Processing scheme determination method and device based on selection probability, equipment and medium

Abstract

The present application relates to the technical field of intelligent decision-making, and discloses a processing scheme determination method and device based on selection probability, equipment and medium, comprising: constructing a processing scheme set based on target loss information of a to-be-processed object; generating candidate scheme feature data and feature field structure; inputting the candidate scheme feature data into a preset discrete selection model to generate utility scores and selection probabilities, and determining a target processing scheme; generating actual scheme feature data based on actual processing scheme, actual resource consumption data and feature field structure to obtain a rationality verification result. The present application can be applied to business scenarios such as financial technology and medical health, unifies the data caliber of candidate scheme feature data and actual scheme feature data through the feature field structure, and determines the target processing scheme through the discrete selection model, so that the resource consumption rationality verification can undertake the scheme recommendation process, thereby reducing the degree of fragmentation between recommendation and verification.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology, and in particular to a method, apparatus, device, and medium for determining processing schemes based on selection probability. Background Technology

[0002] In loss handling operations, for damage to vehicles, property, equipment, business facilities, or service resources, it is usually necessary to determine a handling solution from multiple candidate solutions and verify the reasonableness of actual resource consumption data. Existing methods often rely on human experience, fixed judgment conditions, or historical average ranges. There is a lack of unified data standards between the generation of candidate handling solutions, feature evaluation, and verification of actual resource consumption, making it difficult to seamlessly connect the handling solution recommendation process with the actual resource consumption verification process.

[0003] In the fintech business, claims review, loss assessment review, risk control review, and fund disbursement review typically require determining whether the processing plan matches the actual resource consumption data for the same loss event. Current processing methods, when faced with multiple candidate processing plans, often separate plan selection from resource consumption verification. The loss information, resource sources, processing time, and resource consumption composition on which the candidate processing plans are based are difficult to reconcile with the field structure of the verification data for the actual processing plan, affecting the consistency between the recommended results and the verification results for the same loss event.

[0004] In the healthcare sector, when medical equipment, rehabilitation aids, nursing facilities, or health service resources are damaged, it is necessary to select from candidate treatment options such as repair, replacement, or partial repair, and to review actual resource consumption data. Current methods typically rely on manual judgment or historical records to handle option selection and resource consumption review separately. This lacks a unified feature base that connects the characteristics of candidate treatment options with the characteristics of actual treatment options, making it easy for the verification of the reasonableness of actual resource consumption data to lack an evaluation caliber consistent with that of candidate treatment options. Summary of the Invention

[0005] The main objective of this invention is to provide a method, apparatus, device, and storage medium for determining a processing scheme based on selection probability. This aims to solve the technical problem in the prior art where there is no unified evaluation basis for recommending processing schemes and verifying resource consumption for the same loss event, which makes it difficult to continuously connect the recommendation results and the verification results.

[0006] To achieve the above objectives, the present invention provides a method for determining a processing scheme based on selection probability, comprising: Obtain the target loss information of the object to be processed, and construct a set of processing schemes containing multiple candidate processing schemes based on the target loss information; Based on the set of processing schemes and the target loss information, candidate scheme feature data corresponding to each candidate processing scheme is generated, and the feature field structure is determined based on the candidate scheme feature data; The feature data of the candidate solutions are input into a preset discrete selection model to generate a utility score for each candidate processing solution. Based on the utility scores corresponding to all candidate processing schemes, a selection probability is generated for each candidate processing scheme, and a target processing scheme is determined from the set of processing schemes based on the selection probability. Obtain the actual processing plan and actual resource consumption data corresponding to the object to be processed, and generate actual plan feature data based on the actual processing plan, the actual resource consumption data, the target loss information and the feature field structure; Based on the actual scheme feature data, the rationality of resource consumption is verified, the rationality verification result of the actual resource consumption data is generated, and the rationality verification result of the target processing scheme and the actual resource consumption data is output.

[0007] Furthermore, to achieve the above objectives, the present invention provides a processing scheme determination apparatus based on selection probability, comprising: The processing scheme set construction module is used to obtain the target loss information of the object to be processed, and construct a processing scheme set containing multiple candidate processing schemes based on the target loss information; The candidate solution feature generation module is used to generate candidate solution feature data corresponding to each candidate processing solution based on the processing solution set and the target loss information, and to determine the feature field structure based on the candidate solution feature data; The utility score generation module is used to input the feature data of the candidate schemes into a preset discrete selection model to generate a utility score corresponding to each candidate processing scheme. The target processing scheme determination module is used to generate the selection probability of each candidate processing scheme based on the utility scores corresponding to all candidate processing schemes, and determine the target processing scheme from the set of processing schemes based on the selection probability; The actual solution feature generation module is used to obtain the actual processing solution and actual resource consumption data corresponding to the object to be processed, and generate actual solution feature data based on the actual processing solution, the actual resource consumption data, the target loss information and the feature field structure; The resource consumption verification output module is used to verify the rationality of resource consumption based on the actual scheme feature data, generate the rationality verification result of the actual resource consumption data, and output the rationality verification result of the target processing scheme and the actual resource consumption data.

[0008] Furthermore, to achieve the above objectives, the present invention also provides a computer device, the computer device including a memory, a processor, and a selection probability-based processing scheme determination program stored in the memory and executable on the processor, wherein when the selection probability-based processing scheme determination program is executed by the processor, it implements the steps of the selection probability-based processing scheme determination method as described above.

[0009] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a probabilistic-based processing scheme determination program, wherein when the probabilistic-based processing scheme determination program is executed by a processor, it implements the steps of the probabilistic-based processing scheme determination method described above.

[0010] Beneficial Effects: This invention relates to the field of intelligent decision-making technology, and discloses a method, apparatus, device, and medium for determining processing schemes based on selection probability. The method includes: acquiring target loss information of the object to be processed, and constructing a processing scheme set containing multiple candidate processing schemes; generating candidate scheme feature data based on the processing scheme set and target loss information, and determining the feature field structure; inputting the candidate scheme feature data into a preset discrete selection model to generate utility scores and selection probabilities, and determining the target processing scheme; generating actual scheme feature data based on the actual processing scheme corresponding to the object to be processed, actual resource consumption data, target loss information, and feature field structure; verifying the rationality of resource consumption based on the actual scheme feature data, and generating and outputting the rationality verification result of the actual resource consumption data. This invention can be applied to business scenarios such as fintech and healthcare. It establishes a unified evaluation basis through candidate scheme feature data and feature field structure, generates utility scores and selection probabilities through a discrete selection model to determine the target processing scheme, and uses the same feature field structure for verifying the rationality of resource consumption in the actual scheme feature data, thereby reducing the disconnect between processing scheme recommendation and resource consumption verification. Attached Figure Description

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a schematic diagram of an application environment for a processing scheme determination method based on selection probability according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an embodiment of the method for determining a processing scheme based on selection probability according to the present invention. Figure 3 for Figure 2 A detailed flowchart of step S10; Figure 4 for Figure 2 A detailed flowchart of step S20; Figure 5 for Figure 2 A detailed flowchart of step S30; Figure 6 for Figure 2 A detailed flowchart of step S40; Figure 7 for Figure 2 A detailed flowchart of step S50; Figure 8 for Figure 2 A detailed flowchart of step S60; Figure 9 This is a schematic diagram of the functional modules of a preferred embodiment of the device for determining a processing scheme based on selection probability according to the present invention; Figure 10 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 11 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0013] The processing scheme determination method based on selection probability provided in this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server can obtain the target loss information of the object to be processed from the client, construct a set of processing solutions containing multiple candidate processing solutions; generate candidate solution feature data based on the processing solution set and target loss information, and determine the feature field structure; input the candidate solution feature data into a preset discrete selection model to generate utility scores and selection probabilities, and determine the target processing solution; generate actual solution feature data based on the actual processing solution corresponding to the object to be processed, actual resource consumption data, target loss information, and feature field structure; verify the rationality of resource consumption based on the actual solution feature data, and generate and output the rationality verification result of the actual resource consumption data. This invention can be applied to business scenarios such as fintech and healthcare. It establishes a unified evaluation basis through candidate solution feature data and feature field structure, generates utility scores and selection probabilities through a discrete selection model to determine the target processing solution, and uses the same feature field structure for resource consumption rationality verification of the actual solution feature data, thereby reducing the disconnect between processing solution recommendation and resource consumption verification. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0014] Please see Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the probabilistic-based processing scheme determination method provided by the present invention. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0015] like Figure 2 As shown, the method for determining a processing scheme based on selection probability proposed in this invention includes the following steps: S10, obtain the target loss information of the object to be processed, and construct a processing scheme set containing multiple candidate processing schemes based on the target loss information; In this embodiment, the object to be processed is located using an object identifier, an object category identifier, and a business task identifier. The object identifier distinguishes the specific loss object, the object category identifier limits the available processing scope, and the business task identifier is used to bind the current loss event. The target loss information consists of the loss location, loss form, loss severity, and handling constraints. After data access, text descriptions, form fields, and attachment tags are organized into structured fields. The loss location, loss form, and loss severity are used to determine the initial handling method, while the handling constraints are used to filter out handling methods that do not meet the resource source range, processing time limit, or review requirements. The retained handling methods are supplemented with resource source category, handling procedure combination, and resource consumption fields to form candidate handling schemes. Multiple candidate handling schemes are bound to the object to be processed and the target loss information to generate a set of handling schemes.

[0016] When the target loss information has been fully submitted via a business form, a set of processing solutions is generated using a field mapping table. After the object category, loss location, loss type, loss severity, and handling constraints information from the business form are read, they are converted into structured loss records according to the field mapping table. A processing solution generation index table matches multiple candidate processing methods based on the structured loss records, and writes fields for resource source category, handling procedure combination, and resource consumption for each candidate processing method, forming a set of processing solutions containing multiple candidate processing solutions.

[0017] When the target loss information includes both text descriptions and attachment tags, a set of treatment options is generated by combining text extraction and attachment tag mapping. The text extraction component identifies the loss location, loss form, and loss extent from the loss description, while the attachment tag mapping component converts inspection photo tags, maintenance record tags, or review material tags into treatment constraint information. The loss location, loss form, loss extent, and treatment constraint information are combined to match candidate treatment methods, and the matching results are organized into a set of treatment options.

[0018] In the fintech business, after a claims task enters the loss assessment and review stage, the platform organizes the target loss information based on the report description, inspection records, and review restrictions, generates candidate processing solutions such as replacement, repair, or partial treatment, and binds them into a set of processing solutions.

