajectory, the shadow trajectory states corresponding to payment nodes, operation nodes, replenishment nodes, maintenance nodes, and recycling nodes are connected sequentially according to the leasing cycle nodes. The target chain shadow leasing trajectory maintains a correspondence with the target risk control action chain, and registers the target risk control action chain identifier, leasing cycle nodes, occupancy status, and risk status.

[0206] When performing clearing and cumulative violation processing on the target chain shadow leasing trajectory, the risk release positions, risk retention positions, and risk enhancement positions in the target chain shadow leasing trajectory are processed in the order of the leasing cycle nodes. Closed terms in the risk release positions form a closed coverage relationship in the target chain; unclosed terms in the risk retention positions that are not covered by the closed coverage relationship in the target chain form unclosed violation terms in the target chain; excessive terms in the risk enhancement positions form excessive violation terms in the target chain; and contradictory terms in the risk enhancement positions form contradictory violation terms in the target chain.

[0207] When generating the target chain clearing violation penalty, the target chain unclosed violation, target chain over-limit violation, and target chain counter-evidence violation are matched with the corresponding preset penalty weights. The target chain node clearing violation penalty is generated according to the lease cycle node, and then the target chain node clearing violation penalty is accumulated according to the lease cycle node order to obtain the target chain clearing violation penalty.

[0208] The preset risk level threshold table includes a risk level, a lower threshold, an upper threshold, and a level identifier. Risk levels include low risk, high risk, restricted risk, and high risk. Each risk level corresponds to a threshold range, and these ranges do not overlap.

[0209] When comparing the penalty for breach of the target chain clearing with the preset risk level threshold table, the penalty for breach of the target chain clearing is compared with the lower and upper limits of the threshold corresponding to each risk level. When the penalty for breach of the target chain clearing falls within a certain threshold range, the risk level corresponding to that threshold range is determined as the comprehensive leasing risk level.

[0210] When outputting related data, the overall lease risk level, target risk control action chain, penalty for violation of target chain clearing, and corresponding clearing item number are written into the same risk assessment output record. The risk assessment output record registers the lease period, target risk control action chain, penalty for violation of target chain clearing, overall lease risk level, and corresponding clearing item number.

[0211] In this embodiment, the operation of refreshing the potential status of the three-domain clearing for the next lease period includes:

[0212] Execute the target risk control action chain and receive leasing feedback data;

[0213] Based on the candidate risk control action number and the clearing item number, the leasing feedback data will be written back to the corresponding clearing items in the performance clearing ledger, vehicle asset clearing ledger and energy supply clearing ledger respectively;

[0214] Based on the rewritten performance settlement ledger, vehicle asset settlement ledger and energy supply settlement ledger, the unclosed items, closed items, over-limit items and counter-evidence items were re-evaluated;

[0215] The risk retention position, risk release position, and risk enhancement position are updated according to the reassessment results to form the potential three-domain liquidation status for the next lease cycle.

[0216] When executing the target risk control action chain, the reserved actions are executed sequentially according to the action sequence identifier in the target risk control action chain, and a target risk control action execution record is generated. The target risk control action execution record registers the candidate risk control action number, clearing item number, lease cycle node, action sequence identifier, and execution status identifier.

[0217] When receiving leasing feedback data, the candidate risk control action numbers in the leasing feedback data are matched with the candidate risk control action numbers in the target risk control action execution record, and the clearing item numbers in the leasing feedback data are matched with the clearing item numbers in the target risk control action execution record. If both the candidate risk control action numbers and the clearing item numbers match, the leasing feedback data is identified as the feedback data for the corresponding reserved action.

[0218] When leasing feedback data corresponds to a performance-related candidate risk control action, the leasing feedback data is written back to the performance clearing item corresponding to the same clearing item number in the performance clearing ledger. The leasing feedback data corresponding to the performance-related candidate risk control action includes payment feedback flags, deposit feedback flags, guarantee feedback flags, and vehicle return feedback flags.

