Model training method and device and prediction method of work order delivery time

Abstract

The invention discloses a model training method and device and a prediction method of work order delivery time. The method comprises the steps that a plurality of feature sequences corresponding to a plurality of historical work orders are acquired, and the feature sequences at least comprise the distribution time, the construction time, the installation reservation time, the work order starting time and the delivery time of the work orders; sequentially inputting the plurality of feature sequences into a neural network model according to a time sequence, predicting the target probability of the distribution time of the historical work order corresponding to each feature sequence in the installation reservation time through forward propagation, and recursively updating the target probability; and under the condition that a loss function between the target probability and the real probability meets a preset convergence condition, training of the neural network model is completed. According to the method and the device, the technical problem that the prediction accuracy of the work order delivery time is relatively low due to the fact that the influence of the overtime work order on the time consumption of the undelivered work order is not considered in a related prediction method of the work order delivery time is solved.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and more specifically, to a model training method, device, and a method for predicting work order delivery time. Background Art

[0002] In the current telecommunications services, logistics distribution, maintenance and installation industries, work order management is a key link to ensure service quality and efficiency. However, the timeout problem in the work order delivery process has always been a pain point. It not only affects the customer experience, but also may cause waste of enterprise resources and increase costs. The traditional work order timeout prediction method mainly relies on the threshold judgment of a single work order, that is, judging whether the work order may time out based on the difference between the average processing time of the work order and the customer's appointment time. Although this method is simple and easy to implement, it has obvious limitations when dealing with multiple work orders in a continuous delivery scenario. Specifically, the traditional method ignores the mutual influence between work orders, that is, the timeout of a work order may have a chain reaction, affecting the on-time delivery of subsequent work orders. For example, if the installation and maintenance personnel encounter unforeseen complex situations when processing the first work order, resulting in a time consumption exceeding expectations, then this will directly affect the delivery time of subsequent work orders, and even cause a series of timeouts.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a model training method, device and work order delivery time prediction method to at least solve the technical problem of low accuracy in predicting work order delivery time due to the fact that the relevant work order delivery time prediction method does not consider the impact of time consumed by timed work orders on undelivered work orders.

[0005] According to one aspect of an embodiment of the present application, a model training method is provided, including: obtaining multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequences include at least: delivery time, construction time, installation appointment time, work order start time and delivery time of the work order; inputting the multiple feature sequences into a neural network model in chronological order, predicting the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation, and recursively updating the target probability; and completing the training of the neural network model when the loss function between the target probability and the true probability meets a preset convergence condition.

[0006] Optionally, the target probability of the delivery time of the historical work order corresponding to each feature sequence within the installation appointment time is predicted by forward propagation, and the target probability is recursively updated, including: according to the installation appointment time of the first historical work order, the starting time in the installation appointment time of the first historical work order, the preset start time, and the delivery time of the first historical work order, the target probability of the first historical work order is determined; according to the installation appointment time of the nth historical work order, the end time in the installation appointment time of the nth historical work order, and the starting time in the installation appointment time of the n-1th historical work order , the delivery time of the nth historical work order, the construction time of the n-1th historical work order, determine the target probability of the nth historical work order, where n is a positive integer greater than 1; according to the deadline in the installation appointment time of the n+1th historical work order, the start time in the installation appointment time of the nth historical work order, the delivery time of the n+1th historical work order, the construction time of the nth historical work order, the installation appointment time of the nth historical work order, the target probability of the nth historical work order, and the installation appointment time of the n+1th historical work order, determine the target probability of the n+1th historical work order.

[0007] Optionally, the target probability of the first historical work order is determined using the following formula: Among them, PV1 is the target probability of the first historical work order, Y1 is the installation appointment time of the first historical work order, and Y end(1) is the starting time of the installation appointment time of the first historical work order, CurT is the preset start time, and p1 is the delivery time of the first historical work order.

[0008] Optionally, when the target probability of the n-1th historical work order is greater than 1, the target probability of the nth historical work order is determined using the following formula: Among them, PV n is the target probability of the nth historical work order, Y n The installation appointment time for the nth historical work order, Y end(n) Y is the deadline of the installation appointment time of the nth historical work order. star(n-1) is the starting time of the installation appointment time of the n-1th historical work order, P n is the delivery time of the nth historical work order, M (n-1) It is the construction time of the n-1th historical work order.