[0019] In the field of healthcare, rehabilitation aids service platforms generate target loss information based on equipment type, damaged location, damage type, and maintenance limitations, and then form candidate processing solutions such as component replacement, structural repair, or functional testing.

[0020] This embodiment unifies the collection of information on the objects to be processed and the target loss information, and generates multiple candidate processing schemes based on the loss location, loss form, loss degree and disposal constraint information. This can reduce the attribution bias caused by the dispersed sources of candidate processing schemes, and make the set of processing schemes under the same loss event have a consistent data source and structure.

[0021] S20, Based on the set of processing schemes and the target loss information, generate candidate scheme feature data corresponding to each candidate processing scheme, and determine the feature field structure based on the candidate scheme feature data; In this embodiment, candidate processing schemes in the processing scheme set are read one by one according to their candidate scheme identifiers. The candidate scheme identifiers are used to maintain a one-to-one correspondence between each candidate processing scheme and its feature data. The resource source, treatment combination, estimated cycle, and resource consumption composition in the candidate processing schemes are converted into scheme-side fields. The object category, damaged location, loss form, loss degree, and treatment constraints in the target loss information are converted into loss-side fields. After merging the scheme-side fields and loss-side fields according to the candidate scheme identifiers, the category content is converted into an enumeration field, the numerical content into a unified dimension field, and the text content into a label field, thereby generating the candidate scheme feature data corresponding to each candidate processing scheme. After the candidate scheme feature data is generated, the field name, field type, field order, and missing field processing flags are extracted to form a feature field structure, ensuring that different candidate processing schemes within the same processing scheme set are recorded using consistent field definitions.

[0022] When the set of processing solutions already contains structured fields, a field mapping table is used to map the resource source, treatment combination and estimated cycle in the candidate processing solutions to fixed fields, and the object category, damaged location and loss degree in the target loss information are written into the shared fields, and then candidate solution feature data is generated according to the candidate solution identifier.

[0023] When the set of processing solutions contains text descriptions, a label extraction component is used to identify location words, morphological words, degree words, and treatment words in the candidate processing solutions and target loss information. Synonymous labels are merged into the same field name through a field dictionary, and then candidate solution feature data is generated.

[0024] In cases where field names from multiple sources are inconsistent, a field standardization table is used to map different input fields to a unified field name, and the feature field structure is determined according to the field type and field order to constrain the organization of the feature data of candidate schemes.

[0025] In the fintech business, the loss assessment and review platform reads multiple candidate processing solutions under the same claims task, converts content such as replacement, repair or partial processing into solution-side fields, and converts the report description, investigation record and review restrictions into loss-side fields, generating candidate solution feature data corresponding to each candidate processing solution.

[0026] In the field of healthcare, the rehabilitation aids service platform reads candidate processing solutions such as component replacement, structural repair, and functional testing post-processing, converts equipment category, damaged location, damage morphology, and maintenance limitations into loss-side fields, and generates corresponding candidate solution feature data and feature field structures.

[0027] This embodiment identifies candidate processing schemes in the set of organizational processing schemes by identifying candidate schemes, and merges the loss conditions in the target loss information with the fields of candidate processing schemes into candidate scheme feature data, which enables each candidate processing scheme to have a consistent data recording caliber; then, the field names, field types and field order are extracted from the candidate scheme feature data to form a feature field structure, which can reduce the comparison bias caused by field missing, naming differences or type inconsistencies between different candidate processing schemes.

[0028] S30, input the feature data of the candidate schemes into a preset discrete selection model to generate a utility score corresponding to each candidate processing scheme; In this embodiment, candidate solution feature data is written into a preset discrete selection model according to candidate solution identifiers. Candidate solution identifiers are used to maintain the correspondence between candidate processing solutions and utility scores. Category fields in the candidate solution feature data are converted into enumerated values ​​or one-hot vectors, numerical fields are standardized in terms of dimensions, and label fields are converted into model-readable markers. The preset discrete selection model stores the model field sequence, field weight parameters, direction markers, and baseline parameters. The model reads field values ​​according to the model field sequence and combines the field weight parameters, direction markers, and baseline parameters to generate the utility score corresponding to each candidate processing solution.

[0029] The pre-defined discrete selection model can include an input parsing unit, a field encoding unit, a field alignment unit, a direction transformation unit, a utility mapping unit, and a result binding unit. The input parsing unit reads the candidate solution identifier, field name, field type, and field value from the candidate solution feature data. The field encoding unit converts category fields into enumerated values ​​or one-hot vectors, converts label fields into model-readable markers, and unifies the dimensions of numerical fields. The field alignment unit adjusts the arrangement of fields according to the model field sequence, ensuring that different candidate solutions enter subsequent processing with the same field caliber. The direction transformation unit adjusts the field value direction based on direction markers, ensuring that fields such as resource consumption, processing time, resource source, and loss degree have a consistent evaluation direction in utility score generation. The utility mapping unit generates utility scores corresponding to candidate solutions based on field weight parameters and baseline parameters. The result binding unit binds the utility scores to the candidate solution identifiers, forming a utility output record for each candidate solution.

[0030] The preset discrete choice model can adopt a multi-choice model structure or a utility function structure with field grouping weights. The multi-choice model structure is used to evaluate multiple candidate treatment options concurrently. The field grouping weighted utility function structure assigns weights to resource consumption, resource source, treatment procedure, processing time, loss severity, and object attribute fields, allowing different types of fields to participate in utility score generation according to their corresponding weights. Field weight parameters can be trained using historical processing records, expert selection records, or completed review records, or they can be pre-configured by business review criteria and then updated based on processing results. Direction markers control the direction of field values' influence on utility scores; for example, higher resource consumption or longer processing time reduces utility scores, while stable resource sources, complete treatment procedures, or matching loss severity increase utility scores.

[0031] When the candidate solution's feature data fields are complete, a linear utility function is used to generate utility scores. Field values ​​are arranged according to the model field sequence, weighted by field weight parameters, and then superimposed with baseline parameters to obtain the utility score for each candidate solution. The linear utility function directly reflects the contribution of each field to the overall evaluation result of the candidate solutions, facilitating the subsequent generation of selection probabilities based on the utility scores.

[0032] When candidate solution feature data contains category fields or missing fields, field encoding and missing value labeling are used to generate utility scores. Category fields are converted into model input values, and missing fields are written with missing values ​​and substitute values ​​before participating in model processing. For fields such as resource source category, treatment process combination, loss degree label, and object category identifier, the field encoding unit converts different field contents into a unified input format. For candidate treatment solutions lacking fields such as processing time, resource consumption composition, or label, the model reads substitute values ​​according to the missing values ​​or reduces the weight of the corresponding field to avoid missing fields directly disrupting the utility score generation process.

[0033] When there are significant differences in the field values ​​of candidate solution feature data, the pre-defined discrete selection model can perform interval compression processing before utility mapping. After interval compression, the resource consumption component field and processing time field enter the direction conversion unit, reducing the excessive influence of extreme field values ​​on utility scores. The loss degree field and object attribute field can be converted into segmented values ​​according to the field type, enabling different candidate solutions to generate utility scores on the same evaluation scale.

[0034] In the fintech business, loss assessment and review platforms input resource sources, processing time, resource consumption composition, and loss severity into a pre-defined discrete selection model. The model then generates utility scores for candidate processing options such as replacement, repair, or partial repair through input parsing, field encoding, field alignment, direction transformation, and utility mapping. These utility scores express the overall suitability of each repair option for the current loss event. For example, original equipment manufacturer (OEM) parts replacement scores higher in terms of resource sources and completeness of disposal, while partial repair scores higher in terms of resource consumption and processing time.

[0035] In the healthcare sector, rehabilitation aids service platforms input equipment type, damaged location, maintenance resources, and processing time into a pre-defined discrete selection model. The model generates utility scores for candidate treatment options such as component replacement, structural repair, or functional testing post-treatment based on field weight parameters and direction markers. For candidate treatment options that consume significant maintenance resources, have long processing times, or mismatched damage levels, the utility score decreases accordingly; conversely, for candidate treatment options with stable resource sources, complete treatment procedures, and matching damage levels, the utility score increases accordingly.

[0036] This embodiment converts the feature data of candidate solutions into a data format that can be read by a discrete selection model, and generates utility scores based on the model field sequence, field weight parameters, direction markers, and baseline parameters. This enables each candidate processing solution to obtain a corresponding evaluation result under the same field caliber, reducing the matching deviation between candidate processing solutions and evaluation results.

[0037] S40, generate the selection probability for each candidate processing scheme based on the utility scores corresponding to all candidate processing schemes, and determine the target processing scheme from the set of processing schemes based on the selection probability; In this embodiment, utility scores are aggregated into a utility score set according to candidate solution identifiers. The candidate solution identifiers are used to maintain the binding relationship between each candidate processing solution and the subsequent selection probability. The scores in the utility score set undergo probability transformation processing, converting score differences into probability differences; then, all intermediate probability values ​​are normalized so that the selection probabilities of each candidate processing solution are in the same probability space. After the selection probability is generated, it forms a solution probability record together with the candidate solution identifiers. When the target processing solution is determined from the processing solution set, the identifier of the candidate solution with the highest selection probability is read according to the solution probability record, and the corresponding candidate processing solution is extracted from the processing solution set as the target processing solution.

[0038] When the utility score distribution is stable, exponential transformation and normalization are used to generate selection probabilities. The utility scores of each candidate treatment are first converted into intermediate probability values. The sum of all intermediate probability values ​​is used as the normalization benchmark, and the ratio of each intermediate probability value to the normalization benchmark is used as the selection probability.

[0039] When the utility scores vary widely, score shifting and compression are used to generate selection probabilities. First, the maximum score in the utility score set is used as the shift benchmark, and all utility scores are shifted. Then, the shift results are subjected to probability transformation to reduce the excessive influence of extreme scores on the selection probability distribution.

[0040] When multiple candidate processing schemes have similar selection probabilities, a probability ranking record and a candidate scheme stability marker are used to determine the target processing scheme. The probability ranking record is used to rank the candidate processing schemes, and the candidate scheme stability marker is used to select the candidate processing scheme with more stable processing results when the selection probabilities are small.

[0041] In the fintech business, the loss assessment and review platform converts the utility scores of candidate processing solutions such as replacement, repair and partial processing into selection probabilities, and determines the target processing solution from the same set of processing solutions based on the selection probabilities.