[0219] When leasing feedback data corresponds to candidate risk control actions for asset categories, the leasing feedback data is written back to the vehicle asset clearing item with the same clearing item number in the vehicle asset clearing ledger. Leasing feedback data corresponding to candidate risk control actions for asset categories includes load limit feedback indicators, operating area restriction feedback indicators, autonomous driving review feedback indicators, battery status feedback indicators, vehicle fault diagnosis feedback indicators, and maintenance execution feedback indicators.

[0220] When leasing feedback data corresponds to a candidate risk control action for energy replenishment, the leasing feedback data is written back to the energy replenishment clearing item with the same clearing item number in the energy replenishment clearing ledger. The leasing feedback data corresponding to the candidate risk control action for energy replenishment includes charging adjustment feedback flags, battery swapping adjustment feedback flags, energy replenishment pending processing feedback flags, photovoltaic energy supply allocation feedback flags, energy storage support allocation feedback flags, and energy replenishment execution feedback flags.

[0221] When writing leasing feedback data back to the corresponding clearing items, update the relevant feedback fields in the corresponding clearing items, and retain the original record number, clearing item number, and clearing item type of the corresponding clearing item. The corresponding clearing items include performance clearing items, vehicle asset clearing items, and energy supply clearing items.

[0222] When re-evaluating unclosed items, closed items, over-limit items, and counter-evidence items, the execution status of each of the written-back performance settlement items, vehicle asset settlement items, and energy supply settlement items is determined.

[0223] If the corresponding clearing item does not receive the corresponding feedback field, the corresponding feedback field is a pending execution flag, or the corresponding feedback field is a processing flag, the corresponding clearing item will be reclassified as an unclosed item.

[0224] When the corresponding feedback field of a clearing item meets the clearing item agreement, the corresponding clearing item will be re-determined as a closed item.

[0225] When the collected value of a corresponding clearing item exceeds the clearing item constraint, the corresponding clearing item will be re-determined as an over-limit item.

[0226] If the corresponding feedback field of a clearing item is contrary to the clearing item agreement, the corresponding clearing item will be re-evaluated as a counter-evidence item.

[0227] When updating the risk retention position, risk release position, and risk enhancement position based on the reassessment results, the corresponding potential performance slot, potential asset slot, and potential energy slot are located using the performance settlement item number, vehicle asset settlement item number, and energy supply settlement item number as slot indexes.

[0228] When the reassessment result is an unclosed item, the unclosed item is written to the risk reservation position of the corresponding potential slot, and the old closure identifier of the same clearing item number in the risk release position is replaced with the reassessment result.

[0229] When the reassessment result is a closed item, the closed item is written into the risk release position of the corresponding potential slot, and the old unclosed identifier of the same clearing item number in the risk retention position is replaced with the reassessment result.

[0230] When the reassessment result is an item exceeding the limit, the item exceeding the limit is written into the risk enhancement bit of the corresponding potential slot, forming an item exceeding the limit identifier corresponding to the clearing item number.

[0231] When the reassessment result is a counter-evidence item, the counter-evidence item is written into the risk enhancement bit of the corresponding potential slot, forming a counter-evidence item identifier corresponding to the clearing item number.

[0232] When determining the three-domain clearance potential status for the next lease cycle, the updated risk retention, risk release, and risk enhancement positions are arranged in the domain order of performance potential slots, asset potential slots, and energy potential slots. The arranged performance potential slots, asset potential slots, and energy potential slots together constitute the three-domain clearance potential status for the next lease cycle.

[0233] Example 1: To verify the feasibility of this invention in practice, it was applied to an electric heavy-duty truck leasing operation scenario. In this scenario, there is a continuous leasing relationship between the lessee, the vehicle, the battery pack, and the charging station. The operation system generates contract payment records, vehicle driving records, load records, work area records, unmanned takeover records, battery status records, charging and battery swapping records, maintenance feedback records, and vehicle recovery feedback records. Existing static scoring methods use individual indicators such as overdue payments, abnormal location, and abnormal maintenance to form risk scores. Incomplete payment closures, vehicle overuse, and abnormal charging are easily scattered across different records. Manual reviewers need to repeatedly check multiple business tables, which can easily lead to missed risk identification and duplicate risk control actions.