[0009] Optionally, when the target probability of the nth historical work order is in the interval (0,1), the target probability of the n+1th historical work order is determined using the following formula: Among them, PV n+1 is the target probability of the n+1th historical work order, Y end(n+1)Y is the deadline of the installation appointment time of the n+1th historical work order. star(n) is the starting time of the installation appointment time of the nth historical work order, P n+1 is the delivery time of the n+1th historical work order, M (n) is the construction time of the nth historical work order, Y (n) The installation appointment time for the nth historical work order, PV (n) is the target probability of the nth historical work order, Y n+1 Schedule the installation time for the n+1th historical work order.

[0010] Optionally, before obtaining the characteristic sequence of historical work orders, the method also includes: determining the delivery time between physical grid units based on the physical grid to which the work order address in the historical work order belongs and the continuous delivery information of multiple work orders; determining the first physical grid unit to which the start delivery position of the historical work order belongs based on the work order delivery track of the delivery object in the historical work order, and determining the second physical grid unit to which the end delivery position of the historical work order belongs; determining the delivery time of the historical work order based on the first physical grid unit, the second physical grid unit and the delivery time between the physical grid units.

[0011] According to another aspect of the embodiments of the present application, a method for predicting the delivery time of a work order is also provided, including: obtaining multiple work orders to be predicted; inputting the multiple work orders to be predicted into a neural network model in chronological order, and obtaining the target probability of the delivery time of each work order to be predicted output by the neural network model being within the installation appointment time, wherein the neural network model is obtained by training using the above model training method.

[0012] Optionally, after obtaining the target probability of the delivery time of each work order to be predicted within the installation appointment time output by the neural network model, the following steps can also be performed: if the target probability of any work order is in the interval (0,1), and the target probability of the work order is less than a preset threshold, a first alarm message is generated, and the first alarm message is sent to the delivery object; if the target probability of any work order is less than 0, a second alarm message is generated, and the work order information of the work order and the second alarm message are sent to the management platform and the delivery object.

[0013] According to another aspect of the embodiment of the present application, a model training device is also provided, including: an acquisition module, used to acquire multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequence includes at least: the delivery time, construction time, installation appointment time, work order start time and delivery time of the work order; a prediction module, used to input the multiple feature sequences into the neural network model in chronological order, predict the target probability of the delivery time of the historical work order corresponding to each feature sequence within the installation appointment time through forward propagation, and recursively update the target probability; a training module, used to complete the training of the neural network model when the loss function between the target probability and the true probability meets the preset convergence condition.

[0014] According to another aspect of an embodiment of the present application, a non-volatile storage medium is also provided, and the storage medium includes a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the above model training method.

[0015] According to another aspect of an embodiment of the present application, an electronic device is provided, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the above model training method is executed when the program is running.

[0016] According to another aspect of the embodiments of the present application, a computer program is also provided, wherein the above model training method is implemented when the computer program is executed by a processor.

[0017] According to another aspect of the embodiments of the present application, a computer program product is also provided, which includes a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above model training method is implemented.

[0018] In an embodiment of the present application, a plurality of feature sequences corresponding to a plurality of historical work orders are obtained, wherein the feature sequence includes at least: the delivery time, construction time, installation appointment time, work order start time and delivery time of the work order; the plurality of feature sequences are input into the neural network model in chronological order, and the target probability of the delivery time of the historical work order corresponding to each feature sequence within the installation appointment time is predicted through forward propagation, and the target probability is recursively updated; when the loss function between the target probability and the true probability satisfies the preset convergence condition, the training of the neural network model is completed, so as to achieve the purpose of fully considering the influence of the time consumption of the time-consuming work order of the time-consuming work order on the undelivered work order in the process of predicting the work order delivery time, thereby achieving the technical effect of improving the prediction accuracy of the work order delivery time, and further solving the technical problem of low prediction accuracy of the work order delivery time caused by the fact that the prediction method of the relevant work order delivery time does not consider the influence of the time consumption of the time-consuming work order on the undelivered work order. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 is a flow chart of a model training method according to an embodiment of the present application;