[0042] In the healthcare business, the rehabilitation aids service platform converts the utility scores of candidate treatment options such as component replacement, structural repair, and functional testing post-processing into selection probabilities, and determines the target treatment option from the set of treatment options corresponding to the same injury.

[0043] This embodiment converts the utility scores corresponding to all candidate processing schemes into selection probabilities and binds the selection probabilities to the candidate scheme identifiers. This enables the target processing scheme to be determined from the same set of processing schemes based on probability results, reducing selection bias caused by relying solely on a single score or manual judgment.

[0044] S50, obtain the actual processing plan and actual resource consumption data corresponding to the object to be processed, and generate actual plan feature data based on the actual processing plan, the actual resource consumption data, the target loss information and the feature field structure; In this embodiment, the actual processing plan is recorded using actual plan identifier, actual disposal type, actual resource source, and actual disposal content to represent the disposal result that has been formed or submitted for the object to be processed. Actual resource consumption data is recorded using fields such as material resources, working hours, service resources, testing resources, or replacement resources to reflect the resource occupancy corresponding to the actual disposal result. Target loss information is used to supplement the loss location, loss form, and loss degree corresponding to the actual processing plan, avoiding situations where the actual processing plan only reflects the disposal result without loss conditions. The feature field structure is used to constrain field names, field types, and field order, enabling the actual processing plan, actual resource consumption data, and target loss information to be converted into unified field records. After conversion, the actual plan feature data includes an actual plan identifier, loss field, disposal field, and resource consumption field, and is bound to the object to be processed.

[0045] When both the actual treatment plan and actual resource consumption data are submitted in structured forms, a field mapping table is used to generate the actual plan feature data. The field mapping table reads the actual treatment type, actual resource source, material resources, working hours resources, and service resources, and writes them into the corresponding fields according to the feature field structure. The loss location and loss degree in the target loss information are simultaneously written into the loss field.

[0046] When the actual treatment plan includes textual descriptions, text extraction and field completion are used to generate feature data for the actual plan. The text extraction component identifies the actual treatment type, treatment object, and resource source, while the field completion component completes the loss location, loss form, or loss degree based on the target loss information, and then completes the field arrangement according to the feature field structure.

[0047] When actual resource consumption data comes from multiple records, record merging and field deduplication are used to generate actual solution feature data. Material records, work hour records, and service records are merged according to the actual solution identifier, identical resource fields are merged into the same field value, and conflicting fields are marked with a validation flag before forming the actual solution feature data.

[0048] In the fintech business, the claims assessment and review platform reads actual repair records and actual resource consumption data, and writes the repair type, parts resources, labor hours, damaged parts and extent of damage into the same actual solution feature data to keep the data standards of the actual damage assessment records consistent.

[0049] In the field of healthcare, rehabilitation aids service platforms read actual maintenance records and resource consumption records, converting component replacement information, maintenance hours, service resources, damaged parts, and damage severity into actual solution feature data to form structured records for the same injury.

[0050] This embodiment converts the actual processing plan, actual resource consumption data, and target loss information into actual plan feature data according to the feature field structure. This enables the actual handling results, resource consumption content, and loss conditions to be organized under the same field caliber, reducing data deviations caused by scattered field sources, inconsistent field order, or inconsistent field types.

[0051] S60, Based on the actual scheme feature data, perform resource consumption rationality verification, generate the rationality verification result of the actual resource consumption data, and output the target processing scheme and the rationality verification result of the actual resource consumption data.

[0052] In this embodiment, after the actual disposal field, resource consumption field, and loss field in the actual solution feature data are read, a verification input record is formed according to the field name and field type. The actual disposal field is used to represent the submitted disposal content, the resource consumption field is used to represent the occupancy of material resources, working hours resources, service resources, or replacement resources, and the loss field is used to represent the location, form, and extent of damage. The reasonableness verification of resource consumption is completed through field alignment, value conversion, and judgment processing. Field alignment is used to ensure that the actual solution feature data is consistent with the verification field caliber, value conversion is used to convert category values, numerical values, and label values ​​into comparable data, and judgment processing is used to output reasonableness verification results such as reasonable, abnormal, or requiring review based on the verification parameters. The target treatment plan and the reasonableness verification results are output together, so that the recommended results and the actual resource consumption review results are presented in the same task record.

[0053] When the actual solution's feature data fields are complete, field weight verification is used to generate a reasonableness verification result. The actual disposal field, resource consumption field, and loss field are arranged in the order of verification fields. The values ​​of each field are weighted with the verification weight parameters to obtain a reasonableness score for resource consumption. Then, a reasonableness verification result is generated based on the judgment threshold.

[0054] When the actual solution's feature data includes category and range fields, grouping threshold verification is used to generate reasonableness verification results. The category field is used to determine the verification group, and the range field is used to determine the allowable range of resource consumption. When the actual resource consumption data falls within the corresponding range, a reasonable result is generated; when it exceeds the corresponding range, an abnormal result or a result requiring review is generated.

[0055] When missing fields exist in the feature data of the actual solution, missing field flag verification is used to generate reasonableness verification results. Missing fields are marked with missing flags, while complete fields continue to participate in resource consumption reasonableness verification. Missing field prompts are synchronously written into the verification results to facilitate the output results in expressing the data integrity status.

[0056] In the fintech business, the claims assessment and review platform reads the actual repair type, parts resources, labor resources, damaged location and extent of loss, generates a reasonableness verification result of the actual resource consumption data, and outputs the target processing plan and verification results for claims review and fund expenditure review.

[0057] In the healthcare business sector, the rehabilitation aids service platform reads the actual maintenance type, component resources, maintenance hours, service resources, and damage level, generates resource consumption rationality verification results, and outputs target treatment plans and verification results for maintenance review and service resource verification.

[0058] This embodiment reads the actual disposal field, resource consumption field, and loss field from the actual solution feature data, and generates a reasonableness verification result through field alignment, value transformation, and judgment processing. This enables the actual resource consumption data to be reviewed under a unified field standard. The target processing solution and reasonableness verification result are output synchronously, which can reduce the information fragmentation caused by the scattered display of recommendation results and review results.

[0059] In one embodiment, such as Figure 3 As shown, step S10 above includes: S101, Obtain the target loss information of the object to be processed, and parse the object category identifier, loss location identifier, loss form identifier, loss degree marker and disposal constraint marker of the object to be processed from the target loss information to generate a loss element record; S102, based on the loss location identifier, loss form identifier and loss degree mark in the loss element record, match and filter from the preset candidate scheme index table to obtain an initial disposal method set, and determine the resource source category, disposal procedure combination and resource consumption composition fields corresponding to each initial disposal method in the initial disposal method set; S103, based on the object category identifier and disposal constraint mark in the loss element record, perform feasibility screening on the initial disposal method set to obtain the available disposal method set, and retain the resource source category, disposal process combination and resource consumption composition fields corresponding to each available disposal method in the available disposal method set; S104, Based on the resource source category, disposal procedure combination and resource consumption composition fields corresponding to each available disposal method in the set of available disposal methods, generate multiple candidate processing schemes; S105, the multiple candidate processing schemes, the object to be processed, and the target loss information are bound together to generate a set of processing schemes.

[0060] In this embodiment, the target loss information is obtained by summarizing external business records, loss registration data, or material analysis results, and is then merged according to the objects to be processed. During the merging process, the object identifier of the object to be processed is used as the aggregation basis, and the loss description, damaged location, loss manifestation, loss degree, and disposal restrictions under the same object are written into the same data record. When parsing the target loss information, the object category identifier is extracted from the object type field, asset type field, or service object type field to distinguish the processing scope corresponding to different objects; the loss part identifier is extracted from the location field, component field, or area field to locate the damaged location; the loss form identifier is extracted from descriptions such as breakage, deformation, missing, functional abnormality, and appearance damage to indicate the loss manifestation; the loss degree marker is converted from the loss area, damage quantity, functional impact range, or manual review level to distinguish the disposal intensity; and the disposal constraint marker is converted from resource source restrictions, processing time limit restrictions, review requirement restrictions, or service window restrictions to control the range of subsequent available disposal methods. The above fields are combined into a loss element record, transforming the target loss information from a non-uniform data source into a searchable and filterable data record.

[0061] A pre-defined candidate solution index table is used to store the correspondence between loss elements and initial treatment methods. The index table can be built using a combined index based on loss location identifier, loss morphology identifier, and loss severity marker, or it can be partitioned by object category before establishing the mapping relationship between loss location and loss morphology. When querying the index table, the loss location identifier is used to narrow down the treatment location range, the loss morphology identifier is used to determine the treatment type, and the loss severity marker is used to distinguish different treatment intensities such as repair, replacement, post-inspection treatment, local treatment, or combined treatment. Upon index matching, a set of initial treatment methods is generated. Each initial treatment method synchronously reads the resource source category, treatment process combination, and resource consumption composition fields. The resource source category records the source of treatment resources, the treatment process combination records the sequence or combination of processes required to complete the treatment, and the resource consumption composition field records the components such as material resources, labor resources, inspection resources, service resources, or replacement resources.

[0062] Feasibility screening uses object category identifiers and disposal constraint tags in the loss element records as filtering criteria. Object category identifiers are used to exclude initial disposal methods that do not match the object category to be processed; for example, if the object category does not support a specific disposal procedure, the corresponding initial disposal method will not be retained. Disposal constraint tags are used to exclude initial disposal methods that do not meet resource source range, processing time limit, review requirements, or service window limitations. After screening, a set of available disposal methods is obtained. Each retained available disposal method continues to carry fields for resource source category, disposal procedure combination, and resource consumption composition to avoid losing the fields needed to generate subsequent candidate processing solutions during the screening process.

[0063] Candidate processing solutions are generated by combining available disposal methods and corresponding fields. For each available disposal method, the resource source category is written into the resource field, the disposal procedure combination into the procedure field, and the resource consumption composition field into the consumption composition field. A candidate solution identifier is also assigned to the generated result. Multiple candidate processing solutions correspond to different disposal methods or different resource combinations, covering the optional processing paths under the same target loss information. When generating a set of processing solutions, multiple candidate processing solutions, the object to be processed, and the target loss information are bound together. The binding fields can include the object identifier, the target loss information identifier, the candidate solution identifier, and the set identifier. After binding, each candidate processing solution in the processing solution set can be traced back to the same object to be processed and the same target loss information, reducing the attribution bias caused by the dispersion of candidate processing solution sources.

[0064] For example, the formula for representing a set of processing solutions is:

[0065] in, Represents a set of processing solutions; Let represent the i-th candidate processing scheme; n represents the number of candidate processing schemes in the processing scheme set; and i represents the sequence number of the candidate processing scheme.