[0234] In application scenarios, the system aggregates leasing business data, vehicle operation data, energy replenishment data, and leasing feedback data according to the leasing period, forming a basic record table for leasing risks. The system then organizes performance clearing ledgers, vehicle asset clearing ledgers, and energy replenishment clearing ledgers from this basic record table. The performance clearing ledger records clearing items corresponding to rent, deposits, guarantees, and vehicle recovery; the vehicle asset clearing ledger records clearing items corresponding to mileage, load, operating area, autonomous driving anomalies, battery status, vehicle malfunctions, and maintenance; and the energy replenishment clearing ledger records clearing items corresponding to charging, battery swapping, waiting, photovoltaic power supply, energy storage support, and energy replenishment feedback. These three types of ledgers are then categorized by clearing item number into performance potential slots, asset potential slots, and energy potential slots. Unclosed items are placed in risk retention slots, closed items in risk release slots, and items exceeding limits and those providing counter-evidence in risk enhancement slots. After this processing, the risk status scattered across business records, vehicle records, and energy replenishment records is organized into three domains of potential clearing status.

[0235] During the risk control action generation process, the system identifies action trigger slots from risk retention and risk enhancement positions, and generates candidate risk control actions for performance, assets, and replenishment. Each candidate risk control action is bound to a candidate risk control action number and a clearing item number. The system replicates a set of three-domain clearing potential states for each candidate risk control action, forming a shadow potential state. After reading the shadow potential state, the improved Dreamer model writes the candidate risk control action into the payment occupancy position, asset occupancy position, and replenishment occupancy position, and generates a shadow leasing trajectory along the payment node, operation node, replenishment node, maintenance node, and recovery node. The system accumulates clearing violation penalties for unclosed items, over-limit items, and counter-evidence items in the shadow leasing trajectory, and compares them with the baseline clearing violation penalties before the candidate risk control action was executed. Candidate risk control actions with reduced penalties enter the target risk control action chain, while candidate risk control actions with increased penalties are blocked. After the target risk control action chain is executed, the system receives leasing feedback data, writes the feedback data back to the three types of clearing ledgers, re-determines unclosed items, closed items, over-limit items and counter-evidence items, and refreshes the potential clearing status of the three domains for the next leasing cycle.

[0236] This verification sample included 126 electric heavy-duty trucks, 58 lessees, 184 rental cycle records, 3126 vehicle operation records, 1498 energy replenishment records, and 736 rental feedback records. Existing static scoring methods were used as a control, while the method of this invention was used as a test method. Both methods used the same input data, and manual verification conclusions served as the validation benchmark. Test results show that the method of this invention outperforms existing static scoring methods in terms of risk identification accuracy, completeness of unclosed item identification, completeness of out-of-limit item identification, completeness of counter-evidence item identification, completeness of feedback write-back, and success rate of next cycle status refresh. It also outperforms existing static scoring methods in terms of the proportion of invalid risk control actions, the proportion of duplicate risk control actions, the number of manual reviews, and the average evaluation time. The comparison results are shown in the table below:

[0237] Table 1: Comparison of Risk Assessment Effectiveness

[0238] Risk identification accuracy 78.6% 92.4% Completeness of unclosed item recognition 74.8% 94.1% Completeness of out-of-limit item recognition 76.3% 93.6% Completeness of counter-evidence identification 69.5% 90.8% Early detection rate of vehicle recycling anomalies 71.2% 88.7% Energy replenishment abnormal closure rate 73.9% 91.2% Consistency rate of comprehensive leasing risk levels 80.1% 93.5% Target risk control action closure rate 67.8% 89.4% Feedback write-back completeness rate 76.5% 96.2% Success rate of status update for the next rental period 72.4% 95.7% Percentage of Ineffective Risk Control Actions 26.8% 9.6% Percentage of repetitive risk control actions 18.4% 6.1% Number of high-risk lessees not identified 17 articles 5 items Number of manual reviews 43 times / 100 records 16 times / 100 records Average evaluation time 38.6 minutes / batch 14.8 minutes / batch Average penalty for breach of accounting 42.7 points 24.3 points