[0021] Figure 2 is a flow chart of a method for predicting work order delivery time according to an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of a work order status process according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a work order delivery process according to an embodiment of the present application;

[0024] Figure 5 It is a schematic diagram comparing the differences in business capabilities of a method for predicting work order delivery time according to an embodiment of the present application;

[0025] Figure 6 is a structural diagram of a model training device according to an embodiment of the present application;

[0026] Figure 7 It is a hardware structure block diagram of a computer terminal according to a model training method of an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] According to an embodiment of the present application, a method embodiment of a model training method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Figure 1 is a flow chart of a model training method according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0031] Step S102, obtaining multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequences at least include: the delivery time, construction time, installation appointment time, work order start time and delivery time of the work order.

[0032] A work order, also known as a work order or service request, is a document or electronic record used within an enterprise or service industry to record and manage specific work tasks. In the telecommunications industry, work orders play a vital role. They are not only the carrier of service requests, but also the baton of the entire service process, involving the whole process management from the receipt of customer requests, the assignment of tasks, execution to final completion. The role of work orders in the telecommunications industry can be summarized as follows: When a customer makes a service request through telephone, email, online chat or customer service center, such as installing a new broadband line, fixing a network fault, replacing equipment, etc., the work order system will create a corresponding record, detailing the customer's needs, location, contact information and any special requirements. The work order system automatically or manually assigns the work order to the most appropriate installation and maintenance personnel or service team based on information such as the nature, priority and geographical location of the request. The work order system provides real-time work order status tracking, from work order creation, receipt, preparation, execution to final completion, each stage has clear records and timestamps.

[0033] In this step, a series of feature data related to work order delivery are extracted from the historical work order database. These data constitute a feature sequence for subsequent model training. The feature sequence contains at least the following information: Delivery time: refers to the time from the start of work order processing to the arrival of the installation and maintenance personnel at the customer's designated location. It is a key indicator to measure the efficiency of work order processing. Construction time: The time from the arrival of the installation and maintenance personnel at the customer's location to the start of construction to the completion of construction, reflecting the time-consuming actual installation or maintenance process. Installation appointment time: The construction start time scheduled by the customer is an important basis for work order processing time planning. Work order start time: The specific time point when the work order is assigned to the installation and maintenance personnel, which is used to calculate the total time for work order processing. Work order delivery time: The time when the work order is marked as completed, including the completion of construction and the end time of all subsequent review processes.

[0034] Step S104, multiple feature sequences are input into the neural network model in chronological order, and the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time is predicted through forward propagation, and the target probability is recursively updated.

[0035] In this step, the feature sequence extracted in step S102 is used as input and sent to the pre-designed neural network model for processing. The neural network model uses a forward propagation algorithm to calculate the probability (target probability) that the work order will be delivered within the scheduled time. This target probability reflects the possibility that the work order can be delivered on time within the specified installation appointment time. Based on the feature sequence of the current work order, the model predicts the probability that it can be delivered before the appointment time, and recursively updates this predicted probability with the predicted results of subsequent work orders to reflect the impact of time dependence in the processing of continuous work orders. Specifically, the recursively updated target probability is achieved by using the predicted results of the current work order as the input condition for the prediction of subsequent work orders, and so on, until all work orders in the sequence have been processed. In this way, the model can take into account the order and dependencies between work orders and provide more accurate overtime risk predictions.

[0036] It should be noted that in step S104, through the forward propagation of the neural network, the model can learn the interdependence between different work orders. For example, the delivery time of work order A will affect the start time of work order B, and the processing of work order B will further affect work order C, and so on. This interdependence is encoded in the neural network, so that when the model predicts whether work order C is overdue, it will take into account the actual processing time of work orders A and B, rather than just their average processing time.

[0037] Step S104 introduces a mechanism for recursively updating the target probability. This means that every time a work order is processed, the model will adjust the predicted probability of timeout for subsequent work orders based on the latest work order processing results. For example, if the actual processing time of work order A is longer than expected, this will directly affect the predicted probability of work order B, because the start time of work order B will be postponed. By learning this dynamic change, the model can more accurately predict the timeout risk of subsequent work orders.

