An ai-based service cost prediction method and system

Abstract

The present application relates to the technical field of computer, in particular to a kind of service cost prediction method and system based on AI, method includes: obtaining service order and available performance personnel in target performance period, establish order feature and personnel feature, form candidate execution relationship, and predict candidate operation duration and first visit completion probability.According to service location, candidate operation duration and available operation duration, establish temporary service path, determine incremental path occupancy, and update candidate operation duration in combination with performance personnel information and order sequence.Construct performance state representation and input state adjustment model, obtain resource occupancy weight and time limit influence weight, according to path occupancy, field operation occupancy, expected revisit occupancy and service time limit, calculate candidate service cost representation value.Select the candidate execution relationship that satisfies available operation duration constraint and minimum cost, iteratively obtain target performance scheme, and generate service cost prediction result.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an AI-based method and system for predicting service costs. Background Technology

[0002] With the development of on-site service delivery such as installation, repair, and delivery, companies typically need to arrange service personnel and service routes based on the location, time limit, personnel skills, and available working hours of service orders. For example, patent application No. 2019100488893, entitled "Personnel Scheduling Method and Computer Equipment," generates task scheduling tables, task sequences, and route planning based on personnel skill attributes, task skill requirements, working hours, and address; patent application No. 2021109357743, entitled "An On-Site Service Demand-Driven IT Personnel Scheduling Method and System," reduces on-site service delays through task allocation and route optimization; and patent application No. 2022114103262, entitled "A Job Duration Prediction Method, System, Equipment, and Storage Medium," selects a prediction model and predicts job duration based on job characteristics.

[0003] However, existing solutions often focus on a single aspect of personnel scheduling, route optimization, or work duration prediction. They typically treat work order processing time as a predetermined input or use the predicted time independently for scheduling. This makes it difficult to reflect the combined impact of factors such as service recipients, personnel experience, personnel skill matching, work order location on the route, and incomplete initial service on actual resource usage. When different personnel execute the same work order, on-site work time and the risk of revisiting may differ. Furthermore, changes in work order order sequence and route can alter personnel availability and the timeliness of subsequent work orders. This leads to service costs estimated based on static working hours or single-time scheduling results easily deviating from the actual performance plan, making it difficult to provide enterprises with accurate work order-level and overall performance cost predictions. Summary of the Invention

[0004] In view of the above technical problems, the present invention provides an AI-based service cost prediction method and system, which aims to solve the problem that the service cost prediction results are inconsistent with the actual performance plan due to the disconnect between on-site operation time, personnel capabilities, route arrangement, risk of revisit and service time limit in the existing on-site performance service.

[0005] Other features and advantages of the invention will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0006] According to one aspect of the present invention, an AI-based service cost prediction method is proposed, the method comprising: Obtain the set of service work orders and the set of available personnel within the target fulfillment period, establish the work order characteristics of each service work order and the personnel characteristics of each available personnel. The work order characteristics include service location, service time limit, object attributes, job type and capability requirements, and the personnel characteristics include quantified capability level, job experience and available job duration. The work order features and personnel features are combined to form multiple candidate execution relationships. A pre-trained on-site operation duration prediction model is used to determine the candidate operation duration of each candidate execution relationship. The probability of first visit completion is determined based on the capability matching between the service work order and the available personnel. A temporary service path is established based on the service location, the candidate job duration, and the available job duration. The incremental path occupancy corresponding to each candidate execution relationship is determined. The candidate job duration and the incremental path occupancy are updated based on the personnel information and work order order in the temporary service path. Based on the service work order set, the available fulfillment personnel set, the updated candidate job duration, and the first visit completion probability, a fulfillment status representation is constructed. The fulfillment status representation is then input into a pre-trained state adjustment model to obtain state adjustment parameters for coordinating the impact of timeliness and resource consumption. Based on the incremental path occupancy, the updated candidate job duration, and the expected revisit occupancy caused by the incomplete first visit, the expected resource occupancy of each candidate execution relationship is determined. Based on the first visit completion probability and service time limit, the expected timeliness impact of each candidate execution relationship is determined. The resource occupancy weight and timeliness impact weight are determined using the state adjustment parameters. The expected resource occupancy and the expected timeliness impact are weighted and summed to obtain the candidate service cost representation value. From the candidate execution relationships that satisfy the available job duration constraint, select the candidate service cost representation value with the smallest value as the target execution relationship. After determining the target execution relationship, update the temporary service path, the candidate job duration, and the performance status representation until all service work orders have a target execution relationship, or the remaining candidate execution relationships do not satisfy the available job duration constraint, to obtain the target performance plan. Based on the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact in the target fulfillment plan, the service cost prediction result for the on-site fulfillment service is generated.

[0007] Furthermore, the training samples of the on-site operation duration prediction model are formed by associating historical service work orders, historical performance personnel, historical service paths, and historical execution results; the object identifiers in the training samples are merged according to object categories, categorical fields are discretely encoded, continuous fields are scaled, high-dimensional fields are feature compressed, abnormal performance records are removed, and the sample distribution of different operation duration intervals is balanced.

[0008] Furthermore, the on-site operation duration prediction model is implemented using a multi-layer neural computing structure, including a basic duration prediction unit, a condition correction unit, and an interval confidence unit. The basic duration prediction unit generates a basic operation duration based on object attributes, operation type, weight, volume, service area, and capacity requirements. The condition correction unit corrects the basic operation duration based on the capability level, work experience, vehicles used, and work order order of the corresponding available personnel, generating the candidate operation duration. The interval confidence unit generates a duration interval confidence representation based on the output probability of different operation duration intervals.

[0009] Furthermore, the temporary service path and the candidate job duration are coordinated in an alternating update manner, including generating an initial service path using the basic job duration, determining the work order order and the personnel responsible for fulfilling the path based on the initial service path, inputting the personnel characteristics of the work order order and the personnel responsible for fulfilling the path into the condition correction unit, and re-verifying the feasibility of the path and adjusting the temporary service path based on the corrected candidate job duration, until the continuously updated personnel allocation relationship and work order order meet the preset stability conditions.

[0010] Furthermore, the service cost prediction results include service work order-level cost predictions and target fulfillment plan-level cost predictions. The service work order-level cost predictions are calculated by converting the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact of the corresponding service work order according to preset cost conversion parameters, and risk correction is performed based on the duration interval confidence characterization. The target fulfillment plan-level cost predictions are obtained by summing up the cost predictions of each service work order. After the actual fulfillment is completed, the training samples are updated using the actual operation time, first visit completion results, and actual path occupancy, and the fulfillment status characterization for the next target fulfillment cycle is established using incomplete service work orders, newly added service work orders, and the updated set of available fulfillment personnel.

[0011] Furthermore, the probability of first visit completion is determined based on the degree of matching between the capability requirements and the capability level, as well as the operational experience, through a preset risk mapping relationship; the probability of first visit failure is the complementary probability of the first visit completion probability; the expected occupancy for revisit is determined by calculating the path occupancy and on-site operation occupancy required for revisit based on the probability of first visit failure.

[0012] Furthermore, the performance status representation includes the current performance stage, the number of service work orders of different job types, the number of service work orders with different capacity requirements, the number of available performance personnel of different capacity levels, the spatial distribution of service work orders, the number of service work orders that have not reached the service time limit, the number of service work orders that have reached the service time limit, and the waiting level after reaching the service time limit.

[0013] Furthermore, the state adjustment model is trained based on state transition training samples, which are formed from at least one of historical performance data and simulated performance data. During training, state adjustment parameters are sampled for the performance state representation. The state adjustment parameters are evaluated using the difference between the subsequent cumulative service cost and the state value estimate formed after adopting the state adjustment parameters, and the variation range of parameter sampling distribution in consecutive training rounds is limited, thereby updating the state adjustment model. The state adjustment parameters are used to determine the timeliness impact weight and resource occupancy weight, and the candidate service cost representation value is determined based on the weighted result of the expected resource occupancy and the weighted result of the expected timeliness impact.

[0014] Furthermore, establishing the temporary service path includes: Using the empty path of each available fulfillment personnel starting from and returning to the service center as the initial path, for each service work order for which the target execution relationship has not yet been determined, the incremental path occupancy generated by inserting the service work order into the current path of the corresponding available fulfillment personnel is determined. The candidate execution relationship that satisfies the available job duration constraint and has the smallest candidate service cost representation value is written into the temporary service path, and after writing, the candidate service cost representation values ​​of the remaining candidate execution relationships are re-determined.