[0066] This embodiment extracts object category identifiers, loss location identifiers, loss form identifiers, loss severity markers, and disposal constraint markers from target loss information to form loss element records, thus converting scattered loss descriptions into searchable data records. It determines the initial disposal method set by generating an index table of preset candidate solutions and simultaneously obtains fields for resource source category, disposal procedure combination, and resource consumption composition, ensuring a clear data source for the candidate processing solution generation process. By filtering initial disposal methods using object category identifiers and disposal constraint markers, it reduces disposal methods that do not match the object category, resource source range, or processing limitations. Furthermore, by binding multiple candidate processing solutions, objects to be processed, and target loss information, it ensures that candidate processing solutions in the processing solution set maintain a unified affiliation, thereby improving data consistency in the construction of candidate processing solutions for the same loss event.

[0067] In one embodiment, such as Figure 4 As shown, step S20 above includes: S201, extract the candidate solution identifier, resource source category, disposal procedure combination, resource consumption composition field and processing time field of each candidate processing solution from the processing solution set, and generate the candidate solution original record; S202, generate solution resource consumption features based on the resource consumption composition field in the original record of the candidate solution, generate solution resource features based on the resource source category and disposal procedure combination in the original record of the candidate solution, and generate solution duration features based on the processing time field in the original record of the candidate solution. S203, extract the loss degree marker and object category identifier from the target loss information, generate loss degree features based on the loss degree marker, and generate object attribute features based on the object category identifier; S204, according to the candidate scheme identifier, the scheme resource consumption feature, the scheme resource feature, the scheme duration feature, the loss degree feature and the object attribute feature are organized into fields, and missing filling marks are set for the fields after field organization to generate candidate scheme feature data corresponding to each candidate processing scheme; S205, Based on the field names, field types, field order, and missing fill tags in the candidate scheme feature data, determine the feature field structure.

[0068] In this embodiment, the processing scheme set stores multiple candidate processing schemes, each distinguished by a candidate scheme identifier. The candidate scheme identifier can be generated by combining the object identifier, target loss information identifier, and sequence marker, or it can be generated by a unique marker assigned by the processing scheme set. When extracting the candidate scheme identifier, the resource source category, processing procedure combination, resource consumption composition field, and processing time field are read simultaneously, and these fields are written into the candidate scheme original record. The candidate scheme original record is used to maintain the original field format in the candidate processing schemes, enabling subsequent feature generation to trace back to the corresponding candidate processing scheme and avoiding field misuse between different candidate processing schemes.

[0069] Resource consumption components can include items such as material resources, replacement resources, testing resources, working hours, service resources, and external collaboration resources. When generating resource consumption characteristics for a solution, the resource consumption components can be split, converted into units, grouped into intervals, and marked for missing values, converting different components into field values ​​that can participate in subsequent data processing. Resource source category is used to distinguish the range of resource sources, and disposal procedure combination is used to represent the disposal process, disposal intensity, and operation combination required by the candidate treatment solution. When generating solution resource characteristics, resource source category can be converted into a category field, and disposal procedure combination can be converted into a procedure quantity field, procedure type field, or combination label field. The processing time field can originate from the estimated time, service window, procedure cycle, or resource occupancy cycle in the candidate treatment solution. When generating solution time characteristics, different time units are converted into a unified time field, and written into a numerical time field or an interval time field according to the field type.

[0070] The loss severity marker in the target loss information reflects the scope of the loss's impact and the intensity of the response, while the object category identifier reflects the category attribute of the object to be processed. After extracting the loss severity marker, loss severity features can be generated according to severity level, loss area, quantity of damage, functional impact range, or review level. After extracting the object category identifier, object attribute features can be generated according to object type, structure category, resource category, or service category. Loss severity features and object attribute features are used in conjunction with the candidate processing solution's own fields, ensuring that the candidate solution feature data simultaneously includes solution-side fields and loss-side fields, reducing information bias caused by generating features solely based on the candidate processing solution's internal fields.

[0071] The field organization is based on candidate solution identifiers. Under each candidate solution identifier, solution resource consumption characteristics, solution resource characteristics, solution duration characteristics, loss severity characteristics, and object attribute characteristics are written into the same candidate solution feature record. During field organization, each field is configured with a field name, field type, field order, and field source flag. The field name distinguishes different data items, the field type distinguishes categorical, numeric, range, and label fields, and the field order maintains consistency in the field arrangement across different candidate solution feature data. Missing field imputation flags record whether a field is missing and the imputation method used for missing fields. Missing field imputation methods can include filling in fields of the same type of object, filling in fields within the same processing solution set, writing default placeholder values, or retaining the missing state. Setting missing field imputation flags ensures that missing fields do not directly disrupt the field structure of the candidate solution feature data.

[0072] The feature field structure is determined by the field names, field types, field order, and missing data filling markers in the candidate solution feature data. During this process, the field names of all candidate solution feature data within the same processing solution set can be counted, arranged according to a preset field order or the field generation order, and the field types and missing data filling markers can be written into the structure record. Once the feature field structure is formed, it ensures that the candidate solution feature data within the same processing solution set uses consistent field definitions, facilitating subsequent processing of candidate solution feature data according to the same field names, field types, and field order.

[0073] For example, the formula for representing the feature data of candidate solutions is:

[0074] in, This represents the feature data of the candidate solution corresponding to the i-th candidate processing solution; This represents the resource consumption characteristics of the scheme, and the characteristics generated from the corresponding resource consumption component fields. This represents the resource characteristics of the solution, corresponding to the characteristics generated by the combination of resource source categories and disposal procedures. This indicates the duration feature of the solution, corresponding to the feature generated from the processing duration field. These represent features indicating the degree of loss, corresponding to features generated from loss degree labels. This represents the attribute characteristics of an object, and the characteristics generated corresponding to the object category identifier.

[0075] This embodiment extracts candidate solution identifiers, resource source categories, disposal procedure combinations, resource consumption components, and processing time fields from the processing solution set, and combines them with loss severity markers and object category identifiers from the target loss information to generate candidate solution feature data. This enables each candidate processing solution to carry both solution-side features and loss-side features. By organizing various features into fields according to the candidate solution identifier and setting missing data filling markers, data organization deviations caused by missing fields, inconsistent field types, and inconsistent field orders can be reduced. By determining the feature field structure based on field name, field type, field order, and missing data filling markers, the candidate solution feature data within the same processing solution set can maintain a consistent field caliber, thereby improving the consistency of subsequent comparisons and readings of candidate processing solutions.

[0076] In one embodiment, such as Figure 5 As shown, step S30 above includes: S301, Based on the candidate scheme feature data, determine the candidate scheme identifier, candidate feature field value and candidate feature field sequence corresponding to each candidate processing scheme, and generate a candidate scheme input record; S302, determine the model field sequence, scheme utility weight parameters, feature direction markers and baseline utility parameters from the preset discrete choice model; S303, Based on the model field sequence, perform field alignment on the candidate feature field sequence and candidate feature field values ​​in the candidate solution input record to generate a candidate solution aligned record; S304, Based on the feature direction marker, convert the candidate feature field values ​​in the candidate scheme alignment record into a candidate scheme direction feature record; S305, based on the scheme utility weight parameter, the benchmark utility parameter and the candidate scheme direction feature record, generate the utility score corresponding to each candidate processing scheme.

[0077] In this embodiment, candidate solution feature data is read item by item according to candidate solution identifiers, and candidate feature field values ​​and candidate feature field sequences are separated from each candidate solution feature data item. The candidate solution identifier is used to maintain the unique attribution of candidate processing solutions during input, alignment, direction transformation, and utility score generation processes. Candidate feature field values ​​may include numerical fields, categorical fields, interval fields, and label fields, and the candidate feature field sequence is used to record the arrangement position of these fields in the candidate solution feature data. When generating candidate solution input records, the candidate solution identifier, candidate feature field values, and candidate feature field sequences are written to the same record, so that each candidate processing solution has an independent model input unit.

[0078] The preset discrete choice model stores the model field sequence, alternative utility weight parameters, feature direction markers, and baseline utility parameters. The model field sequence specifies the names and order of fields read by the model, preventing differences in field order among candidate alternative input records from affecting utility score generation. The alternative utility weight parameters record the strength of different fields' influence on the utility score; they can be configured by field name or grouped by field type. Feature direction markers indicate the direction of field value influence on the utility score; for example, when resource consumption increases, processing time increases, or the degree of loss changes, the corresponding field can be transformed into a positive, negative, or segmented direction. The baseline utility parameters provide a fixed utility term for candidate treatments, ensuring that each candidate treatment has a unified baseline term before field values ​​are used in processing.

[0079] Field alignment uses the model field sequence as a reference, rearranging the candidate feature field sequences and values ​​in the candidate solution input records. If a candidate feature field sequence contains a field identical to the model field sequence, the corresponding candidate feature field value is written to the model field position. If a candidate feature field sequence contains a field that the model does not read, the field can be retained in the additional field area or removed from the model input area. If a candidate solution input record contains a missing field in the model field sequence, a missing field marker or a replacement field value can be written. After field alignment, a candidate solution alignment record is generated, and the field order in the candidate solution alignment record is consistent with the model field sequence.

[0080] The direction transformation processes the candidate feature field values ​​in the candidate solution alignment record based on the feature direction markers. Positive fields can retain their original values ​​or be written into the direction field after dimension unification. Negative fields can undergo reverse transformation to align the field's direction of change with the utility score's direction of change. Segmented fields can be converted to corresponding direction values ​​according to their value ranges. Categorical fields can first be converted into enumerated values ​​or vector fields, and then written into the direction field according to the feature direction markers. After the transformation, a candidate solution direction feature record is formed. This record retains the candidate solution identifier and saves the field values ​​that have undergone direction unification.

[0081] The utility score is generated jointly by the solution utility weight parameter, the baseline utility parameter, and the candidate solution directional feature record. During generation, field values ​​from the candidate solution directional feature record are read according to the model field sequence. Each field value is weighted with the corresponding solution utility weight parameter and then merged with the baseline utility parameter to obtain the utility score of the candidate solution. Each utility score is bound to a candidate solution identifier, ensuring that the utility scores of different candidate solutions remain independently assigned.

[0082] For example, the formula for generating utility scores is:

[0083] in, This represents the utility score corresponding to the i-th candidate processing scheme; This represents the candidate scheme direction feature record corresponding to the i-th candidate processing scheme; Indicates the utility weight parameter of the scheme; This indicates that the directional feature records are weighted according to the weight parameter; b represents the benchmark utility parameter.