[0038] The neural network model in step S104 is, for example, a Transformer model. Step S104 can be implemented by the following method: construct a feature sequence, which contains key information of historical work orders, such as work order business type, customer appointment time, work order pickup time, work order door-to-door time, construction completion time, basic information of delivery personnel, grid-related information, etc. These features are organized into a sequence, and each element represents the state of a work order or a key time point in its processing. For the Transformer model, the feature of each work order can be regarded as a position in the sequence, and the order relationship between work orders (such as the order in time) is encoded into the feature representation at the time of model input through positional encoding to ensure that the model can understand the time dependency of work order processing. Once the feature sequence is constructed, it can be input into the Transformer model for forward propagation. During the forward propagation process, the model uses the encoder layer to process the input sequence. The attention mechanism in the encoder can capture the dependency between each work order in the sequence, so as to more accurately understand the dynamic process of work order processing.

[0039] The Transformer model takes each feature (work order status) in the sequence as input, and predicts the target probability of whether the delivery time of the work order is within the installation appointment time through a combination of a multi-head attention mechanism and a feed forward network.

[0040] The recursive update process is as follows: For the first work order in the sequence, the model predicts the target probability that its delivery time is within the installation appointment time. For the nth work order in the sequence (n>1), the model uses the predicted target probability of the n-1th work order, its actual construction time, and the delivery time of the nth work order to update the predicted target probability that the delivery time of the nth work order is within the installation appointment time.

[0041] Step S106, when the loss function between the target probability and the true probability meets the preset convergence condition, the training of the neural network model is completed.

[0042] During the model training phase, the loss function is calculated by comparing the target probability predicted by the model with the actual probability that the work order is indeed on time or overtime in the actual historical data. The loss function is an important indicator for evaluating the prediction accuracy of the model. The smaller the loss function, the smaller the gap between the model prediction result and the actual situation. This embodiment adopts the binary cross entropy loss function, which is suitable for prediction problems with binary classification as the target variable (i.e., whether the work order is on time), and can effectively measure the difference between the model prediction result and the actual label. The goal of model training is to adjust the parameters in the neural network to minimize the loss function, that is, the target probability predicted by the model is as close as possible to the actual delivery timeliness rate of the historical work order. When the value of the loss function changes slightly in continuous iterations and meets the preset convergence conditions, the model training process is considered complete. This means that the model has learned the relationship between the key features of the work order and the delivery timeliness rate, and can be used to predict the timeout risk of newly received work orders.

[0043] The loss function f(x,y) can be expressed by the following formula:

[0044]

[0045] Where n is the number of samples, x i is the true label (0 or 1) of the i-th sample, y i is the predicted probability of the ith sample.

[0046] Through the above steps, historical work order data can be effectively utilized to build a neural network model that can predict whether the work order will be delivered on time within the scheduled time, thereby providing installation and maintenance personnel with early warning of overtime risks, helping them to reasonably arrange the work order processing sequence and optimize customer service levels.

[0047] The following Figure 1 The steps shown are exemplary and explanatory.

[0048] According to some optional embodiments of the present application, predicting the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation and recursively updating the target probability can be achieved by the following method.

[0049] According to the installation appointment time of the first historical work order, the starting time of the installation appointment time of the first historical work order, the preset start timing time, and the delivery time of the first historical work order, the target probability of the first historical work order is determined.

[0050] According to the installation appointment time of the nth historical work order, the deadline in the installation appointment time of the nth historical work order, the start time in the installation appointment time of the n-1th historical work order, the delivery time of the nth historical work order, and the construction time of the n-1th historical work order, the target probability of the nth historical work order is determined, where n is a positive integer greater than 1.

[0051] According to the deadline of the installation appointment time of the n+1th historical work order, the start time of the installation appointment time of the nth historical work order, the delivery time of the n+1th historical work order, the construction time of the nth historical work order, the installation appointment time of the nth historical work order, the target probability of the n+1th historical work order, and the installation appointment time of the n+1th historical work order, the target probability of the n+1th historical work order is determined.