[0015] According to a second aspect of the present invention, an AI-based service cost prediction system is provided, the system comprising: The feature construction module is used to obtain the set of service work orders and the set of available performance personnel within the target performance cycle, and to establish the work order features of each service work order and the personnel features of each available performance personnel. The work order features include service location, service time limit, object attributes, job type and capability requirements, and the personnel features include quantified capability level, job experience and available job duration. The task indicator prediction module is used to combine the work order features and the personnel features to form multiple candidate execution relationships, use a pre-trained on-site task duration prediction model to determine the candidate task duration of each candidate execution relationship, and determine the first visit completion probability based on the capability matching between the service work order and the available personnel. The path planning module is used to establish a temporary service path based on the service location, the candidate job duration, and the available job duration, determine the incremental path occupancy corresponding to each candidate execution relationship, and update the candidate job duration and the incremental path occupancy based on the personnel information and work order order in the temporary service path. The status representation module is used to construct a performance status representation based on the service work order set, the available performance personnel set, the updated candidate job duration and the first visit completion probability, and input the performance status representation into the pre-trained status adjustment model to obtain status adjustment parameters for coordinating the impact of timeliness and resource consumption. The cost calculation module is used to determine the expected resource consumption of each candidate execution relationship based on the incremental path consumption, the updated candidate job duration, and the expected revisit consumption caused by the failure of the first visit. It also determines the expected timeliness impact of each candidate execution relationship based on the first visit completion probability and service time limit, and uses the state adjustment parameters to determine the resource consumption weight and timeliness impact weight. Finally, it weights and sums the expected resource consumption and the expected timeliness impact to obtain the candidate service cost representation value. The solution iteration module is used to select the candidate service cost representation value with the smallest value from the candidate execution relationships that satisfy the available job duration constraint as the target execution relationship, and update the temporary service path, the candidate job duration and the performance status representation after determining the target execution relationship, until all service work orders have determined the target execution relationship, or the remaining candidate execution relationships do not satisfy the available job duration constraint, to obtain the target performance solution; The cost prediction module is used to generate a service cost prediction result for the on-site fulfillment service based on the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact in the target fulfillment plan.

[0016] The technical solution of the present invention has the following beneficial effects: This invention establishes candidate execution relationships between service work orders and fulfilling personnel. It utilizes an artificial intelligence model to determine the appropriate on-site operation duration for each candidate relationship and evaluates fulfillment resource consumption by considering personnel capability matching, path occupancy, and initial service completion status. This allows for corresponding cost predictions when the same work order is executed by different personnel or located at different path positions. By coordinating resource occupancy and timeliness impact based on fulfillment status, it avoids pursuing shorter paths that lead to high-risk revisits and reduces path dispersion and decreased personnel utilization caused by prioritizing urgent work orders. As the target execution relationship is determined, the operation duration, path status, and cost representation are updated, ensuring that the resulting service cost predictions correspond to the actual personnel allocation and work order sequence. This improves the accuracy of predicting on-site labor occupancy, travel occupancy, revisit occupancy, and timeliness losses, providing a basis for enterprise personnel allocation, fulfillment scheme selection, and labor cost control. Attached Figure Description

[0017] Figure 1 This is a flowchart of an AI-based service cost prediction method as described in the embodiments of this specification; Figure 2 This is a structural block diagram of an AI-based service cost prediction system as described in the embodiments of this specification. Detailed Implementation

[0018] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make the invention more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of the invention. However, those skilled in the art will recognize that the technical solutions of the invention may be practiced with one or more of these specific details omitted, or other methods, components, apparatus, steps, etc., may be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of the invention.

[0019] Furthermore, the accompanying drawings are merely illustrative of the invention. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0020] This invention provides an AI-based method for predicting service costs. (Refer to...) Figure 1 The diagram shown illustrates a flowchart of an AI-based service cost prediction method according to an embodiment of the present invention. This method can be applied to devices such as personal computers and servers. The method can be executed by a device, which can be implemented by software and / or hardware. Specifically, the method may include the following steps S101-S107: In step S101, the set of service work orders and the set of available fulfillment personnel within the target fulfillment period are obtained, and the work order characteristics of each service work order and the personnel characteristics of each available fulfillment personnel are established. The work order characteristics include service location, service time limit, object attributes, job type and capability requirements, and the personnel characteristics include quantified capability level, job experience and available job duration.

[0021] The target fulfillment cycle can be divided by calendar days, work shifts, or multiple consecutive workdays. At the beginning of the cycle, the service work orders that need to be processed are retrieved from the work order management system, and the personnel who can participate in the fulfillment of the current cycle are retrieved from the personnel management system. The service work order set can simultaneously include newly added service work orders in the current cycle, unfinished service work orders from the previous cycle, and service work orders that require a second visit due to incomplete initial visits. The available fulfillment personnel set is a subset of the full fulfillment personnel set who are able to attend work normally and still have remaining work hours, which can be represented as: ; ; ; Where t represents the target performance period, and W represents the total set of personnel involved in the performance. This represents the set of available personnel for contract fulfillment within the target contract fulfillment period. This indicates the attendance status of the personnel (w) during the target performance period. This indicates the available working hours for the contract fulfillment personnel (w). Indicates the number of available personnel for contract fulfillment. This represents a set of service orders. This indicates the number of service orders. Personnel who are on leave, absent, have reached their maximum working hours, or are temporarily unable to fulfill their obligations are not included in the available personnel pool. The set of service work orders can be represented as: ; in, This indicates new service orders added within the target fulfillment period. This indicates service work orders that were transferred from previous fulfillment cycles but have not yet been completed. This indicates a service order that needs to be processed again because a previous visit was not completed. By retaining the work order identifiers from different sources, new requests, backlogged requests, and repeat visit requests can be distinguished in subsequent processing. However, the work order identifiers are only used for data association and are not used as characteristics to reflect operation time or service cost.

[0022] The characteristics of service order i can be represented as follows: ; in, Service locations can be indicated using latitude and longitude, postal codes, service area codes, or road network nodes. The service time limit can be expressed as a deadline, a deadline for fulfillment, or the remaining time until the deadline. Represents object properties, Indicates the job type. This indicates capability requirements. Object attributes can include service object category, object quantity, weight, volume, and the delivery, installation, or maintenance items to be completed. For fulfillment scenarios with many object types, specific object identifiers can be grouped into object categories to avoid an excessive number of object identifiers leading to overly complex work order features. Order location, object and quantity, service items, service area, weight, and volume can all be obtained from structured fields in the order planning system. The personnel characteristics of available personnel w can be represented as follows: ; in, Indicates the level of quantifiable ability. Indicates work experience. This indicates the available work hours. Capability levels and capability requirements use the same capability dimensions and grade standards. Values ​​can be assigned based on dimensions such as installation capability, electrical work capability, maintenance capability, or handling capability. Alternatively, when there are few types of work, it can be divided into general capability levels and professional capability levels. Work experience can be quantified based on at least one of the following: length of service, number of similar service orders completed, or recent completion status of similar work. Available work hours are the remaining time after deducting confirmed tasks, rest time, and other unavailable time from the planned work hours for the current period. This avoids including personnel registered on the roster but unable to actually perform services in subsequent candidate performance relationships.

[0023] In step S102, the work order features and personnel features are combined to form multiple candidate execution relationships. A pre-trained on-site operation duration prediction model is used to determine the candidate operation duration of each candidate execution relationship. The probability of first visit completion is determined based on the capability matching between the service work order and the available personnel.

[0024] Wherein, the first visit completion probability is determined according to the matching degree between the capability requirement and the capability level and the work experience via a preset risk mapping relationship, and the first visit non-completion probability is a complementary probability of the first visit completion probability. The on-site work duration prediction model is implemented by a multi-layer neural computing structure.

[0025] The candidate execution relationship is used to represent to-be-evaluated combinations between service work orders and available fulfillment personnel. The same service work order may form candidate execution relationships with a plurality of available fulfillment personnel respectively, and the same available fulfillment personnel may also form candidate execution relationships with a plurality of service work orders. Combinations that do not meet legal qualifications, safety conditions or mandatory capability requirements may not be included in the candidate execution relationships. Combinations where the capability level does not fully meet the capability requirement but still have basic execution conditions may be retained and assigned a high first visit non-completion risk, so as to facilitate subsequent comprehensive evaluation in combination with path occupancy, work duration and timeliness impact.

[0026] The on-site work duration is the time required for a fulfillment personnel to complete delivery, installation, maintenance, handling or other on-site work after arriving at the service location, and does not include the travel time required for the fulfillment personnel to move between service locations. Object attributes, work type, weight, volume and capability requirement are used to reflect the work scale and complexity of the service work order, and capability level and work experience are used to reflect the estimated efficiency of the corresponding available fulfillment personnel in completing the service work order. The multi-layer neural computing structure learns the nonlinear relationship between the above features, and outputs candidate work durations corresponding to each candidate execution relationship.

[0027] Since a complete service path has not yet been formed, the features available at the service work order creation stage and personnel features may be used to generate an initial value of the candidate work duration first.