[0239] The data in the table shows that the method of this invention outperforms existing static scoring methods in risk identification indicators. The accuracy rate of risk identification increased from 78.6% to 92.4%, an increase of 13.8 percentage points; the completeness rate of identifying incomplete payments increased from 74.8% to 94.1%, an increase of 19.3 percentage points; the completeness rate of identifying over-limit items increased from 76.3% to 93.6%, an increase of 17.3 percentage points; and the completeness rate of identifying counter-evidence items increased from 69.5% to 90.8%, an increase of 21.3 percentage points. These data demonstrate that the method of this invention can integrate incomplete payment, vehicle over-limit, abnormal energy replenishment, and counter-evidence feedback into the same risk status structure for processing.

[0240] In terms of closed-loop processing of leasing business, the method of this invention also shows significant improvement. The early identification rate of vehicle recovery anomalies increased from 71.2% to 88.7%, an increase of 17.5 percentage points; the closure rate of charging anomalies increased from 73.9% to 91.2%, an increase of 17.3 percentage points; the consistency rate of comprehensive leasing risk levels increased from 80.1% to 93.5%, an increase of 13.4 percentage points; and the closure rate of target risk control actions increased from 67.8% to 89.4%, an increase of 21.6 percentage points. These results indicate that there is a high degree of matching between the target risk control action chain and the potential state of the three-domain clearing.

[0241] Regarding feedback updates, the completeness rate of feedback write-back increased from 76.5% to 96.2%, an increase of 19.7 percentage points; the success rate of status refresh for the next lease cycle increased from 72.4% to 95.7%, an increase of 23.3 percentage points. This data indicates that lease feedback data can be written back relatively completely to the performance settlement ledger, vehicle asset settlement ledger, and energy replenishment settlement ledger, and the risk retention, risk release, and risk enhancement positions for the next lease cycle can be continuously refreshed.

[0242] Regarding the efficiency of risk control actions, the proportion of invalid risk control actions decreased from 26.8% to 9.6%, a reduction of 17.2 percentage points; the proportion of duplicate risk control actions decreased from 18.4% to 6.1%, a reduction of 12.3 percentage points; the number of high-risk lessees missed was reduced from 17 to 5, a reduction of 12; the number of manual reviews decreased from 43 times / 100 records to 16 times / 100 records, a reduction of 27; the average assessment time decreased from 38.6 minutes / batch to 14.8 minutes / batch, a reduction of 23.8 minutes; and the average penalty for breach of clearing rules decreased from 42.7 points to 24.3 points, a reduction of 18.4 points. The data shows that the method of this invention can reduce invalid judgments and duplicate processing, thereby improving the efficiency of lease risk assessment.

[0243] The above description is only a preferred embodiment 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 risk assessment method for electric heavy-duty truck leasing based on big data analysis and reinforcement learning, characterized in that, Includes the following steps: Acquire electric heavy-duty truck leasing operation data according to the leasing period and generate a basic record table of leasing risks; Based on the lease risk basic record form, generate performance settlement ledgers, vehicle asset settlement ledgers, and energy supply settlement ledgers; Establish potential performance slots, potential asset slots, and potential energy slots according to the clearing item number. Write unclosed items in the three types of ledgers into the risk retention position, closed items into the risk release position, and over-limit items and counter-evidence items into the risk enhancement position to form a three-domain clearing potential status. Read the risk retention bit and risk enhancement bit, enumerate candidate risk control actions, and copy a set of three-domain clearing potential states for each candidate risk control action to form a shadow potential state. In the improved Dreamer model, the payment occupancy, asset occupancy, and energy replenishment occupancy in the shadow potential state are rewritten according to the candidate risk control actions, and the shadow leasing trajectory is obtained by rolling. By comparing the risk retention position, risk release position and risk enhancement position in the shadow leasing trajectory, for the cumulative clearing violation penalty of unclosed, over-limit and counter-evidence, the clearing violation penalty before the candidate risk control action is executed is used as the benchmark, the candidate risk control action with increased penalty is blocked, the candidate risk control action with reduced penalty is retained, the target risk control action chain is generated and the comprehensive leasing risk level is output. After executing the target risk control action chain, read the leasing feedback data, write the leasing feedback data back to the three types of ledgers, and refresh the potential status of the three-domain clearing in the next leasing cycle.

2. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 1, characterized in that, The operation of generating the basic lease risk record table includes: Read leasing business data, vehicle operation data, energy replenishment data, and leasing feedback data according to the leasing period; Using lessee, vehicle, battery pack, charging station, and lease period as the aggregation index, data belonging to the same aggregation index are merged into a basic record of lease risk; Configure record numbers according to the generation order of the lease risk basic records to form a lease risk basic record table.

3. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 2, characterized in that, The operations for generating the performance settlement ledger, vehicle asset settlement ledger, and energy supply settlement ledger include: Read the leasing business field, vehicle operation field, energy supply field, and leasing feedback field from the basic leasing risk record table by record number; Register the data corresponding to the lessee's performance in the lease business field and lease feedback field as performance settlement items to form a performance settlement ledger; Register the data corresponding to vehicle asset occupancy in the vehicle operation field and leasing feedback field as vehicle asset clearing items to form a vehicle asset clearing ledger. The data corresponding to the energy replenishment resource occupation in the energy replenishment field and the leasing feedback field are registered as energy replenishment clearing items, forming an energy replenishment clearing ledger.

4. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 3, characterized in that, The operations that form the potential state of the three-domain clearing include: Using the settlement item numbers of performance settlement items, vehicle asset settlement items, and energy supply settlement items as slot indexes, potential performance slots, potential asset slots, and potential energy slots are established respectively; Within each potential slot, set up risk retention positions, risk release positions, and risk enhancement positions; Data for which no corresponding lease feedback field has been obtained or where the corresponding lease feedback field is incomplete is marked as an unclosed item and written to the risk reserved bit; Data that meets the settlement agreement in the corresponding lease feedback field is marked as a closed item and written into the risk release position; data whose collected value exceeds the settlement constraint is marked as an over-limit item; data whose lease feedback field is contrary to the settlement agreement is marked as a counter-evidence item; over-limit items and counter-evidence items are written into the risk enhancement position. The risk retention position, risk release position, and risk enhancement position are arranged in the domain order of performance potential slots, asset potential slots, and energy potential slots to form a three-domain clearing potential state.

5. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 4, characterized in that, The operations for forming the shadow latent state include: In the potential states of the three-domain clearing, potential slots for writing unclosed items, over-limit items, or counter-evidence items are selected as action trigger slots. Based on the performance potential slot, asset potential slot, or energy potential slot to which the action trigger slot belongs, generate corresponding performance-related candidate risk control actions, asset-related candidate risk control actions, or energy replenishment-related candidate risk control actions. Configure candidate risk control action numbers for candidate risk control actions, and bind the candidate risk control action numbers to the clearing item numbers of the action trigger slots; Copy the three domain clearing potential states according to the candidate risk control action number to obtain the shadow potential state corresponding to the candidate risk control action number; The correspondence between performance potential slots, asset potential slots, energy potential slots, risk retention slots, risk release slots, and risk enhancement slots is preserved in the shadow potential state.

6. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 5, characterized in that, The operation of obtaining the shadow rental trajectory by scrolling includes: Input the shadow potential state corresponding to the candidate risk control action number into the improved Dreamer model; In the shadow potential state, set the payment occupancy position corresponding to the performance potential slot, the asset occupancy position corresponding to the asset potential slot, and the energy replenishment occupancy position corresponding to the energy potential slot; According to the clearing item number bound to the candidate risk control action number, write the performance-related candidate risk control action into the payment occupancy position, write the asset-related candidate risk control action into the asset occupancy position, and write the energy replenishment-related candidate risk control action into the energy replenishment occupancy position. The improved Dreamer model performs continuous state transitions on the shadow potential state after writing the occupied bit according to the lease cycle node order, and obtains the shadow trajectory state corresponding to each lease cycle node. By connecting the shadow trajectory status according to the candidate risk control action number, a shadow leasing trajectory is formed.

7. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 6, characterized in that, The operations for cumulative clearing violation penalties include: Establish a clearing violation matrix based on candidate risk control action numbers and lease cycle nodes; Based on the closed term of the risk release position in the shadow trajectory state, generate the closed coverage relationship of the corresponding clearing item number in the clearing violation matrix; Compare the unclosed items in the risk reservation position with the closed coverage relationship, and identify the unclosed items that are not covered by the closed coverage relationship as unclosed violation items; The out-of-limit items in the risk-enhancing positions are identified as out-of-limit violations, and the counter-evidence items in the risk-enhancing positions are identified as counter-evidence violations. Preset penalty weights are matched for unclosed violations, excessive violations, and violations that are proven by contradiction, and node clearing violation penalties are generated for each lease cycle node. Accumulate the clearing violation penalties for the same candidate risk control action number according to the order of the lease cycle nodes to obtain the clearing violation penalty corresponding to the candidate risk control action number.

8. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 7, characterized in that, The operations for generating the target risk control action chain include: The shadow potential state where the payment occupancy bit, asset occupancy bit, and energy replenishment occupancy bit are not written into the candidate risk control action number is determined as the baseline shadow potential state. The benchmark shadow lease trajectory is obtained by performing continuous state transitions on the benchmark shadow potential state in the order of lease cycle nodes using the improved Dreamer model. Accumulated liquidation violations are applied to the risk release, risk retention, and risk enhancement positions in the benchmark shadow leasing trajectory to obtain benchmark liquidation violation penalties. Compare the penalty for each candidate risk control action number corresponding to the clearing violation with the benchmark clearing violation penalty; Candidate risk control actions with a comparison result greater than or equal to the benchmark clearing violation penalty are marked as blocked actions, while candidate risk control actions with a comparison result less than the benchmark clearing violation penalty are marked as reserved actions. Arrange the retained actions according to the order of the lease cycle nodes and the order of the clearing item numbers corresponding to the retained actions, and generate the target risk control action chain.

9. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 8, characterized in that, The operation of outputting the comprehensive leasing risk level includes: Load the reserved actions in the target risk control action chain into the baseline shadow potential state according to the order of the chain; According to the candidate risk control actions of performance, asset and energy replenishment categories to which the reserved action belongs, update the payment occupancy bit, asset occupancy bit and energy replenishment occupancy bit in the benchmark shadow potential state respectively. The target chain shadow leasing trajectory is obtained by performing continuous state transitions on the baseline shadow latent state after updating the occupied bits using the improved Dreamer model. Accumulated liquidation violations are performed on the risk release, risk retention, and risk enhancement positions in the target chain shadow leasing trajectory to obtain the target chain liquidation violation penalty. The penalty for breach of target chain clearing is compared with the preset risk level threshold table to determine the overall leasing risk level; The risk level of the comprehensive lease is associated with the target risk control action chain, the penalty for violation of the target chain clearing, and the corresponding clearing item number.

10. The method for risk assessment of electric heavy-duty truck leasing based on big data analysis and reinforcement learning according to claim 9, characterized in that, The operation to refresh the potential status of the three-domain clearing for the next lease cycle includes: Execute the target risk control action chain and receive leasing feedback data; Based on the candidate risk control action number and the clearing item number, the leasing feedback data will be written back to the corresponding clearing items in the performance clearing ledger, vehicle asset clearing ledger and energy supply clearing ledger respectively; Based on the rewritten performance settlement ledger, vehicle asset settlement ledger and energy supply settlement ledger, the unclosed items, closed items, over-limit items and counter-evidence items were re-evaluated; The risk retention position, risk release position, and risk enhancement position are updated according to the reassessment results to form the potential three-domain liquidation status for the next lease cycle.

Images