[0052] Specifically, the target probability of the first historical work order can be determined using the following formula: Among them, PV1 is the target probability of the first historical work order, Y1 is the installation appointment time of the first historical work order, and Y end(1) is the starting time of the installation appointment time of the first historical work order, CurT is the preset start time, and p1 is the delivery time of the first historical work order.

[0053] During the model training process, the target probability calculated by the above formula will be used as the label of the training data to guide the neural network model to learn how to predict the possibility of completing the work order within the appointment time based on the work order characteristics. In this way, the model can not only learn the processing characteristics of a single work order, but also understand the dynamic relationship between the work order processing time and the appointment time, so as to more accurately evaluate and predict the risk of overtime of the work order.

[0054] Specifically, when the target probability of the n-1th historical work order is greater than 1, the target probability of the nth historical work order is determined using the following formula: Among them, PV n is the target probability of the nth historical work order, Y n The installation appointment time for the nth historical work order, Y end(n) Y is the deadline of the installation appointment time of the nth historical work order. star(n-1) is the starting time of the installation appointment time of the n-1th historical work order, P n is the delivery time of the nth historical work order, M (n-1) It is the construction time of the n-1th historical work order.

[0055] For example, when the target probability of the first historical work order is greater than 1, the target probability of the second historical work order is determined using the following formula: Among them, PV2 is the target probability of the second historical work order, Y2 is the installation appointment time of the second historical work order, and Y end(2) The deadline for the installation appointment of the second historical work order, Y star(1) is the starting time of the installation appointment time of the first historical work order, P2 is the delivery time of the second historical work order, and M (1) It is the construction time of the first historical work order.

[0056] Specifically, when the target probability of the nth historical work order is in the interval (0,1), the target probability of the n+1th historical work order is determined using the following formula: Among them, PV n+1 is the target probability of the n+1th historical work order, Y end(n+1) Y is the deadline of the installation appointment time of the n+1th historical work order. star(n) is the starting time of the installation appointment time of the nth historical work order, P n+1 is the delivery time of the n+1th historical work order, M (n) is the construction time of the nth historical work order, Y (n) The installation appointment time for the nth historical work order, PV (n) is the target probability of the nth historical work order, Y n+1 Schedule the installation time for the n+1th historical work order.

[0057] For example, when the target probability of the second historical work order is in the interval (0,1), the target probability of the third historical work order is determined using the following formula:

[0058]

[0059] Among them, PV3 is the target probability of the third historical work order, Y end(3) The deadline for the installation appointment of the third historical work order, Y star(2) is the starting time of the installation appointment time of the second historical work order, P3 is the delivery time of the third historical work order, and M (2) is the construction time of the second historical work order, Y (2) Schedule the installation time for the second historical work order, PV (2) is the target probability of the second historical work order, and Y3 is the installation appointment time of the third historical work order.

[0060] During the recursive update process, the above formula can be used to dynamically update the target probability of each subsequent work order, taking into account the dependencies between work orders and the fluctuations in actual processing time, providing installation and maintenance personnel with a more accurate timeout risk prediction.

[0061] Optionally, before obtaining the characteristic sequence of historical work orders, the following steps can also be performed: determine the delivery time between physical grid units based on the physical grid to which the work order address in the historical work order belongs and the continuous delivery information of multiple work orders; determine the first physical grid unit to which the start delivery position of the historical work order belongs based on the work order delivery track of the delivery object in the historical work order, and determine the second physical grid unit to which the end delivery position of the historical work order belongs; determine the delivery time of the historical work order based on the delivery time between the first physical grid unit, the second physical grid unit and the physical grid units.

[0062] Preprocess the historical work order data to extract the address information of each work order and the physical grid unit associated with it. The historical work order data may contain key information such as the pickup point, customer address, pickup time, door-to-door time, and construction completion time of the work order. Then, by analyzing the delivery trajectory of the work order from one physical grid unit to another, the average delivery time and time distribution between different grid units are counted. For example, the average delivery time from physical grid A to physical grid B is counted. These inter-grid delivery time data will be used to build a delivery time database to provide a reference for the prediction of subsequent work orders. From the historical work order data, identify the delivery start location (first physical grid unit) and end location (second physical grid unit) of each work order. Then use the inter-grid time to calculate the work order delivery time.