[0028] When a two-level capability division manner is adopted, the first visit non-completion probability may be expressed as: ; wherein, represents the first visit non-completion probability when service work order i is executed by available fulfillment personnel w within the target fulfillment period t, is a preset risk value, and 0<p<1. When the capability level meets the capability requirement, the corresponding candidate execution relationship may be identified as a low-risk relationship. When the capability level does not fully meet the capability requirement but still has execution conditions, the corresponding candidate execution relationship may be identified as a risk relationship.

[0029] When employing multiple capability levels, multiple risk levels can be set based on the difference between capability requirements and capability levels, and these risk levels can be adjusted by considering the similar operational experience of available personnel. A higher number of completed similar service orders or greater experience in similar tasks can reduce the probability of initial failure. Conversely, a larger capability gap and less experience in similar tasks can increase the probability of initial failure. Therefore, when the same service order is executed by different available personnel, corresponding candidate operation times and initial completion probabilities can be obtained respectively.

[0030] In step S103, a temporary service path is established based on the service location, the candidate job duration, and the available job duration. The incremental path occupancy corresponding to each candidate execution relationship is determined. The candidate job duration and the incremental path occupancy are updated based on the personnel information and work order order in the temporary service path.

[0031] The temporary service path and the candidate job duration are coordinated in an alternating update manner. When the temporary service path is established, the empty path from the service center to the service center for each available fulfillment personnel is used as the initial path.

[0032] In this system, the service center serves as both the start and end point of each temporary service path, represented by node zero. Each temporary service path consists of service work orders visited sequentially by available personnel. When establishing a path, the feasibility of inserting the service work orders to be assigned into different positions within the current path of each available personnel is evaluated sequentially, and the insertion position that minimizes the total path occupancy increment is selected. The corresponding path update process can be represented as follows: ; This indicates the temporary service path before inserting the service order. Let ψ represent the available personnel to be evaluated, i represent the service order to be inserted, and ψ represent the path insertion function. This represents the temporary service path after insertion. The path insertion function compares the total path occupancy change caused by inserting the service order into the current path at each candidate location and returns the insertion result with the smallest incremental path occupancy.

[0033] Incremental path occupancy characterizes the additional occupancy caused by a new service order on the current path travel portion. When a service order is inserted between two adjacent nodes, the direct travel between the original adjacent nodes is replaced by the travel from the starting node to the new service location and from the new service location to the ending node. When a service order is inserted at the beginning or end of a path, the incremental path occupancy is determined by combining the travel from the service center to the first service location or from the last service location to the service center, respectively. The path travel can be determined based on road network distance, estimated travel time, or historical travel time for the same period.

[0034] Path feasibility is verified based on the total time required for available personnel to complete the current path. The total time includes the travel time from the service center, through each service location, and back to the service center, as well as the candidate job duration for each service order. If the total time does not exceed the available job duration, the current insertion is considered to meet the path feasibility requirements. If the total time exceeds the available job duration, the current insertion position is abandoned, or the relevant service order is transferred to other available personnel's candidate paths for evaluation.

[0035] Once the initial temporary service path is formed, the actual candidate personnel, vehicles used, path location, and work order order for each service work order can be obtained. The path location indicates the position of the service work order within the corresponding temporary service path. Inputting this information as the performance context into the on-site operation time prediction model allows for the correction of previously obtained candidate operation times. For example, the continuous working status of the personnel may differ at the beginning or end of the path for the same service unit, and the time required for object handling, loading / unloading, and tool preparation may differ when using different vehicles; therefore, the corresponding candidate operation times can have different correction results.

[0036] After the candidate job duration is updated, the total time of the corresponding temporary service path is recalculated and its feasibility is verified. If the change in the candidate job duration causes the path to be unable to meet the available job duration, the work order order can be adjusted, the insertion position changed, or the corresponding available fulfillment personnel replaced. After each adjustment of the temporary service path, the updated fulfillment personnel information and work order order are obtained again, and the candidate job duration and incremental path occupancy are corrected accordingly. Alternating updates can stop when the personnel allocation relationship and work order order no longer change, or when the difference between the candidate job durations obtained from two consecutive updates is lower than a preset threshold or when a preset number of updates is reached.

[0037] In step S104, a performance status representation is constructed based on the service work order set, the available performance personnel set, the updated candidate job duration, and the first visit completion probability. The performance status representation is then input into a pre-trained state adjustment model to obtain state adjustment parameters for coordinating the impact of timeliness and resource consumption.

[0038] The performance status representation includes the current performance stage, the number of service work orders of different job types, the number of service work orders with different capacity requirements, the number of available performance personnel of different capacity levels, the spatial distribution of service work orders, the number of service work orders that have not reached the service time limit, the number of service work orders that have reached the service time limit, and the waiting level after reaching the service time limit.

[0039] The performance status representation is used to convert scattered work order information, personnel information, and path calculation results within the current performance cycle into a fixed-length input that the performance adjustment model can process. When the number of service work orders or the number of available performance personnel changes, the dimensions of the performance status representation remain unchanged, allowing the performance adjustment model to be reused across different performance cycles. For quantitative information, statistics can be compiled according to job type, capacity requirements, and capacity level. For candidate job durations and first-visit completion probabilities, the mean, maximum, number of segments, or the proportion of work orders in the corresponding interval can be used for aggregation, avoiding directly inputting candidate execution relationships with varying quantities one by one into the performance adjustment model.

[0040] The current fulfillment stage indicates the temporal position of the target fulfillment cycle within the continuous fulfillment process. It can be represented by the number of cycles elapsed since the start of the fulfillment plan, the current cycle number, or the number of cycles remaining until the planned deadline. This feature reflects whether the fulfillment task is in a concentrated arrival phase, a continuous processing phase, or a remaining task cleanup phase, causing the status adjustment parameters to change as the fulfillment progresses.

[0041] The spatial distribution of service orders reflects the concentration and dispersion of orders within the service area. It can be formed based on the average distance from service orders of different job types or with different capacity requirements to the service center, as well as the average distance between similar service orders. When service orders are mainly concentrated near or adjacent to the service center, more orders can be completed through compact paths. When service orders are more dispersed or there are isolated orders far from the service center, simply reducing path occupancy may cause isolated orders to be continuously delayed. Therefore, it is necessary to adjust the emphasis between the impact on timeliness and the impact on resource consumption in conjunction with service time limits.

[0042] Service orders that have not reached their service deadline can be further grouped according to the remaining cycles before the deadline, enabling the status adjustment model to distinguish between orders with far-reaching service deadlines and those nearing their deadlines. The waiting level after reaching the service deadline can be represented by the number of cycles or the duration exceeding the deadline. A higher waiting level generally indicates a more significant impact on timeliness due to further delays. For different job types and capacity requirements, the number of orders that have not reached their service deadline, the number of orders that have reached their service deadline, and the average waiting level can be separately tallied to reflect the distribution of urgent orders across different service categories.

[0043] Before the performance status representation is input into the state adjustment model, the scale can be unified according to the value range of each feature. This ensures that the number of personnel, work orders, distance, time, and probability are within similar numerical ranges, avoiding inappropriate influence of features with large values ​​on the model output. The transformation parameters used for scale unification are determined by the state transition training samples and remain consistent during actual prediction.

[0044] The state adjustment model maps the performance state representation, and the state adjustment parameters can be expressed as: ; in, This represents the performance status characteristic corresponding to the target performance period t. This represents a state adjustment model. This represents the status adjustment parameters determined based on the current performance status, and satisfying the following: ; When the status adjustment parameter is close to 0, the calculation of subsequent candidate service cost representation values ​​can pay more attention to the impact on timeliness, giving higher processing priority to service orders that have reached or are close to their service deadlines. When the status adjustment parameter is close to 1, it can pay more attention to the impact on resource consumption, making it easier to combine service orders with similar paths that can make full use of available job time for processing. When the fulfillment status is between the two, the status adjustment parameter takes an intermediate value to balance service deadlines and path utilization efficiency.

[0045] When there are many work orders that have not reached their service deadline within the current cycle and path resources are relatively scarce, the weight of resource usage impact can be increased to reduce the backlog of work orders in subsequent cycles. When the number of work orders that have reached their service deadline increases or the waiting level increases, the weight of timeliness impact can be increased to prevent a small number of urgent work orders from being continuously delayed due to remote service locations. When there is a significant imbalance between the number of work orders with different capability requirements and the number of available fulfillment personnel at different capability levels, the status adjustment parameter can also reflect the scarcity of high-capability fulfillment personnel, preventing the limited number of high-capability fulfillment personnel from being over-occupied by low-capability demand work orders.