[0084] This embodiment organizes the candidate solution feature data into candidate solution input records and aligns the fields according to the model field sequence, which can reduce input bias caused by inconsistent field order and missing fields. By converting the direction of candidate feature field values ​​through feature direction markers, the changes in values ​​of different fields can have a consistent meaning in the generation of utility scores. By generating the utility score corresponding to each candidate treatment solution through the solution utility weight parameter and the benchmark utility parameter, the candidate treatment solutions can obtain comparable numerical results under the unified model field caliber.

[0085] In one embodiment, such as Figure 6 As shown, step S40 above includes: S401, determine the candidate scheme identifier corresponding to each candidate processing scheme from the processing scheme set, generate a utility score set based on the utility scores corresponding to all candidate processing schemes and the candidate scheme identifier, and establish an identifier mapping record between the utility score set and the processing scheme set; S402, determine the probability transformation parameter and the probability normalization parameter from the discrete choice model, and convert each utility score in the utility score set into a probability intermediate value based on the probability transformation parameter to generate a probability intermediate value set; S403, Based on the probability normalization parameter, normalize the probability intermediate value set to generate the selection probability corresponding to each candidate processing scheme, and bind the selection probability with the candidate scheme identifier to generate a scheme probability record; S404, Generate a probability sorting record based on the selection probability in the scheme probability record, and determine the target processing scheme from the processing scheme set based on the probability sorting record and the identifier mapping record.

[0086] In this embodiment, the candidate solution identifier corresponding to each candidate processing solution is read from the processing solution set, and the candidate solution identifier and its corresponding utility score are written into the same utility score record. The candidate solution identifier is used to maintain the correspondence between the utility score and the candidate processing solutions within the processing solution set, preventing the utility score from deviating from the source of the candidate processing solution during subsequent probability transformation and sorting. All utility score records are merged according to the same object to be processed or the same processing solution set to form a utility score set. An identifier mapping record is established between the utility score set and the processing solution set. The identifier mapping record can store the set identifier, candidate solution identifier, utility score position, and candidate processing solution storage location, enabling the specific candidate processing solution in the processing solution set to be retrieved later based on the probability results.

[0087] In the discrete choice model, probability transformation parameters and probability normalization parameters are read. The probability transformation parameters control the magnitude of the transformation from utility scores to probability medians, and can employ exponential transformation coefficients, temperature adjustment coefficients, score shifting coefficients, or piecewise transformation coefficients. Each utility score in the utility score set is converted to a probability median based on the probability transformation parameters, and the differences in utility scores are retained as differences between probability medians. After the probability median set is generated, each probability median remains bound to a candidate solution identifier, facilitating the continued identification of candidate solutions after subsequent normalization processing.

[0088] The probability normalization parameter controls the normalization caliber of the probability intermediate value set. It can include the selection method for the normalization denominator, the minimum probability cutoff value, the number of digits retained for probability, and a flag for handling outlier intermediate values. When normalizing the probability intermediate value set, all normalizable probability intermediate values ​​are used as a benchmark, and each probability intermediate value is converted into a corresponding selection probability. After generating the selection probabilities, they are bound to candidate solution identifiers to form a solution probability record. The solution probability record can simultaneously store candidate solution identifiers, utility scores, probability intermediate values, selection probabilities, and normalization status, facilitating the identification of field sources and numerical changes during the probability generation process.

[0089] The selection probabilities in the scheme probability record are used to generate the probability ranking record. The ranking process can proceed from highest to lowest selection probability, or, when selection probabilities are the same or differ by less than a set range, a stable ranking result can be formed by combining the candidate scheme identifier order, utility score, or storage order in the identifier mapping record. After the probability ranking record is generated, the identifiers of candidate schemes that meet the selection criteria are read from the ranking result, and the corresponding candidate processing schemes are extracted from the processing scheme set according to the identifier mapping record to obtain the target processing scheme. The target processing scheme retains the resource source category, disposal procedure combination, and resource consumption composition fields from the original candidate processing schemes, ensuring that the output object comes from existing candidate processing schemes within the processing scheme set.

[0090] For example, the formula for generating the probability of selection is:

[0091] in, This represents the selection probability corresponding to the i-th candidate processing scheme; This represents the utility score corresponding to the i-th candidate processing scheme; This represents the utility score corresponding to the j-th candidate processing scheme; This represents the probability transformation parameter, used to adjust the sensitivity when converting utility scores to intermediate probability values; n represents the number of candidate treatment options. This represents the denominator after normalizing the median probability of all candidate processing schemes.

[0092] The formula for determining the target processing solution is:

[0093] in, C represents the target processing solution; C represents the set of processing solutions. This represents the i-th candidate processing solution in the set of processing solutions; This represents the selection probability corresponding to the i-th candidate processing scheme; This indicates that the candidate processing solution with the highest probability of selection is chosen from the set of processing solutions.

[0094] This embodiment establishes a mapping relationship between the utility score set and the processing scheme set by identifying candidate schemes. It converts the utility scores of all candidate processing schemes into selection probabilities under the same probability caliber by using probability transformation parameters and probability normalization parameters. This enables the utility differences between different candidate processing schemes to be transformed into rankable probability results. By determining the target processing scheme through scheme probability records and probability ranking records, the matching deviation between utility scores and candidate processing schemes can be reduced, and the target processing scheme can be extracted from the processing scheme set according to clear identifiers, thereby improving the data consistency of the candidate processing scheme selection process.

[0095] In one embodiment, such as Figure 7 As shown, step S50 above includes: S501, parse the fields of actual resource source category, actual disposal procedure combination and actual processing time from the actual processing plan corresponding to the object to be processed, and generate the actual resource consumption composition field based on the actual resource consumption data; S502, extract the actual loss level marker and actual object category identifier from the target loss information, and generate an actual loss element record; S503, based on the field names and field order in the feature field structure, convert the actual resource consumption composition field, the actual resource source category, the actual disposal process combination, the actual processing time field, and the actual loss degree marker and actual object category identifier in the actual loss element record into actual field value records; S504, based on the field type and missing field filling mark in the feature field structure, perform field type conversion and missing field filling on the actual field value record to generate the actual scheme alignment feature record; S505, bind the actual solution alignment feature record, the object to be processed, the actual processing solution, and the actual resource consumption data to generate actual solution feature data.

[0096] In this embodiment, after the actual processing plan is read, it is aggregated according to the objects to be processed, and the fields of actual resource source category, actual disposal procedure combination, and actual processing time are parsed from the actual processing plan. The actual resource source category records the range of resource sources already used; the field value can be converted from source number, source type, resource ownership marker, or supply channel marker. The actual disposal procedure combination records the completed disposal content; the field value can be converted from procedure number sequence, disposal item combination, maintenance item combination, or service item combination. The actual processing time field records the actual processing time; the field value can be converted from start time and finish time, or from working hours records, service cycle records, or task completion cycle.

[0097] Actual resource consumption data is broken down into components such as material resources, replacement resources, testing resources, working hours, service resources, and external collaboration resources, and then organized according to component name, quantity, unit, and source. The organized components are then merged into the actual resource consumption component field. This field preserves the composition of the actual resource consumption data, allowing the disposal details in the actual processing plan to be included in the same feature record along with the resource occupancy field.

[0098] The target loss information extracts the actual loss severity marker and the actual object category identifier. The actual loss severity marker can be derived from loss level, damage scope, functional impact, or audit level, reflecting the intensity of the loss corresponding to the actual handling plan. The actual object category identifier can be derived from object type, asset category, equipment category, or service object category, distinguishing the meaning of field values ​​under different object categories. After the actual loss severity marker and actual object category identifier are written into the actual loss element record, they participate in field transformation along with the actual handling plan and actual resource consumption data.

[0099] The field names and order in the feature field structure control how the actual field value records are organized. The actual resource consumption composition field, actual resource source category, actual disposal procedure combination, actual processing time field, actual loss level marker, and actual object category identifier are matched item by item according to their field names and written to the corresponding positions according to the field order. When field names are inconsistent, they are converted to consistent names through a field mapping table; when field orders are inconsistent, they are rearranged according to the feature field structure to generate actual field value records.

[0100] The field type and missing field filling markers in the feature field structure are used to standardize the actual field value records. When the field type is numeric, the field value is converted to a unit, format, or range; when the field type is categorical, the field value is enumerated and encoded or converted to a category number; when the field type is tag-based, the field value is merging tags or converting to a tag dictionary. When a missing field exists, a supplementary value, placeholder value, or missing status value is written according to the missing field filling markers. After the field type conversion and missing field filling are completed, the actual scheme aligned feature record is generated.

[0101] The actual solution alignment feature record is bound to the object to be processed, the actual processing plan, and the actual resource consumption data to form actual solution feature data. The bound fields can include object identifier, actual solution identifier, resource consumption data identifier, and feature record identifier. After binding, the actual solution feature data can simultaneously trace back the actual disposal content, resource consumption composition, and loss conditions, reducing matching discrepancies between different data sources.

[0102] For example, the formula for representing the feature data of an actual solution is:

[0103] in, This represents the characteristic data of the actual solution; This indicates the characteristics corresponding to the fields that represent the actual resource consumption components; This indicates the characteristics corresponding to the combination of actual resource source categories and actual disposal procedures; This indicates the feature corresponding to the actual processing time field; The features that indicate the actual degree of loss are marked; This indicates the feature corresponding to the actual object category identifier.

[0104] This embodiment parses the fields of actual resource source category, actual disposal procedure combination, and actual processing time from the actual processing plan, and converts the actual resource consumption data into actual resource consumption component fields, enabling the unified organization of actual disposal content and resource occupancy. By extracting the actual loss degree marker and actual object category identifier from the target loss information, the actual plan feature data can include the corresponding loss conditions. By controlling the field name, field order, field type, and missing data filling method through the feature field structure, data deviations caused by inconsistent field naming, inconsistent field arrangement, inconsistent field type, and missing fields can be reduced. By binding the actual plan to align feature records, objects to be processed, actual processing plans, and actual resource consumption data, the consistency and traceability of the actual plan feature data can be improved.