[0063] The embodiment of the present application also provides a method for predicting the delivery time of a work order, comprising: obtaining a plurality of work orders to be predicted; inputting the plurality of work orders to be predicted into a neural network model in chronological order, and obtaining a target probability that the delivery time of each work order to be predicted output by the neural network model is within the installation appointment time, wherein the neural network model is based on Figure 1 The model training method shown is used for training.

[0064] In some optional embodiments, after obtaining the target probability of the delivery time of each work order to be predicted within the installation appointment time output by the neural network model, the following steps can also be performed: if the target probability of any work order is in the interval (0,1), and the target probability of the work order is less than a preset threshold, a first alarm message is generated, and the first alarm message is sent to the delivery object; if the target probability of any work order is less than 0, a second alarm message is generated, and the work order information of the work order and the second alarm message are sent to the management platform and the delivery object.

[0065] When the target probability of any work order output by the prediction model (i.e., the probability that the work order will be delivered within the appointment time) is in the interval (0,1) and is less than the preset threshold, the system will generate the first alarm information. The threshold here can be set according to business needs and historical data analysis. For example, if the threshold is set to 0.5, it means that when the model predicts that the probability of completion of the work order within the appointment time is less than 50%, an alarm will be triggered. The purpose of the first alarm message is to inform the installation and maintenance personnel in advance that a work order may not be completed in time, so that they can take measures before the actual service, such as optimizing the route, contacting the customer in advance to explain the situation, or requesting a change in the appointment time. The system sends the first alarm message to the installation and maintenance personnel so that they can adjust their work plans in time and reduce the occurrence of timeout events.

[0066] If the target probability of any work order output by the prediction model is less than 0, it means that the model predicts that the work order cannot be completed within the appointment time under any circumstances. In practical applications, the situation where the target probability is less than 0 rarely occurs because the probability value is usually limited to between 0 and 1. However, if this situation occurs due to the prediction error of the model or data anomalies, the system will generate a second alarm message, indicating that the work order has a very high risk of timeout. The second alarm message not only notifies the installation and maintenance personnel, but is also sent to the management platform. The management platform can be a company's dispatch center or customer service department, which is responsible for supervising the overall situation of work order processing and intervening when necessary, such as reallocating work orders, providing additional resources, or coordinating customer communications. By sending the work order information and the second alarm message to the management platform, management can be prompted to take more urgent measures to resolve potential timeout issues.

[0067] Figure 2 is a flowchart of a method for predicting work order delivery time according to an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0068] Step S201, historical work order keyword definition.

[0069] By analyzing the work order processing logs of the delivery personnel, the processing time point information of each work order is extracted, including the construction order acceptance time, pickup time, door-to-door time, completion time, and appointment time. At the same time, the physical grid units to which the work order pickup point and delivery address belong are recorded, which helps to understand the movement pattern and time consumption of work orders between different grids.

[0070] According to the pickup point grid and delivery address grid of the work order, the average delivery time between physical grids is calculated. For the delivery time within the grid, an error range is set, usually 3 minutes. By analyzing the continuous delivery of multiple work orders, the delivery time model between grids can be further refined. For example, the time from the pickup point to physical grid A, the time from physical grid A to physical grid B, etc. These data help to understand the time consumption pattern of installation and maintenance personnel moving between different grids.

[0071] Define order information. Order information includes the order acceptance time, pickup time, door-to-door time, completion time, and customer appointment time of the construction order. At the same time, considering the address route from the store to the customer's home, the delivery time and the time of the order on hand, this information is crucial for predicting the total time required for work order processing. Personnel information: including the current coordinates of the installation and maintenance personnel, the physical grid unit they are in, and the number of work orders on hand. This information can reflect the immediate working status and possible time pressure of the installation and maintenance personnel.

[0072] Define the installation time of the installation and maintenance personnel. Installation time: By calculating the difference between the completion time of the construction order and the door-to-door time, the installation time of each work order can be obtained. These data help to build an average installation time model for installation and maintenance personnel, and further improve the accuracy of delivery time prediction.