[0046] In step S105, the expected resource consumption of each candidate execution relationship is determined based on the incremental path occupancy, the updated candidate job duration, and the expected revisit occupancy caused by the failure of the first visit. The expected timeliness impact of each candidate execution relationship is determined based on the first visit completion probability and service time limit. The resource consumption weight and timeliness impact weight are determined using the state adjustment parameters. The expected resource consumption and the expected timeliness impact are weighted and summed to obtain the candidate service cost representation value.

[0047] The expected revisit occupancy is determined by calculating the path occupancy and on-site operation occupancy required for the revisit based on the probability of the first visit not being completed.

[0048] The expected resource consumption reflects the anticipated time consumption within the current fulfillment cycle after assigning a service order to an available fulfillment personnel. Incremental path consumption reflects the additional travel time added after the service order is inserted into the current path. Updated candidate job duration reflects the on-site time occupied by the corresponding fulfillment personnel at the service location. Expected revisit consumption reflects the additional subsequent fulfillment time that may be incurred if the initial visit is not completed. The lower the probability of initial visit completion, the greater the expected resource consumption caused by revisiting.

[0049] In one implementation, the expected repeat visit occupancy can be expressed as: ; in, This represents the expected repeat visit occupancy of service order i and available fulfillment personnel w within the target fulfillment period t. This indicates the probability of completing the first visit. This indicates the set of resources required for a subsequent visit. Indicates resource project The corresponding occupancy. Resource items can include travel time for revisiting and on-site work time. This calculation uses a probability-weighted approach, transforming the uncertain revisit into an expected occupancy that can be evaluated together with current route occupancy and on-site work occupancy.

[0050] Expected resource occupancy can be uniformly converted from incremental path occupancy, updated candidate job duration, and expected revisit occupancy into time occupancy. Alternatively, it can be converted separately based on vehicle usage costs, unit time costs of fulfilling personnel, and additional revisit costs, and then summed. Using time occupancy for evaluation maintains dimensional consistency across different candidate execution relationships. When using monetary costs for evaluation, each type of occupancy can be multiplied by its corresponding unit cost parameter.

[0051] The expected timeliness impact reflects the waiting impact when a service order fails to be completed on time within the current fulfillment period. For service orders that have reached or exceeded their service time limit, the timeliness impact can increase with the overdue time. In one implementation, the basic timeliness impact of service order i in fulfillment period t can be expressed as: ; in, This indicates the basic timeliness impact of service order i within the fulfillment period t. This indicates the fulfillment period corresponding to the service time limit. This indicates the growth coefficient affected by timeliness, and When the service deadline has not yet been reached, the basic timeliness impact can be set to zero. When the service deadline has been reached or exceeded, the basic timeliness impact increases with the waiting period, resulting in a higher timeliness impact for long-backlogged service orders.

[0052] When considering the possibility that the first visit may not be completed, the expected timeliness impact that the current candidate execution relationship can mitigate can be determined using the probability of the first visit being completed. The expression is as follows: ; in, This represents the expected reduction in timeliness impact when service order i is executed by available personnel w. When the probability of initial visit completion is high, the current execution relationship can significantly reduce the impact of delays. When the probability of initial visit completion is low, even if the current service order is scheduled, waiting impacts may continue due to the initial visit not being completed. For service orders that have not yet reached their service deadline but are close to it, a non-zero advance timeliness impact can be set based on the remaining fulfillment period to differentiate between different levels of urgency; however, this advance timeliness impact is lower than that of service orders that have already reached their service deadline.

[0053] The status adjustment parameter is used to change the relative roles of resource consumption and timeliness impact in the candidate service cost representation value according to the current performance status. In one implementation, the candidate service cost representation value can be expressed as: ; in, This represents the cost representation value of the candidate service. Represents the set of evaluation dimensions. Indicates the candidate execution relationship in the evaluation dimension The amount of influence on, This indicates the corresponding weight determined by the state adjustment parameters. Evaluation dimensions include at least resource consumption and timeliness impact. When the resource consumption weight is large, there is a greater tendency to select candidate execution relationships with lower incremental path consumption, shorter candidate job duration, and lower risk of re-visiting. When the timeliness impact weight is large, there is a greater tendency to process service orders that have reached or are close to their service time limits.

[0054] Before weighting, the evaluation dimensions can be standardized to avoid any one dimension dominating in the long term due to differences in numerical ranges, such as time consumption, probability, and timeliness. The smaller the candidate service cost representation value, the lower the expected overall cost of the corresponding candidate execution relationship under the current performance status.

[0055] In step S106, the candidate service cost representation value with the smallest value is selected from the candidate execution relationships that satisfy the available job duration constraint as the target execution relationship. After determining the target execution relationship, the temporary service path, the candidate job duration, and the performance status representation are updated until all service work orders have a target execution relationship, or the remaining candidate execution relationships do not satisfy the available job duration constraint, thus obtaining the target performance plan.

[0056] Then, the service work order corresponding to the target execution relationship is inserted into the current path of the corresponding available fulfillment personnel. After the insertion is completed, the incremental path occupancy, candidate job duration and candidate service cost representation value of the remaining candidate execution relationships are re-determined.

[0057] When determining the target execution relationship, a successive evaluation method can be used to screen the remaining service work orders. For each service work order for which no executor has been assigned, the candidate execution relationships formed when executed by different available personnel are evaluated, and the corresponding candidate service cost representation value is calculated in conjunction with the current temporary service path. Satisfying the available job duration constraint means that after adding the service work order to the current path of the corresponding personnel, the estimated time required for that personnel to complete all path tasks does not exceed their remaining available job duration.

[0058] The candidate execution relationship with the lowest candidate service cost representation value indicates that, under the current performance status, this execution method can achieve a lower overall performance cost. Since the candidate service cost representation value considers both resource consumption and timeliness impacts, the selected target execution relationship does not necessarily correspond to the shortest path or the closest distance service arrangement. Instead, it is a comprehensive evaluation result formed under the combined effects of current personnel capabilities, service time limits, path structure, and the risk of revisiting.

[0059] In one implementation, the selection process for the target execution relationship can be implemented using an incremental insertion method. For a service order i and an available fulfillment worker w, the process iterates through the different positions of the current path of the available fulfillment worker w, inserting service order i into each position. The candidate service cost representation value corresponding to each insertion position is calculated, and the position that satisfies the path feasibility constraint and has the smallest candidate service cost representation value is selected as the target insertion position. By using a successive insertion method, a full search of all service orders and all path combinations can be avoided, reducing the computational load.

[0060] The path change before and after inserting a service order can be represented by incremental path occupancy. If the current path is X, inserting service order i at position j will form the path. The corresponding incremental path occupancy can be expressed as: ; in, This represents the incremental path occupancy generated after service order i is inserted at position j. This indicates the total path usage after inserting the service order. This indicates the total path occupancy before insertion. The total path occupancy can be determined based on the estimated travel time between service locations and the on-site operation time corresponding to each service order.

[0061] After determining the target execution relationship, the corresponding service work order is inserted into the current path of the available fulfillment personnel, and the remaining available job time for that fulfillment personnel is updated. Since adding a new service work order will change the arrival order, path location, and available time of subsequent service work orders, it is necessary to redetermine the incremental path occupancy and candidate job time corresponding to the remaining candidate execution relationships.

[0062] During the update process, the personnel, vehicle, and work order information in the current path can be retrieved again. The updated fulfillment context is then input into the on-site operation time prediction model to correct the candidate operation times for the remaining candidate execution relationships. If a service unit corresponding to a candidate execution relationship is in a continuous operation path, its on-site operation time may be affected by the completion status of the preceding service work order, the continuous operation status of personnel, and the path arrangement. Therefore, by re-predicting after the path update, the consistency between the candidate operation time and the actual fulfillment process can be improved.

[0063] Each time a target execution relationship is determined, the performance status representation is updated synchronously. The updates include the number of service orders with confirmed execution relationships, the number of remaining unassigned service orders, changes in the number of available performance personnel at each capability level, changes in the spatial distribution of service orders, and changes in service time limit status. The updated performance status representation is then re-input into the status adjustment model to obtain status adjustment parameters applicable to the current status, enabling subsequent candidate service cost representation values ​​to be dynamically adjusted according to changes in the performance process.

[0064] Once the target execution relationships for all service work orders have been determined, the service paths, service order sequences, and personnel allocation relationships corresponding to each available fulfillment personnel are summarized to form a target fulfillment plan. If the remaining service work orders cannot meet the available work time constraints of any available fulfillment personnel, the currently determined target execution relationships are used as the target fulfillment plan, and the service work orders that cannot be scheduled are reserved for processing in subsequent fulfillment cycles. This approach avoids increasing the probability of path timeouts, service delays, or repeat visits due to forcibly assigning service work orders beyond the personnel's work capacity.

[0065] In step S107, the service cost prediction result of the on-site fulfillment service is generated based on the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact in the target fulfillment plan.