[0105] In one embodiment, such as Figure 8 As shown, step S60 above includes: S601, Based on the actual scheme feature data, determine the actual feature field values ​​and the actual feature field sequence, and generate the actual verification input record; S602, determine the verification field sequence, preset verification weight parameters, resource consumption direction marker and rationality judgment threshold from the preset verification parameter table; S603, based on the verification field sequence, perform field alignment on the actual feature field sequence and actual feature field value in the actual verification input record to generate an actual verification aligned record; S604, Based on the resource consumption direction marker, convert the actual feature field value in the actual verification alignment record into an actual verification direction feature record; S605, Based on the preset verification weight parameters and the actual verification direction feature record, a resource consumption rationality score is generated, and based on the resource consumption rationality score and the rationality judgment threshold, a rationality verification result of the actual resource consumption data is generated; S606, output the rationality verification results of the target processing scheme and the actual resource consumption data.

[0106] In this embodiment, after the actual scheme feature data is read, it is split into actual feature field values ​​and actual feature field sequences according to the actual scheme identifier, field name, and field order. The actual feature field values ​​carry information such as resource consumption composition, actual resource source, actual disposal procedures, actual processing time, actual loss level, and actual object category. The actual feature field sequences record the arrangement position of each field within the actual scheme feature data. When generating the actual verification input record, the actual feature field values ​​and actual feature field sequences are written into the same record, enabling the actual resource consumption data to participate in rationality verification in a structured form.

[0107] The preset verification parameter table stores the verification field sequence, preset verification weight parameters, resource consumption direction markers, and reasonableness judgment thresholds. The verification field sequence specifies the names and order of fields read during the verification process. The preset verification weight parameters record the degree of influence of different fields on the resource consumption reasonableness score. The resource consumption direction markers indicate the direction of influence of field value changes on the resource consumption reasonableness score. The reasonableness judgment threshold converts the resource consumption reasonableness score into a reasonableness verification result. The preset verification parameter table can be configured with different parameter records according to object category, loss severity range, or disposal procedure type, ensuring that the actual scheme characteristic data can match the corresponding verification criteria.

[0108] During field alignment, the actual verification input record is read with reference to the verification field sequence. If a field in the actual feature field sequence matches the verification field sequence, the corresponding actual feature field value is written to the verification field position. If a field in the actual feature field sequence is not read during the verification process, the corresponding field is retained as an additional field or excluded from the verification input area. If a corresponding actual feature field value is missing in the verification field sequence, a missing value marker or substitute field value is written. After alignment, an actual verification alignment record is generated to ensure that the field arrangement in the actual scheme feature data is consistent with the preset verification parameter table.

[0109] Resource consumption direction markers are used to transform the direction of actual feature field values ​​in the actual validation alignment records. For resource consumption composition fields, actual processing time fields, and actual loss degree fields, the direction can be transformed into positive validation values, negative validation values, or segmented validation values ​​based on the increase or decrease of field values. Numerical fields are written into the direction field after unit conversion and interval compression, categorical fields are written into the direction field after enumeration transformation, and label fields are written into the direction field after label mapping. After the transformation is completed, actual validation direction feature records are generated, enabling different fields to participate in score generation under the same directional caliber.

[0110] The resource consumption rationality score is generated based on preset verification weight parameters and actual verification direction feature records. During generation, field values ​​from the actual verification direction feature records are read according to the verification field sequence. These values ​​are weighted with the corresponding preset verification weight parameters and then combined to obtain the resource consumption rationality score. The resource consumption rationality score is compared with a rationality judgment threshold to generate the rationality verification result of the actual resource consumption data. The rationality verification result may include a rationality marker, an anomaly marker, a review requirement marker, a score, and a trigger field marker. In the output stage, the rationality verification results of the target processing plan and the actual resource consumption data are written to the same output record, allowing the verification conclusions of the target processing plan and the actual resource consumption data to be presented together.

[0111] For example, the formula for generating the resource consumption rationality score is:

[0112] in, This indicates a score representing the rationality of resource consumption; This represents the actual verification direction feature record after field alignment and direction transformation; N represents the preset verification weight parameter; This indicates that the feature records of the actual verification direction are weighted according to the preset verification weight parameters.

[0113] This embodiment reduces verification bias caused by inconsistencies in field names and arrangements by determining actual feature field values ​​and sequences from actual solution feature data and aligning fields using verification field sequences. By using resource consumption direction markers to reverse the direction of actual feature field values, fields such as resource consumption composition, actual processing time, loss level, and object category participate in score generation in a unified direction. By generating reasonableness verification results for actual resource consumption data through preset verification weight parameters and reasonableness judgment thresholds, resource consumption review transforms from scattered field judgment to unified scores and judgment results. By synchronously outputting the target processing solution and reasonableness verification results, the data correspondence between the recommended processing solution and the actual resource consumption verification results is improved.

[0114] In one embodiment, after receiving a car insurance claim, the insurance claims platform treats the same accident vehicle as a pending case and integrates the report information, inspection records, damage photo markings, repair shop preliminary opinions, and damage assessment review requirements into target loss information. The target loss information records the vehicle category, damaged part, damage type, damage severity, and handling constraints, such as a cracked front bumper, deformed right front fender, damaged headlight mounting structure, moderate accident severity, and repair time limit / claims processing period. The platform parses the object category identifier, damaged part identifier, damage type identifier, damage severity marker, and handling constraint marker from the target loss information to generate a loss element record. Based on the damaged part identifier, damage type identifier, and damage severity marker, the platform queries preset candidate solutions to generate an index table, obtaining initial handling methods such as replacing original parts, using non-original parts, partial repair, and partial repair plus component replacement after inspection. It also reads the resource source category, handling procedure combination, and resource consumption composition fields corresponding to each initial handling method. The platform then filters out initial handling methods that do not meet the requirements of vehicle model, repair resource source range, or claims review based on object category identifiers and handling constraint tags, resulting in a set of available handling methods. These available methods are then organized into multiple candidate handling solutions, forming a set of handling solutions. Each candidate handling solution in the set serves as an optional decision item. For example, candidate handling solution one is replacing original parts, with the resource source category being original parts resources. The handling process includes disassembly, parts replacement, painting, and re-inspection, and the resource consumption components include parts resources, painting resources, labor resources, and testing resources. Candidate handling solution two is replacing non-original parts, with the resource source category being alternative parts resources. The handling process includes disassembly, replacement, adjustment, and re-inspection. Candidate handling solution three is partial repair, with the resource source category being repair service resources. The handling process includes sheet metal repair, painting, and appearance verification.

[0115] The platform extracts the candidate solution identifier, resource source category, treatment procedure combination, resource consumption composition field, and processing time field for each candidate treatment solution from the set of treatment solutions, generating the original record of the candidate solution. Based on the resource consumption composition field, it generates solution resource consumption characteristics, such as organizing parts resources, labor resources, inspection resources, and painting resources into quantifiable fields; based on the resource source category and treatment procedure combination, it generates solution resource characteristics, such as converting original parts, non-original parts, partial repair procedures, and combined repair procedures into category fields or tag fields; based on the processing time field, it generates solution duration characteristics, such as converting estimated repair days, labor hours, and service windows into a unified time field. The platform extracts loss severity markers and object category identifiers from the target loss information, generating loss severity characteristics and object attribute characteristics, so that the repair solution not only reflects repair cost, parts type, and repair time, but also incorporates accident severity, vehicle model information, and vehicle damage structure. The platform organizes candidate solution characteristics into fields based on candidate solution identifiers, including resource consumption characteristics, resource characteristics, duration characteristics, loss severity characteristics, and object attribute characteristics. Missing data is then added to these fields to generate candidate solution feature data for each candidate processing solution. This candidate solution feature data essentially converts the repair solutions into data that the model can read, which may include resource consumption composition, resource source type, repair procedure combination, processing time, accident severity, and vehicle model information. The platform then determines the feature field structure based on the field names, field types, field order, and missing data additions in the candidate solution feature data, enabling comparisons of multiple repair solutions under the same field definition.

[0116] The platform inputs candidate solution feature data into a pre-defined discrete selection model. Each candidate processing solution corresponds to a candidate solution input record, which includes a candidate solution identifier, candidate feature field values, and a sequence of candidate feature fields. The discrete selection model pre-stores the model field sequence, solution utility weight parameters, feature direction markers, and baseline utility parameters. The platform aligns the candidate feature field sequence and candidate feature field values ​​according to the model field sequence, ensuring that different repair solutions, such as original parts replacement, non-original parts replacement, and partial repair, are input into the model in the same field order. After field alignment, the platform transforms the candidate feature field values ​​according to the feature direction markers. For example, if the repair cost is too high, the processing time is too long, or the accident risk is high, the corresponding field is processed in a negative direction; if the repair quality is high, the resource source is reliable, or the treatment process is complete, the corresponding field is processed in a positive direction. The platform generates a utility score for each candidate processing solution based on the solution utility weight parameters, baseline utility parameters, and candidate solution direction feature records. The utility score is used to express the overall matching degree of different repair solutions under the current damage. For example, replacing the original parts may score higher in terms of repair quality and vehicle matching, but it has a negative impact on resource consumption and processing time; partial repair may score higher in terms of resource consumption and processing time, but it is affected by the risk field when the damage is more severe.

[0117] The platform aggregates the utility scores and candidate solution identifiers corresponding to all candidate treatment options into a utility score set, and establishes an identifier mapping record between the utility score set and the treatment option set. The probability transformation parameter in the discrete choice model is used to convert each utility score into a probability median value, and the probability normalization parameter is used to normalize all probability median values ​​into selection probabilities. After generating the selection probabilities, the platform binds the selection probabilities to the candidate solution identifiers, forming a solution probability record. The solution probability record can display the likelihood of adoption for repair options such as original parts replacement, non-original parts replacement, and partial repair. The platform then generates a probability ranking record based on the solution probability record, and determines the candidate treatment option with the highest selection probability from the treatment option set as the target treatment option. Thus, the repair solution selection process transforms from relying solely on human experience judgment into a quantitative decision-making process based on utility scores and selection probabilities, enabling the formation of recommendations among multiple available repair options. For example, when partial repair has a higher probability of being selected in terms of resource consumption, processing time, and degree of loss, the platform will determine partial repair as the target processing solution; when the degree of loss is severe and the risk field of partial repair is high, the platform can determine replacement of original parts or a combination of disposal methods as the target processing solution.