[0073] Figure 3 is a schematic diagram of a work order status process according to an embodiment of the present application, such as Figure 3 As shown, Figure 3 This paper mainly analyzes the key time points and related time consumption when installation and maintenance personnel process work orders. Through the analysis of these time points and time consumption, a more accurate work order processing model can be constructed. It is divided into the following parts:

[0074] 1. Analysis of installation time by installation and maintenance personnel: This is expressed as the completion time of the construction order minus the on-site time of the construction order. This part of time reflects the time required for the installation and maintenance personnel to complete the installation or maintenance service after arriving at the customer's address. The installation time is affected by many factors, including the specific business type of the work order (such as broadband installation, fault repair, etc.), the installation environment in the customer's home, the skill level and experience of the installation and maintenance personnel, etc.

[0075] 2. Analysis of travel time of installation and maintenance personnel: The delivery time from the pickup point to physical grid A reflects the time required for installation and maintenance personnel to reach the first grid unit where the customer is located from the pickup point. The delivery time from physical grid A to physical grid B reflects the time required for installation and maintenance personnel to move to the next grid unit after completing the service in one grid unit. Considering the time required for installation and maintenance personnel to move between multiple grids helps to understand the total time required for continuous processing of multiple work orders.

[0076] 3. Key time points: Construction order acceptance time: the time when the installation and maintenance personnel receive the new work order, that is, the time when the work order is assigned to the installation and maintenance personnel. Construction order pickup time: the time when the installation and maintenance personnel arrive at the designated pickup point and pick up the equipment or materials required for the work order. Construction order door-to-door time: the time when the installation and maintenance personnel arrive at the customer's address and prepare to start installation or maintenance services. Construction order completion time: the time when the installation and maintenance personnel complete the service, including all steps such as installation, maintenance, and testing.

[0077] 4. Other related factors: Pick up at the store: The installation and maintenance personnel may need to pick up the goods at the store, and the time taken for this step is part of the entire work order processing time. Store-customer home address route: The route selection from the pickup point to the customer's home affects the travel time of the installation and maintenance personnel. Delivery personnel arrival time: The time required to get from the installation and maintenance personnel's current location to the store to pick up the goods. Delivery personnel's on-hand order time: The time taken by the installation and maintenance personnel to process other work orders on hand affects the door-to-door time of new work orders. Customer appointment time: The service time agreed upon by the customer and the installation and maintenance personnel is an important basis for the work order processing time planning.

[0078] Step S202: pre-processing key data.

[0079] 1. Preprocessing of key data of work orders. Business type: Determine the type of service involved in the work order, such as broadband installation, maintenance, etc. This information can be automatically obtained through the work order system, or supplemented by manual classification. Delivery type: Classify the delivery method of the work order, such as direct door-to-door, pick up first and then door-to-door, etc. This helps to understand the differences in the work order processing process. Appointment time, pickup time, door-to-door delivery time, construction completion time: These time point data need to be standardized and converted into a unified time format, such as UNIX timestamp, so that the model can understand the time series. At the same time, check and process missing values ​​or outliers to ensure the integrity of the data. The physical grid unit where the delivery personnel and customer address are located: Encode the physical grid unit of the delivery personnel ID and the customer address, and use digital or category codes to represent it, so that the model can identify different delivery personnel and grids. Average construction time of work orders: Calculate the average construction time of work orders of different business types based on historical work order data. This will be used as one of the features to reflect the general law of construction time.

[0080] 2. Pre-processing of key data of installation and maintenance delivery personnel. Current number of work orders to be delivered: Record the number of unfinished work orders currently held by each installation and maintenance personnel, which directly affects their ability and efficiency in handling new work orders. Construction time for different business work orders: Analyze historical data and build a mapping table between business type and average construction time for each installation and maintenance personnel to reflect individual differences. Delivery skills: Encode the skill level or specific skills of the installation and maintenance personnel. For example, some personnel may have higher skills in broadband installation, while others may be better at troubleshooting.