[0066] The service cost forecast results include service work order level cost forecasts and target fulfillment plan level cost forecasts. The service work order level cost forecast is calculated by converting the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact of the corresponding service work order according to preset cost conversion parameters, and is risk-corrected based on the time interval confidence characterization. The target fulfillment plan level cost forecast is obtained by summing up the cost forecasts of each service work order level.

[0067] The service cost forecast results are used to characterize the comprehensive performance costs generated during the execution of the target performance plan. Because different service work orders have different path locations, on-site operational difficulties, initial visit completion status, and service time limits, a single time indicator is not directly used as the basis for cost evaluation. Instead, different types of occupancy and impact are converted to a unified evaluation scale.

[0068] Route occupancy represents the resource consumption incurred by fulfillment personnel moving between different service locations, and can be determined based on actual road distance, estimated travel time, traffic conditions, or historical travel data. On-site operation occupancy represents the operational resources consumed by fulfillment personnel to complete the current service order at a service location; its value can be determined by the updated candidate operation duration. Expected revisit occupancy represents the additional resource consumption incurred when services are rescheduled in subsequent fulfillment cycles due to the initial visit not being completed. Timeliness impact represents the service impact caused when a service order is not completed within the service time limit.

[0069] In one implementation, the service order-level cost forecast can be converted based on the unit cost parameters corresponding to different occupancy levels. The calculation method can be expressed as follows: ; in, This represents the predicted cost at the service order level for service order i. This indicates the resource consumption corresponding to the path. This indicates the resource consumption required for on-site operations. This indicates the resource consumption expected upon a subsequent visit. This indicates the cost impact corresponding to the timeliness effect. , , and These represent various cost conversion parameters.

[0070] Among them, route occupancy and on-site operation occupancy mainly reflect the fulfillment resources directly invested by the enterprise, expected repeat visit occupancy reflects the potential resource consumption caused by the risk of the first visit, and timeliness impact reflects the impact of failure to complete services on fulfillment quality. By setting cost conversion parameters separately, the degree of influence of various factors on the final cost prediction results can be adjusted according to different service business scenarios.

[0071] Because the candidate job durations output by the on-site job duration prediction model may have some prediction bias, a duration interval confidence characterization is used to adjust the service order-level cost prediction value for risk. The duration interval confidence characterization indicates the probability that the prediction result falls into different job duration intervals. When the probability of the prediction result corresponding to a longer job duration interval is high, it indicates that the service order has a higher time occupancy risk, and the corresponding cost prediction value can be increased. When the prediction results are concentrated in shorter job duration intervals, the amount of time risk correction can be reduced.

[0072] In one implementation, the risk-corrected service order-level cost forecast can be expressed as: ; in, This represents the risk-adjusted service order-level cost forecast. This represents the uncorrected service order-level cost forecast. This represents the risk adjustment factor. This represents the risk probability determined based on the confidence level of the time interval. The higher the risk probability, the greater the likelihood of a significant deviation in the on-site operation time, and the larger the corresponding adjustment range for the cost forecast.

[0073] The target fulfillment plan-level cost forecast is used to evaluate the estimated cost of the entire target fulfillment plan. Since the target fulfillment plan consists of target execution relationships corresponding to multiple service work orders, the cost forecasts at the service work order level can be aggregated to obtain the plan-level cost forecast.

[0074] ; in, This represents the projected cost for the target fulfillment plan, where n represents the number of service orders included in the target fulfillment plan. This represents the risk-adjusted cost forecast for the i-th service order.

[0075] After actual performance is completed, actual operation time, initial visit completion results, and actual route occupancy can be collected, and the actual results can be compared with the original prediction results. When there is a deviation between the actual operation time and the predicted operation time, or a difference between the initial visit completion result and the predicted probability, the actual performance data can be used to update subsequent training samples, enabling the on-site operation time prediction model and the state adjustment model to adapt to changes in service type, personnel capabilities, and the performance environment. Through continuous updates, the accuracy of service cost prediction results can gradually improve as the company accumulates performance data.

[0076] In one embodiment, the training samples of the on-site operation duration prediction model are formed by associating historical service work orders, historical personnel, historical service paths, and historical execution results; the object identifiers in the training samples are merged according to object categories, categorical fields are discretely encoded, continuous fields are scaled, high-dimensional fields are feature compressed, abnormal performance records are removed, and the sample distribution of different operation duration intervals is balanced.

[0077] The on-site operation duration prediction model is implemented using a multi-layer neural computing structure and includes a basic duration prediction unit, a condition correction unit, and an interval confidence unit. The basic duration prediction unit generates a basic operation duration based on object attributes, operation type, weight, volume, service area, and capacity requirements. The condition correction unit corrects the basic operation duration based on the capability level, operation experience, vehicles used, and work order order of the corresponding available personnel, generating the candidate operation duration. The interval confidence unit generates a duration interval confidence representation based on the output probability of different operation duration intervals.

[0078] The historical service work orders, historical service personnel, historical service paths, and historical execution results can be linked by work order number, service personnel number, vehicle number, and actual execution date, so that each training sample corresponds to a completed or terminated on-site service record. Historical service work orders provide object attributes, job type, service area, and capability requirements; historical service personnel provide capability level and work experience; historical service paths provide the order of service personnel, vehicles, and work orders; and historical execution results provide the actual on-site operation time. After the linking is completed, the fields scattered in personnel management data and work order management data are organized into sample data oriented towards a single service record.

[0079] Actual on-site operation time can be determined based on the arrival and departure times of the service personnel. Records with missing arrival or departure times, zero on-site operation time, on-site operation time exceeding the preset limit, or interruption due to vehicle malfunction or customer cancellation can be identified as abnormal performance records. When using manually entered duration as training labels, it can be verified in conjunction with location records, vehicle dwell records, and work order completion times to reduce label deviations caused by omissions, late reporting, or reporting according to the planned duration.

[0080] Object identifiers are used to distinguish specific goods, equipment, or services. When there are many types of objects, directly encoding all object identifiers would result in an excessive number of input fields. Therefore, object categories can be formed based on their purpose, size range, installation method, or operational complexity. Category fields can use one-hot encoding, so that each category corresponds to one input position. For fields with a large number of values ​​but similar technical meanings, category merging can be performed first, followed by discrete encoding, to reduce input dimensionality.

[0081] Standardization or interval scaling can be used to unify the scale of continuous fields, ensuring that data with different dimensions such as weight, volume, personnel experience, and work order order have similar value ranges. The scaling parameters are determined only based on the training data, and the same parameters are used for validation data and actual data to be predicted, avoiding the use of different scales for different datasets. For high-dimensional object categories, service areas, and capability requirements fields after discrete encoding, principal component analysis can be used for feature compression, converting multiple highly correlated fields into a smaller number of comprehensive features, reducing model complexity and training time while retaining key differences.

[0082] Historical work order fulfillment records are typically concentrated within shorter on-site operation duration intervals, while longer operation records are fewer. Direct training might cause the model to tend to output shorter duration results, reducing its ability to identify complex work orders. Balanced sample distribution can be achieved by increasing the sampling frequency of shorter duration intervals, setting interval loss weights, or replicating effective samples from fewer intervals. After the training data is balanced, the validation data maintains its original distribution to test the model's predictive ability under actual work order distributions.

[0083] The basic duration prediction unit uses information already determined when the service order is generated, without using the unformed personnel allocation results and route locations. Object attributes reflect the composition and processing difficulty of the service object; weight and volume reflect the handling and loading / unloading burden; job type and capacity requirements reflect the on-site operation content; and service area reflects the differences in building conditions, parking conditions, or working environment in different areas. The basic operation duration serves as the initial time estimate for on-site service, providing a time basis for the formation of temporary service routes.

[0084] After a temporary service path is formed, the condition correction unit receives the corresponding personnel's skill level, work experience, vehicles used, and work order sequence. The degree of matching between skill level and skill requirements affects the proficiency in completing the task; work experience reflects the personnel's familiarity with similar objects and tasks; vehicles used affect tool carrying, object loading and unloading, and on-site preparation; and work order sequence reflects the current work order's path position in the continuous execution process. The condition correction unit can output a correction amount or percentage relative to the basic task duration, and then generate candidate task durations corresponding to specific candidate execution relationships.