[0118] After the actual loss assessment of a claim is completed, the platform obtains the actual handling plan and actual resource consumption data corresponding to the object to be processed. The actual handling plan can be the final repair method submitted by the repair shop, such as actually using original parts replacement plus repainting and inspection, or actually using partial repair plus component replacement. The actual resource consumption data can include parts resources, labor resources, inspection resources, painting resources, service resources, and external collaboration resources. The platform parses the actual resource source category, actual handling procedure combination, and actual handling time fields from the actual handling plan, and generates actual resource consumption component fields based on the actual resource consumption data. The platform extracts the actual loss degree marker and actual object category identifier from the target loss information to generate actual loss element records. Based on the field names and field order in the feature field structure, the platform converts the actual resource consumption component fields, actual resource source category, actual handling procedure combination, actual handling time field, actual loss degree marker, and actual object category identifier into actual field value records. Based on the field types and missing field filling markers in the feature field structure, the platform performs field type conversion and missing field filling on the actual field value records to generate actual solution alignment feature records. These actual solution alignment feature records, the objects to be processed, the actual processing solutions, and the actual resource consumption data are then bound together as actual solution feature data. The actual solution feature data and the candidate solution feature data use the same field structure, enabling the actual loss assessment results to be incorporated into the same evaluation system.

[0119] The platform verifies the rationality of resource consumption based on actual solution feature data. It determines the actual feature field values ​​and sequences from the actual solution feature data, generates actual verification input records, and then determines the verification field sequence, preset verification weight parameters, resource consumption direction markers, and rationality judgment thresholds from a preset verification parameter table. The platform aligns the actual feature field sequence and values ​​according to the verification field sequence, generating actual verification aligned records, and converts the actual feature field values ​​into actual verification direction feature records based on the resource consumption direction markers. For fields in the actual resource consumption data that are too high in spare parts resources, too high in labor hours, too long in processing time, or where the disposal procedures do not match the degree of loss, the platform converts them into corresponding verification field values ​​according to the resource consumption direction markers. The platform generates a resource consumption rationality score based on the preset verification weight parameters and actual verification direction feature records, and generates the rationality verification result of the actual resource consumption data based on the resource consumption rationality score and the rationality judgment threshold. The rationality verification result can be rated as reasonable, abnormal, or requiring review, and can simultaneously output trigger fields, such as spare parts resources exceeding the range of similar losses, labor hours not matching local repair procedures, and actual processing time not matching the degree of loss. The platform ultimately outputs a reasonableness verification result for the target treatment plan and the actual resource consumption data, allowing claims reviewers to see the reasonableness conclusions corresponding to both the recommended repair plan and the actual damage assessment amount. For example, if the target treatment plan is partial repair, but the actual treatment plan is replacement of original parts and the resource consumption is significantly higher than the corresponding damage level, the platform outputs actual resource consumption data that needs to be reviewed or is abnormal; if the target treatment plan is replacement of original parts and the actual resource consumption data matches the damage level, vehicle model information, parts type, and repair time, the platform outputs a reasonableness verification result.

[0120] In one embodiment, a processing scheme determination apparatus based on selection probability is provided, which corresponds one-to-one with the processing scheme determination method based on selection probability described in the above embodiments. (Refer to...) Figure 9 , Figure 9 This is a functional block diagram of a preferred embodiment of the processing scheme determination device based on selection probability of the present invention. The processing scheme determination device based on selection probability includes: a processing scheme set construction module 10, a candidate scheme feature generation module 20, a utility score generation module 30, a target processing scheme determination module 40, an actual scheme feature generation module 50, and a resource consumption verification output module 60. Detailed descriptions of each functional module are as follows: The processing scheme set construction module 10 is used to obtain the target loss information of the object to be processed, and construct a processing scheme set containing multiple candidate processing schemes based on the target loss information. The candidate solution feature generation module 20 is used to generate candidate solution feature data corresponding to each candidate processing solution based on the processing solution set and the target loss information, and to determine the feature field structure based on the candidate solution feature data; The utility score generation module 30 is used to input the feature data of the candidate schemes into a preset discrete selection model to generate a utility score corresponding to each candidate processing scheme. The target processing scheme determination module 40 is used to generate the selection probability corresponding to each candidate processing scheme based on the utility scores corresponding to all candidate processing schemes, and determine the target processing scheme from the set of processing schemes based on the selection probability. The actual solution feature generation module 50 is used to obtain the actual processing solution and actual resource consumption data corresponding to the object to be processed, and generate actual solution feature data based on the actual processing solution, the actual resource consumption data, the target loss information and the feature field structure; The resource consumption verification output module 60 is used to verify the rationality of resource consumption based on the actual scheme feature data, generate the rationality verification result of the actual resource consumption data, and output the target processing scheme and the rationality verification result of the actual resource consumption data.

[0121] In one embodiment, the processing scheme set construction module 10 is specifically used for: Obtain the target loss information of the object to be processed, and parse the object category identifier, loss location identifier, loss form identifier, loss degree marker, and disposal constraint marker of the object to be processed from the target loss information to generate a loss element record; Based on the loss location identifier, loss form identifier, and loss degree marker in the loss element record, the initial disposal method set is obtained by matching and filtering from the preset candidate solution index table, and the resource source category, disposal procedure combination, and resource consumption composition fields corresponding to each initial disposal method in the initial disposal method set are determined. Based on the object category identifier and disposal constraint marker in the loss element record, the feasibility of the initial disposal method set is screened to obtain the available disposal method set, and the resource source category, disposal process combination and resource consumption composition fields corresponding to each available disposal method in the available disposal method set are retained; Based on the resource source category, disposal procedure combination, and resource consumption composition fields corresponding to each available disposal method in the set of available disposal methods, multiple candidate disposal schemes are generated. The multiple candidate processing schemes, the object to be processed, and the target loss information are bound together to generate a set of processing schemes.

[0122] In one embodiment, the candidate scheme feature generation module 20 is specifically used for: Extract the candidate solution identifier, resource source category, disposal procedure combination, resource consumption composition field and processing time field for each candidate processing solution from the set of processing solutions to generate the original record of the candidate solution; Based on the resource consumption composition field in the original record of the candidate solution, generate the solution resource consumption feature; based on the resource source category and disposal procedure combination in the original record of the candidate solution, generate the solution resource feature; and based on the processing time field in the original record of the candidate solution, generate the solution time feature. Extract loss severity markers and object category identifiers from the target loss information, generate loss severity features based on the loss severity markers, and generate object attribute features based on the object category identifiers; According to the candidate solution identifier, the solution resource consumption feature, solution resource feature, solution duration feature, loss degree feature and object attribute feature are organized into fields, and missing filling tags are set for the fields after field organization to generate candidate solution feature data corresponding to each candidate processing solution; The feature field structure is determined based on the field names, field types, field order, and missing fill tags in the candidate scheme feature data.

[0123] In one embodiment, the utility score generation module 30 is specifically used for: Based on the candidate scheme feature data, determine the candidate scheme identifier, candidate feature field value and candidate feature field sequence for each candidate processing scheme, and generate a candidate scheme input record; Determine the model field sequence, alternative utility weight parameters, feature direction markers, and baseline utility parameters from the preset discrete choice model; Based on the model field sequence, the candidate feature field sequence and candidate feature field value in the candidate solution input record are aligned to generate a candidate solution aligned record; Based on the feature direction marker, the candidate feature field values ​​in the candidate scheme alignment record are converted into candidate scheme direction feature records; Based on the utility weight parameter of the proposed solution, the baseline utility parameter, and the directional feature record of the candidate proposed solution, a utility score is generated for each candidate processing solution.

[0124] In one embodiment, the target processing scheme determination module 40 is specifically used for: From the set of processing schemes, determine the candidate scheme identifier corresponding to each candidate processing scheme, generate a utility score set based on the utility scores corresponding to all candidate processing schemes and the candidate scheme identifier, and establish an identifier mapping record between the utility score set and the set of processing schemes; The probability transformation parameter and probability normalization parameter are determined from the discrete choice model. Based on the probability transformation parameter, each utility score in the utility score set is converted into a probability intermediate value to generate a probability intermediate value set. The probability intermediate value set is normalized based on the probability normalization parameter to generate the selection probability corresponding to each candidate processing scheme, and the selection probability is bound to the candidate scheme identifier to generate a scheme probability record. A probability sorting record is generated based on the selection probability in the scheme probability record, and a target processing scheme is determined from the processing scheme set based on the probability sorting record and the identifier mapping record.

[0125] In one embodiment, the actual scheme feature generation module 50 is specifically used for: Parse the fields of actual resource source category, actual disposal procedure combination and actual processing time from the actual processing plan corresponding to the object to be processed, and generate the actual resource consumption composition field based on the actual resource consumption data; Extract the actual loss level marker and actual object category identifier from the target loss information to generate an actual loss element record; Based on the field names and field order in the feature field structure, the actual resource consumption composition field, the actual resource source category, the actual disposal process combination, the actual processing time field, and the actual loss degree marker and actual object category identifier in the actual loss element record are converted into actual field value records. Based on the field types and missing field filling markers in the feature field structure, the actual field value records are converted into field types and missing fields are filled to generate actual scheme alignment feature records. The actual solution alignment feature record, the object to be processed, the actual processing solution, and the actual resource consumption data are bound together to generate actual solution feature data.

[0126] In one embodiment, the resource consumption verification output module 60 is specifically used for: Based on the actual scheme feature data, determine the actual feature field values ​​and the actual feature field sequence, and generate the actual verification input record; Determine the verification field sequence, preset verification weight parameters, resource consumption direction markers, and reasonableness judgment thresholds from the preset verification parameter table; Based on the verification field sequence, the actual feature field sequence and actual feature field value in the actual verification input record are aligned to generate an actual verification aligned record; Based on the resource consumption direction marker, the actual feature field value in the actual verification alignment record is converted into an actual verification direction feature record; Based on the preset verification weight parameters and the actual verification direction feature records, a resource consumption rationality score is generated, and based on the resource consumption rationality score and the rationality judgment threshold, a rationality verification result of the actual resource consumption data is generated; Output the rationality verification results of the target processing scheme and the actual resource consumption data.

[0127] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides determination and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When executed by the processor, the computer program implements the functions or steps of a probabilistic-based processing scheme determination method on the server side.

[0128] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides determination and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements client-side functions or steps of a probabilistic-based processing scheme determination method.

[0129] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain the target loss information of the object to be processed, and construct a set of processing schemes containing multiple candidate processing schemes based on the target loss information; Based on the set of processing schemes and the target loss information, candidate scheme feature data corresponding to each candidate processing scheme is generated, and the feature field structure is determined based on the candidate scheme feature data; The feature data of the candidate solutions are input into a preset discrete selection model to generate a utility score for each candidate processing solution. Based on the utility scores corresponding to all candidate processing schemes, a selection probability is generated for each candidate processing scheme, and a target processing scheme is determined from the set of processing schemes based on the selection probability. Obtain the actual processing plan and actual resource consumption data corresponding to the object to be processed, and generate actual plan feature data based on the actual processing plan, the actual resource consumption data, the target loss information and the feature field structure; Based on the actual scheme feature data, the rationality of resource consumption is verified, the rationality verification result of the actual resource consumption data is generated, and the rationality verification result of the target processing scheme and the actual resource consumption data is output.