[0085] The number of nodes in the input layer of the multi-layer neural computation structure can be consistent with the number of input features after encoding and feature compression. Hidden layers are used to extract the nonlinear relationships between the input features. The output layer for continuous duration prediction can be configured with one output node to output the duration of the base job or candidate job in minutes. The number of hidden layers and nodes is determined based on the training sample size. In one optional approach, the number of nodes in each hidden layer can be set to half the number of nodes in the input layer. Training and validation data can be divided according to a preset ratio, and cross-validation is used to compare different numbers of hidden layers, activation methods, and training parameters. The interval confidence unit divides the on-site operation duration into multiple consecutive duration intervals and sets a corresponding output node for each interval. Each output node provides the probability that a candidate operation duration falls into the corresponding interval, and the output probabilities of all intervals constitute the duration interval confidence representation. Duration intervals can be divided according to ten minutes, fifteen minutes, or other time granularities used by the enterprise scheduling. When the interval width is small, it can provide a finer duration range, but it requires a larger sample size and higher prediction accuracy. When the interval width is large, the prediction stability is higher, but the ability to distinguish path times is correspondingly reduced.

[0086] When the output probability is mainly concentrated within the interval of the candidate job duration and its adjacent intervals, the prediction result can be considered to have a high degree of confidence. When the output probability is dispersed across multiple duration intervals, or when a higher output probability is found over a longer duration interval, it can be considered that the current service order has significant duration uncertainty, and the corresponding risk correction level should be increased in subsequent service cost predictions. By simultaneously outputting the duration of consecutive candidate jobs and the confidence representation of duration intervals, temporary service paths can be calculated using specific durations, and the service cost prediction results can retain a reflection of prediction uncertainty.

[0087] In one embodiment, the temporary service path and the candidate job duration are coordinated in an alternating update manner, including generating an initial service path using the basic job duration, determining the work order order and the personnel responsible for fulfilling the path based on the initial service path, inputting the personnel characteristics of the work order order and the personnel responsible for fulfilling the path into the condition correction unit, and re-verifying the feasibility of the path and adjusting the temporary service path based on the corrected candidate job duration, until the continuously updated personnel allocation relationship and work order order meet the preset stability conditions.

[0088] Establishing the temporary service path includes: using the empty path of each available fulfillment personnel starting from and returning to the service center as the initial path; for each service work order for which the target execution relationship has not yet been determined, determining the incremental path occupancy generated by inserting the service work order into the current path of the corresponding available fulfillment personnel; writing the candidate execution relationship that satisfies the available job duration constraint and has the smallest candidate service cost representation value into the temporary service path; and re-determining the candidate service cost representation values ​​of the remaining candidate execution relationships after writing.

[0089] The basic operation duration reflects only the initial site occupancy corresponding to the service work order's own attributes. It is used to form the initial service path before the fulfillment personnel and work order sequence are determined. Once the initial service path is determined, the fulfillment personnel, path location, preceding work order, and succeeding work order for each service work order can be obtained, and a fulfillment context can be formed accordingly. The condition correction unit adjusts the basic operation duration based on the fulfillment context, allowing the same service work order to have different candidate operation durations under different fulfillment personnel or different work order sequences.

[0090] The route fulfillment personnel are the available personnel corresponding to the current temporary service route. The work order order is the sequential arrangement of each service work order within the same route. When the work order order changes, the status of site preparation, object loading and unloading, tool retrieval, and continuous operation may change, thus requiring a re-determination of candidate operation durations. After the candidate operation duration changes, the total occupancy and remaining available operation duration of the corresponding route also change, and a previously feasible route may become an infeasible route.

[0091] Path feasibility verification includes calculating the estimated total time required for available personnel to complete all service orders within the current path. The estimated total time includes the path occupancy from the service center, movement between service locations, and return to the service center, as well as the candidate job durations for each service order within the path. If the estimated total time does not exceed the available job duration, the current path meets the feasibility requirements. If the estimated total time exceeds the available job duration, the insertion position of the relevant service orders can be changed, the order sequence adjusted, or the relevant service orders can be transferred to the paths of other available personnel.

[0092] Alternating updates can include two interconnected processes: path updates and duration updates. Path updates adjust personnel allocation and work order order based on the current candidate job duration, while duration updates correct the candidate job duration based on the adjusted path fulfillment personnel and work order order. After each duration update, path feasibility is re-verified, and temporary service paths are adjusted again if infeasible paths or lower-cost insertion locations exist.

[0093] The preset stability condition is that the personnel allocation relationship and work order order remain unchanged after two consecutive updates. To avoid repeated changes in the duration of individual candidate jobs between adjacent values, the difference between the durations of candidate jobs in two consecutive updates not exceeding a preset duration threshold can be used as an auxiliary stability condition. If the preset number of updates is reached but the stability condition is still not met, the update result with the lowest candidate service cost representation value and meeting the available job duration constraint can be retained.

[0094] When establishing temporary service paths, for each service order whose target execution relationship has not yet been determined, the results of inserting it into different positions within the current paths of each available fulfillment personnel are evaluated. When a service order is inserted into the middle of a path, the path occupancy between the original adjacent service locations is compared with the path occupancy after detouring through the new service location. When a service order is inserted into the beginning or end of a path, the evaluation is based on the path occupancy from the service center to the new service location or from the new service location to the service center, respectively. The position with the smallest increase in path occupancy among all insertion positions is selected as the candidate insertion position for the corresponding candidate execution relationship.

[0095] Once a candidate execution relationship is determined as the current target execution relationship, the corresponding service work order is inserted into the current path of the available personnel, and the personnel allocation result determined in this round is fixed. Since the path structure has changed, the incremental path occupancy generated when the remaining service work orders are inserted into the current path will also change. Therefore, the candidate service cost representation value before insertion cannot be used. It is necessary to recalculate the incremental path occupancy, candidate job duration, and candidate service cost representation value of the remaining candidate execution relationships.

[0096] The above process is executed in a successive insertion manner. In each round, only one target execution relationship with the lowest current overall cost and meeting the available job duration constraint is determined, and the updated temporary service path is used as the basis for the evaluation in the next round. Through round-by-round updates, the evaluation results of subsequent service work orders can reflect the established personnel allocation relationships and path structures, avoiding multiple service work orders being evaluated based on the same unupdated path at the same time, which could lead to job duration exceeding the limit or duplicate estimation of path occupancy.

[0097] In one embodiment, the state adjustment model is trained based on state transition training samples, which are formed from at least one of historical performance data and simulated performance data. During training, state adjustment parameters are sampled for the performance state representation. The state adjustment parameters are evaluated using the difference between the subsequent cumulative service cost and the state value estimate formed after adopting the state adjustment parameters, and the variation range of parameter sampling distribution in consecutive training rounds is limited, thereby updating the state adjustment model. The state adjustment parameters are used to determine the timeliness impact weight and resource occupancy weight, and the candidate service cost representation value is determined based on the weighted result of the expected resource occupancy and the weighted result of the expected timeliness impact.

[0098] The state transition training samples are organized according to continuous performance cycles. Each sample records at least the current performance status representation, the sampled state adjustment parameters, the performance plan formed using these parameters, the current cycle service cost, the performance status representation for the next performance cycle, and the subsequent cumulative service cost generated from the current cycle. State transition refers to the process by which the performance status changes from the current cycle to the next cycle under the combined effects of service work order allocation, path execution, new service work orders, changes in personnel availability, and the completion result of the first visit.

[0099] Historical performance data can be used to form state transition training samples based on actual continuous performance records. Simulated performance data can start from a preset initial performance state and generate a continuous performance process according to the service order arrival pattern, personnel availability status change pattern, and first visit completion probability, recording the status, parameters, costs, and next state corresponding to each performance cycle. Using simulated performance data can supplement historical data with less frequently occurring states such as work order concentration, personnel shortage, long-distance work order backlog, and high risk of repeat visits.

[0100] The state conditioning model can include a parameter generation network and a state value estimation network. The parameter generation network receives the performance state representation and outputs a parameter sampling distribution, while the state value estimation network receives the same performance state representation and outputs the cumulative service cost expected to be generated starting from the current performance state. When the training epochs are k, the sampling process of the state conditioning parameters can be represented as follows: ; in, This represents the performance status characteristic corresponding to the performance period t. Indicated by network parameters Given a defined conditional parameter sampling distribution, This represents the state adjustment parameter obtained from the sampling. The state adjustment parameter is limited to between 0 and 1.

[0101] The parameter sampling distribution can adopt a normal distribution whose mean varies with the performance status, and its expression is: ; ; in, Indicates the parameter generation network, This represents the mean of the distribution obtained based on the current performance status. This represents the sampling standard deviation in the k-th training epoch. A larger sampling standard deviation is used to expand the parameter search range, while a smaller sampling standard deviation is used for stable updates near the already obtained optimal parameters. The sampling results, after boundary constraints, are used as the state adjustment parameters for the current performance cycle.

[0102] After completing the continuous fulfillment process using state adjustment parameters, the cumulative service costs incurred from the current fulfillment state are calculated, and the corresponding state value estimate is obtained using a state value estimation network. The parameter evaluation quantity can be expressed as: ; in, This represents the evaluation quantity of the state conditioning parameters. This represents the cumulative service cost estimate of the state value estimation network for the current performance state. This represents the actual cumulative service cost incurred after adopting the sampling parameters. This represents the network parameters of the state value estimation network in the k-th training epoch. When the actual cumulative service cost is lower than the state value estimate, the evaluation value is positive, increasing the sampling probability of this state adjustment parameter in the corresponding fulfillment state. When the actual cumulative service cost is higher than the state value estimate, the evaluation value is negative, decreasing the sampling probability of this state adjustment parameter.

[0103] The state value estimation network can be updated according to the following formula: ; in, This represents the set of state transition training samples obtained in the k-th training round. Indicates the number of samples. This represents the state value error function, used to measure the difference between the state value estimate and the subsequent cumulative service cost.

[0104] To limit the variation in parameter sampling distribution across consecutive training rounds, the ratio of sampling density to the same state adjustment parameters before and after the update can be calculated: ; The sampling density ratio is constrained by a truncated objective function: ; in, Indicates the parameter limiting the distribution variation. This means limiting the sampling density ratio within a preset range. This limitation can prevent excessive changes in the parameter sampling distribution during a single training session, reducing the possibility of training instability when the state value estimate has not been fully updated.

[0105] The parameter generation network can be updated according to the following formula: ; Through multiple rounds of updates, the sampling density of state adjustment parameters that result in lower subsequent cumulative service costs is increased, while the sampling density of state adjustment parameters that result in higher subsequent cumulative service costs is decreased.

[0106] During training, the scale of each dimension representing the performance status can be standardized, and the subsequent cumulative service cost can be scaled to a preset range to reduce the impact of different feature dimensions and cost fluctuations on network parameter updates. The sampling standard deviation of the state adjustment parameters can be gradually reduced with each training epoch, allowing the training process to transition from exploring parameters over a large range to stable learning near the optimal parameters.

[0107] After the state conditioning model has completed training, it no longer randomly samples the state conditioning parameters. Instead, it selects the parameters with the highest sampling density from the parameter distribution corresponding to the current performance state. The expression for this parameter is: ; When the state adjustment parameter is close to zero, it increases the influence of timeliness on the candidate service cost representation. When the state adjustment parameter is close to 1, it increases the influence of resource usage on the candidate service cost representation. The resource usage weight can be set to... The weight of the timeliness impact can be taken as 1- This ensures that the two weights change with the performance status and remain coordinated with each other.

[0108] In one implementation, after the actual performance is completed, the training samples are updated using the actual operation time, the first visit completion result, and the actual path occupancy, and the performance status representation for the next target performance cycle is established using incomplete service orders, newly added service orders, and the updated set of available performance personnel.

[0109] Actual completion of service fulfillment can be the end of the target fulfillment period, or the completion or confirmation that a relevant service order in the target fulfillment plan cannot be completed. Actual operation time can be determined based on the arrival and departure times of the service personnel at the service location. Actual route occupancy can be determined based on vehicle location records, mobile terminal trajectories, arrival times of route nodes, or itinerary records submitted by the service personnel. The initial visit completion result is used to mark whether the service order was completed on the first visit. Service orders that were not completed due to insufficient capabilities, missing tools, missing parts, or unmet on-site conditions are recorded as requiring a second visit.

[0110] When updating training samples, the actual operation duration is replaced or supplemented with the duration label of the corresponding candidate execution relationship. The actual personnel, vehicles, and work order order are used as performance conditions, and the correspondence between predicted operation duration, duration interval confidence representation, and actual operation duration is retained. The initial visit completion results are used to correct risk records corresponding to different combinations of capability requirements, capability levels, and operational experience. Actual path occupancy is used to update travel records between different service areas, time periods, and path nodes, so that subsequent incremental path occupancy can reflect the recent performance environment.

[0111] For performance records with significantly abnormal actual operation times, missing location records, or unconfirmed work order status, they can be marked as data to be verified and not directly added to the valid training samples. Records confirmed to be normal performance discrepancies will be retained, while records confirmed to be due to incorrect reporting, equipment failure, or non-operational reasons will be removed. New training samples can be added to the model update according to a preset period or after reaching a preset number, to avoid frequent changes in model parameters caused by a small number of accidental records.

[0112] Unfinished service orders include those without execution arrangements, those arranged but not completed in the current cycle, and those requiring a second visit due to an incomplete initial visit. When establishing the performance status representation for the next target performance cycle, retain the service location, service time limit, capacity requirements, waiting time, and initial visit results, and re-determine whether the service time limit has been met and the waiting level after meeting the service time limit based on the end time of the current cycle.

[0113] A new service order is a service request that enters the service system after the current target fulfillment cycle ends and needs to be processed in the next target fulfillment cycle. After merging new service orders with incomplete service orders, the number of service orders of different job types and different capacity requirements is recalculated, and the spatial distribution and service time limit status of service orders are updated. For service orders that require a second visit, the personnel from the previous fulfillment, the reason for incompleteness, and the time consumed can be retained for reference when forming candidate execution relationships in the next cycle.

[0114] The updated set of available personnel for the next target performance cycle is determined based on the attendance plan, remaining available work hours, changes in skill levels, and personnel status. Personnel on leave, undergoing training, temporarily suspended, or whose work time is occupied by other tasks are not included in the available personnel set for the corresponding cycle. Personnel characteristics are updated synchronously when they complete training, accumulate experience in similar tasks, or their skill levels change.

[0115] Once the performance status representation for the next target performance cycle is established, it serves as input for the next round of service cost prediction and target performance plan generation. This creates a continuous feedback loop between actual performance results, training sample updates, and the establishment of the next cycle's status, enabling subsequent candidate job durations, first-visit completion probabilities, and path occupancy predictions to gradually adapt to changes in service recipients, personnel capabilities, and the performance environment.

[0116] Based on the same line of thought, such as Figure 2 The diagram shown is a structural block diagram of an AI-based service cost prediction system provided in an embodiment of the present invention. The system includes: The feature construction module 201 is used to obtain the set of service work orders and the set of available performance personnel within the target performance cycle, and to establish the work order features of each service work order and the personnel features of each available performance personnel. The work order features include service location, service time limit, object attributes, job type and capability requirements, and the personnel features include quantified capability level, job experience and available job duration. The task indicator prediction module 202 is used to combine the work order features and the personnel features to form multiple candidate execution relationships, use a pre-trained on-site task duration prediction model to determine the candidate task duration of each candidate execution relationship, and determine the first visit completion probability based on the capability matching between the service work order and the available personnel. The path planning module 203 is used to establish a temporary service path based on the service location, the candidate job duration and the available job duration, determine the incremental path occupancy corresponding to each candidate execution relationship, and update the candidate job duration and the incremental path occupancy based on the personnel information and work order order in the temporary service path. The status representation module 204 is used to construct a performance status representation based on the service work order set, the available performance personnel set, the updated candidate job duration and the first visit completion probability, and input the performance status representation into a pre-trained status adjustment model to obtain status adjustment parameters for coordinating the impact of timeliness and resource consumption. The cost calculation module 205 is used to determine the expected resource consumption of each candidate execution relationship based on the incremental path consumption, the updated candidate job duration, and the expected revisit consumption caused by the failure of the first visit; to determine the expected timeliness impact of each candidate execution relationship based on the first visit completion probability and service time limit; and to determine the resource consumption weight and timeliness impact weight using the state adjustment parameters. The expected resource consumption and the expected timeliness impact are weighted and summed to obtain the candidate service cost representation value. The solution iteration module 206 is used to select the candidate service cost representation value with the smallest value from the candidate execution relationships that satisfy the available job duration constraint as the target execution relationship, and update the temporary service path, the candidate job duration and the performance status representation after determining the target execution relationship, until all service work orders have determined the target execution relationship, or the remaining candidate execution relationships do not satisfy the available job duration constraint, to obtain the target performance solution; The cost prediction module 207 is used to generate a service cost prediction result for the on-site fulfillment service based on the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact in the target fulfillment plan.

[0117] This system establishes candidate execution relationships based on service work orders and fulfilling personnel. Using an artificial intelligence model, it determines the appropriate on-site operation time for each candidate relationship and evaluates fulfillment resource consumption by considering personnel capability matching, path occupancy, and initial service completion status. This allows for corresponding cost predictions when the same work order is executed by different personnel or located at different path positions. By coordinating resource occupancy and timeliness impact based on fulfillment status, it avoids pursuing shorter paths that lead to high-risk revisits and reduces path dispersion and decreased personnel utilization caused by prioritizing urgent work orders. As the target execution relationship is determined, the operation time, path status, and cost representation are updated, ensuring that the resulting service cost predictions correspond to the actual personnel allocation and work order sequence. This improves the accuracy of predicting on-site labor occupancy, travel occupancy, revisit occupancy, and timeliness losses, providing a basis for enterprise personnel allocation, fulfillment plan selection, and labor cost control.