[0130] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, and a computer program is stored thereon. When the computer program is executed by a processor, it performs the following steps: Obtain the target loss information of the object to be processed, and construct a set of processing schemes containing multiple candidate processing schemes based on the target loss information; Based on the set of processing schemes and the target loss information, candidate scheme feature data corresponding to each candidate processing scheme is generated, and the feature field structure is determined based on the candidate scheme feature data; The feature data of the candidate solutions are input into a preset discrete selection model to generate a utility score for each candidate processing solution. Based on the utility scores corresponding to all candidate processing schemes, a selection probability is generated for each candidate processing scheme, and a target processing scheme is determined from the set of processing schemes based on the selection probability. Obtain the actual processing plan and actual resource consumption data corresponding to the object to be processed, and generate actual plan feature data based on the actual processing plan, the actual resource consumption data, the target loss information and the feature field structure; Based on the actual scheme feature data, the rationality of resource consumption is verified, the rationality verification result of the actual resource consumption data is generated, and the rationality verification result of the target processing scheme and the actual resource consumption data is output.

[0131] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0132] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0133] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0134] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0135] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for determining a processing scheme based on selection probability, characterized in that, Includes the following steps: Obtain the target loss information of the object to be processed, and construct a set of processing schemes containing multiple candidate processing schemes based on the target loss information; Based on the set of processing schemes and the target loss information, candidate scheme feature data corresponding to each candidate processing scheme is generated, and the feature field structure is determined based on the candidate scheme feature data; The feature data of the candidate solutions are input into a preset discrete selection model to generate a utility score for each candidate processing solution. Based on the utility scores corresponding to all candidate processing schemes, a selection probability is generated for each candidate processing scheme, and a target processing scheme is determined from the set of processing schemes based on the selection probability. Obtain the actual processing plan and actual resource consumption data corresponding to the object to be processed, and generate actual plan feature data based on the actual processing plan, the actual resource consumption data, the target loss information and the feature field structure; Based on the actual scheme feature data, the rationality of resource consumption is verified, the rationality verification result of the actual resource consumption data is generated, and the rationality verification result of the target processing scheme and the actual resource consumption data is output.

2. The method for determining a processing scheme based on selection probability as described in claim 1, characterized in that, Obtain the target loss information of the object to be processed, and construct a set of processing solutions containing multiple candidate processing solutions based on the target loss information, including: Obtain the target loss information of the object to be processed, and parse the object category identifier, loss location identifier, loss form identifier, loss degree marker, and disposal constraint marker of the object to be processed from the target loss information to generate a loss element record; Based on the loss location identifier, loss form identifier, and loss degree marker in the loss element record, the initial disposal method set is obtained by matching and filtering from the preset candidate solution index table, and the resource source category, disposal procedure combination, and resource consumption composition fields corresponding to each initial disposal method in the initial disposal method set are determined. Based on the object category identifier and disposal constraint marker in the loss element record, the feasibility of the initial disposal method set is screened to obtain the available disposal method set, and the resource source category, disposal process combination and resource consumption composition fields corresponding to each available disposal method in the available disposal method set are retained; Based on the resource source category, disposal procedure combination, and resource consumption composition fields corresponding to each available disposal method in the set of available disposal methods, multiple candidate disposal schemes are generated. The multiple candidate processing schemes, the object to be processed, and the target loss information are bound together to generate a set of processing schemes.

3. The method for determining a processing scheme based on selection probability as described in claim 1, characterized in that, Based on the set of processing schemes and the target loss information, candidate scheme feature data is generated for each candidate processing scheme, and the feature field structure is determined based on the candidate scheme feature data, including: Extract the candidate solution identifier, resource source category, disposal procedure combination, resource consumption composition field and processing time field for each candidate processing solution from the set of processing solutions to generate the original record of the candidate solution; Based on the resource consumption composition field in the original record of the candidate solution, generate the solution resource consumption feature; based on the resource source category and disposal procedure combination in the original record of the candidate solution, generate the solution resource feature; and based on the processing time field in the original record of the candidate solution, generate the solution time feature. Extract loss severity markers and object category identifiers from the target loss information, generate loss severity features based on the loss severity markers, and generate object attribute features based on the object category identifiers; According to the candidate solution identifier, the solution resource consumption feature, solution resource feature, solution duration feature, loss degree feature and object attribute feature are organized into fields, and missing filling tags are set for the fields after field organization to generate candidate solution feature data corresponding to each candidate processing solution; The feature field structure is determined based on the field names, field types, field order, and missing fill tags in the candidate scheme feature data.

4. The method for determining a processing scheme based on selection probability as described in claim 1, characterized in that, The candidate solution feature data is input into a preset discrete selection model to generate a utility score for each candidate processing solution, including: Based on the candidate scheme feature data, determine the candidate scheme identifier, candidate feature field value and candidate feature field sequence for each candidate processing scheme, and generate a candidate scheme input record; Determine the model field sequence, alternative utility weight parameters, feature direction markers, and baseline utility parameters from the preset discrete choice model; Based on the model field sequence, the candidate feature field sequence and candidate feature field value in the candidate solution input record are aligned to generate a candidate solution aligned record; Based on the feature direction marker, the candidate feature field values ​​in the candidate scheme alignment record are converted into candidate scheme direction feature records; Based on the utility weight parameter of the proposed solution, the baseline utility parameter, and the directional feature record of the candidate proposed solution, a utility score is generated for each candidate processing solution.

5. The method for determining a processing scheme based on selection probability as described in claim 1, characterized in that, The process involves generating a selection probability for each candidate processing scheme based on its utility scores, and determining a target processing scheme from the set of processing schemes based on these selection probabilities. From the set of processing schemes, determine the candidate scheme identifier corresponding to each candidate processing scheme, generate a utility score set based on the utility scores corresponding to all candidate processing schemes and the candidate scheme identifier, and establish an identifier mapping record between the utility score set and the set of processing schemes; The probability transformation parameter and probability normalization parameter are determined from the discrete choice model. Based on the probability transformation parameter, each utility score in the utility score set is converted into a probability intermediate value to generate a probability intermediate value set. The probability intermediate value set is normalized based on the probability normalization parameter to generate the selection probability corresponding to each candidate processing scheme, and the selection probability is bound to the candidate scheme identifier to generate a scheme probability record. A probability sorting record is generated based on the selection probability in the scheme probability record, and a target processing scheme is determined from the processing scheme set based on the probability sorting record and the identifier mapping record.

6. The method for determining a processing scheme based on selection probability as described in claim 1, characterized in that, Obtain the actual processing plan and actual resource consumption data corresponding to the object to be processed, and generate actual plan feature data based on the actual processing plan, the actual resource consumption data, the target loss information, and the feature field structure, including: Parse the fields of actual resource source category, actual disposal procedure combination and actual processing time from the actual processing plan corresponding to the object to be processed, and generate the actual resource consumption composition field based on the actual resource consumption data; Extract the actual loss level marker and actual object category identifier from the target loss information to generate an actual loss element record; Based on the field names and field order in the feature field structure, the actual resource consumption composition field, the actual resource source category, the actual disposal process combination, the actual processing time field, and the actual loss degree marker and actual object category identifier in the actual loss element record are converted into actual field value records. Based on the field types and missing field filling markers in the feature field structure, the actual field value records are converted into field types and missing fields are filled to generate actual scheme alignment feature records. The actual solution alignment feature record, the object to be processed, the actual processing solution, and the actual resource consumption data are bound together to generate actual solution feature data.

7. The method for determining a processing scheme based on selection probability as described in claim 1, characterized in that, Based on the actual scheme feature data, the rationality of resource consumption is verified, a rationality verification result of the actual resource consumption data is generated, and the rationality verification results of the target processing scheme and the actual resource consumption data are output, including: Based on the actual scheme feature data, determine the actual feature field values ​​and the actual feature field sequence, and generate the actual verification input record; Determine the verification field sequence, preset verification weight parameters, resource consumption direction markers, and reasonableness judgment thresholds from the preset verification parameter table; Based on the verification field sequence, the actual feature field sequence and actual feature field value in the actual verification input record are aligned to generate an actual verification aligned record; Based on the resource consumption direction marker, the actual feature field value in the actual verification alignment record is converted into an actual verification direction feature record; Based on the preset verification weight parameters and the actual verification direction feature records, a resource consumption rationality score is generated, and based on the resource consumption rationality score and the rationality judgment threshold, a rationality verification result of the actual resource consumption data is generated; Output the rationality verification results of the target processing scheme and the actual resource consumption data.

8. A processing scheme determination device based on selection probability, characterized in that, The processing scheme determination device based on selection probability includes: The processing scheme set construction module is used to obtain the target loss information of the object to be processed, and construct a processing scheme set containing multiple candidate processing schemes based on the target loss information; The candidate solution feature generation module is used to generate candidate solution feature data corresponding to each candidate processing solution based on the processing solution set and the target loss information, and to determine the feature field structure based on the candidate solution feature data; The utility score generation module is used to input the feature data of the candidate schemes into a preset discrete selection model to generate a utility score corresponding to each candidate processing scheme. The target processing scheme determination module is used to generate the selection probability of each candidate processing scheme based on the utility scores corresponding to all candidate processing schemes, and determine the target processing scheme from the set of processing schemes based on the selection probability; The actual solution feature generation module is used to obtain the actual processing solution and actual resource consumption data corresponding to the object to be processed, and generate actual solution feature data based on the actual processing solution, the actual resource consumption data, the target loss information and the feature field structure; The resource consumption verification output module is used to verify the rationality of resource consumption based on the actual scheme feature data, generate the rationality verification result of the actual resource consumption data, and output the rationality verification result of the target processing scheme and the actual resource consumption data.

9. A computer device, characterized in that, The computer device includes a memory, a processor, and a selection probability-based processing scheme determination program stored in the memory and executable on the processor, wherein the selection probability-based processing scheme determination program, when executed by the processor, implements the steps of the selection probability-based processing scheme determination method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a processing scheme determination program based on selection probability, which, when executed by a processor, implements the steps of the processing scheme determination method based on selection probability as described in any one of claims 1-7.

Images