[0118] The specific details of the above system have been described in detail in the method section of the implementation plan. For any undisclosed details, please refer to the implementation plan of the method section, and therefore will not be repeated here.

[0119] The above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is also readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0120] It should be noted that although several modules or units of the system have been mentioned in the detailed description above, this division is not mandatory. In fact, according to exemplary embodiments of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0121] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0122] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An AI-based service cost prediction method, characterized in that, Includes the following steps: Obtain the set of service work orders and the set of available personnel within the target fulfillment period, establish the work order characteristics of each service work order and the personnel characteristics of each available personnel. The work order characteristics include service location, service time limit, object attributes, job type and capability requirements, and the personnel characteristics include quantified capability level, job experience and available job duration. The work order features and personnel features are combined to form multiple candidate execution relationships. A pre-trained on-site operation duration prediction model is used to determine the candidate operation duration of each candidate execution relationship. The probability of first visit completion is determined based on the capability matching between the service work order and the available personnel. A temporary service path is established based on the service location, the candidate job duration, and the available job duration. The incremental path occupancy corresponding to each candidate execution relationship is determined. The candidate job duration and the incremental path occupancy are updated based on the personnel information and work order order in the temporary service path. Based on the service work order set, the available fulfillment personnel set, the updated candidate job duration, and the first visit completion probability, a fulfillment status representation is constructed. The fulfillment status representation is then input into a pre-trained state adjustment model to obtain state adjustment parameters for coordinating the impact of timeliness and resource consumption. Based on the incremental path occupancy, the updated candidate job duration, and the expected revisit occupancy caused by the incomplete first visit, the expected resource occupancy of each candidate execution relationship is determined. Based on the first visit completion probability and service time limit, the expected timeliness impact of each candidate execution relationship is determined. The resource occupancy weight and timeliness impact weight are determined using the state adjustment parameters. The expected resource occupancy and the expected timeliness impact are weighted and summed to obtain the candidate service cost representation value. From the candidate execution relationships that satisfy the available job duration constraint, select the candidate service cost representation value with the smallest value as the target execution relationship. After determining the target execution relationship, update the temporary service path, the candidate job duration, and the performance status representation until all service work orders have a target execution relationship, or the remaining candidate execution relationships do not satisfy the available job duration constraint, to obtain the target performance plan. Based on the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact in the target fulfillment plan, the service cost prediction result for the on-site fulfillment service is generated.

2. The AI-based service cost prediction method according to claim 1, characterized in that, The training samples of the on-site operation duration prediction model are formed by associating historical service work orders, historical personnel, historical service paths, and historical execution results. The object identifiers in the training samples are merged according to object categories, categorical fields are discretely encoded, continuous fields are scaled, high-dimensional fields are feature compressed, abnormal performance records are removed, and the sample distribution of different operation duration intervals is balanced.

3. The AI-based service cost prediction method according to claim 2, characterized in that, The on-site operation duration prediction model is implemented by a multi-layer neural computing structure and includes a basic duration prediction unit, a condition correction unit, and an interval confidence unit; the basic duration prediction unit generates a basic operation duration based on object attributes, operation type, weight, volume, service area, and capacity requirements; The condition correction unit corrects the basic operation time based on the ability level, work experience, vehicles used, and work order sequence of the corresponding available personnel, and generates the candidate operation time. The interval confidence unit generates a duration interval confidence representation based on the output probability of different operation duration intervals.

4. The AI-based service cost prediction method according to claim 3, characterized in that, The temporary service path and the candidate job duration are coordinated in an alternating update manner, including generating an initial service path using the basic job duration, determining the work order order and the personnel responsible for fulfilling the path based on the initial service path, inputting the personnel characteristics of the work order order and the personnel responsible for fulfilling the path into the condition correction unit, and re-verifying the feasibility of the path and adjusting the temporary service path based on the corrected candidate job duration, until the continuously updated personnel allocation relationship and work order order meet the preset stability conditions.

5. The AI-based service cost prediction method according to claim 3, characterized in that, The service cost prediction results include service work order-level cost predictions and target fulfillment plan-level cost predictions. The service work order-level cost predictions are calculated by converting the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact of the corresponding service work order according to preset cost conversion parameters, and risk correction is performed based on the duration interval confidence characterization. The target fulfillment plan-level cost predictions are obtained by summing up the cost predictions of each service work order. After the actual fulfillment is completed, the training samples are updated using the actual operation time, first visit completion results, and actual path occupancy, and the fulfillment status characterization for the next target fulfillment cycle is established using incomplete service work orders, newly added service work orders, and the updated set of available fulfillment personnel.

6. The AI-based service cost prediction method according to claim 1, characterized in that, The probability of first visit completion is determined based on the degree of matching between the capability requirements and the capability level, as well as the operational experience, through a preset risk mapping relationship. The probability of first visit failure is the complementary probability of the first visit completion probability. The expected occupancy for a second visit is determined by calculating the path occupancy and on-site operation occupancy required for a second visit based on the probability of first visit failure.

7. The AI-based service cost prediction method according to claim 1, characterized in that, The performance status representation includes the current performance stage, the number of service orders of different job types, the number of service orders with different capacity requirements, the number of available performance personnel of different capacity levels, the spatial distribution of service orders, the number of service orders that have not reached the service time limit, the number of service orders that have reached the service time limit, and the waiting level after reaching the service time limit.

8. The AI-based service cost prediction method according to claim 7, characterized in that, The state adjustment model is trained based on state transition training samples, which are formed from at least one of historical performance data and simulated performance data. During training, state adjustment parameters are sampled for the performance state representation. The state adjustment parameters are evaluated using the difference between the subsequent cumulative service cost and the state value estimate formed after adopting the state adjustment parameters, and the variation range of parameter sampling distribution in consecutive training rounds is limited. The state adjustment model is updated accordingly. The state adjustment parameters are used to determine the timeliness impact weight and resource occupancy weight, and the candidate service cost representation value is determined based on the weighted result of the expected resource occupancy and the weighted result of the expected timeliness impact.

9. The AI-based service cost prediction method according to claim 1, characterized in that, Establishing the temporary service path includes: Using the empty path of each available fulfillment personnel starting from and returning to the service center as the initial path, for each service work order for which the target execution relationship has not yet been determined, the incremental path occupancy generated by inserting the service work order into the current path of the corresponding available fulfillment personnel is determined. The candidate execution relationship that satisfies the available job duration constraint and has the smallest candidate service cost representation value is written into the temporary service path, and after writing, the candidate service cost representation values ​​of the remaining candidate execution relationships are re-determined.

10. An AI-based service cost prediction system, the system comprising: The feature construction module is used to obtain the set of service work orders and the set of available performance personnel within the target performance cycle, and to establish the work order features of each service work order and the personnel features of each available performance personnel. The work order features include service location, service time limit, object attributes, job type and capability requirements, and the personnel features include quantified capability level, job experience and available job duration. The task indicator prediction module is used to combine the work order features and the personnel features to form multiple candidate execution relationships, use a pre-trained on-site task duration prediction model to determine the candidate task duration of each candidate execution relationship, and determine the first visit completion probability based on the capability matching between the service work order and the available personnel. The path planning module is used to establish a temporary service path based on the service location, the candidate job duration, and the available job duration, determine the incremental path occupancy corresponding to each candidate execution relationship, and update the candidate job duration and the incremental path occupancy based on the personnel information and work order order in the temporary service path. The status representation module is used to construct a performance status representation based on the service work order set, the available performance personnel set, the updated candidate job duration and the first visit completion probability, and input the performance status representation into the pre-trained status adjustment model to obtain status adjustment parameters for coordinating the impact of timeliness and resource consumption. The cost calculation module is used to determine the expected resource consumption of each candidate execution relationship based on the incremental path consumption, the updated candidate job duration, and the expected revisit consumption caused by the failure of the first visit. It also determines the expected timeliness impact of each candidate execution relationship based on the first visit completion probability and service time limit, and uses the state adjustment parameters to determine the resource consumption weight and timeliness impact weight. Finally, it weights and sums the expected resource consumption and the expected timeliness impact to obtain the candidate service cost representation value. The solution iteration module is used to select the candidate service cost representation value with the smallest value from the candidate execution relationships that satisfy the available job duration constraint as the target execution relationship, and update the temporary service path, the candidate job duration and the performance status representation after determining the target execution relationship, until all service work orders have determined the target execution relationship, or the remaining candidate execution relationships do not satisfy the available job duration constraint, to obtain the target performance solution; The cost prediction module is used to generate a service cost prediction result for the on-site fulfillment service based on the path occupancy, on-site operation occupancy, expected revisit occupancy, and timeliness impact in the target fulfillment plan.

Images