[0081] 3. Preprocessing of key data of physical grids. Physical grid unit where the work order pickup is located, physical grid unit where the work order customer address is located: Encode the physical grid units of the pickup point and customer address for model identification and analysis of the delivery time between grids. Average delivery time between physical grid units: Based on historical work order data, calculate the average delivery time between adjacent physical grids and construct a delivery time matrix between grids.

[0082] 4. Calculation of delivery timeliness rate (PV(n)): The calculation of delivery timeliness rate PV(n) is based on the comparison between the delivery person's door-to-door delivery time and the appointment time. Specifically, PV(n) = available delivery time / appointment time, where available delivery time refers to the length of time available from the pickup time to the appointment time, and the appointment time refers to the time period between the appointment time and the current time. The closer the value of PV(n) is to 1, the higher the possibility that the delivery person will deliver within the appointment time; a value less than 1 indicates a risk of overtime.

[0083] Step S203: construct a historical work order overtime risk prediction model based on Transformer.

[0084] The input layer is responsible for converting the key feature information of the work order into a form that the model can understand. For each work order, including business type, delivery type, appointment time, pickup time, delivery time, construction completion time, delivery personnel ID, physical grid unit of the customer address, and average construction time of the work order, this information needs to be embedded in a sequence in the order of time series. In addition, the information of the installation and maintenance personnel, such as the current number of work orders to be delivered, the construction time of different business work orders, delivery skills, etc., as well as the key data of the physical grid are also serialized so that the model can identify the temporal sequence and dynamic changes in work order processing. The serialization process of the input layer usually involves feature vectorization, that is, mapping each feature to a numerical vector. For example, business type and delivery type can be converted into vectors through one-hot encoding, time features can be converted into timestamps or relative differences from the current time, and installation and maintenance personnel and physical grid information can also be converted into vectors through specific encoding methods. Finally, these vectors will be organized into a sequence as the input of the Transformer model.

[0085] The Transformer encoder is the core part of the model. It can capture the dependencies between each work order and other work orders in the input sequence through the self-attention mechanism. The self-attention mechanism allows the model to consider the information of all work orders in the sequence when processing each work order in the sequence, not just the work orders before and after it. This means that when the model predicts the on-time delivery rate of a work order, it will take into account the processing of all other work orders, including processed and pending work orders, and their possible impact on the target work order. The ability to capture this global dependency is crucial to understanding the changes in work order delivery time in the continuous processing of multiple work orders and predicting potential timeout risks. It enables the model to handle more complex time series problems and improve the accuracy of predictions.

[0086] The output layer is responsible for generating the model's prediction results, namely the delivery timeliness rate PV(n) of each work order. PV(n) is a probability value that indicates the probability of the nth work order being completed within the scheduled time. The closer this value is to 1, the higher the probability that the work order will be completed on time, and vice versa, there is a risk of overtime. The output layer contains a fully connected layer that maps the output of the Transformer encoder to the target probability value.

[0087] Step S204, PV(n) is calculated and updated.

[0088] Based on the order of work orders and the appointment time of the work orders, calculate the probability PV of each work order being delivered within the appointment time n .

[0089] Among them, appointment time: Y n =(Y end(n) -Y star(n) ); Current time: CurT; Delivery time of the nth order: P n ; Construction time of the nth order: M n .

[0090] For the first order: Probability of on-time delivery

[0091] For the nth order: the on-time delivery rate of the n-1th work order>1

[0092]

[0093] For the n+1th order: the on-time delivery rate of the nth work order is in the interval (0, 1)

[0094]

[0095] Step S205, optimizing loss function and model training.

[0096] The loss function is designed based on binary cross entropy to train the model to predict the delivery time P of each work order. n 、Construction time M n , Delivery Timely Rate PV n For PVs that need to be updated recursively n , which can be used as a serialization target.

[0097] During the training process, the current PV can be calculated through forward propagation n , and update the PV of the next step at each step n .

[0098]

[0099] Where f(x,y) is the loss function, n is the number of samples, and x i is the true label (0 or 1) of the i-th sample, y i is the predicted probability of the ith sample.

[0100] Step S206: model deployment and real-time prediction.

[0101] In actual use, the model can gradually input the characteristics of the current work order and the previous PV(n), and predict the timeout probability of the next work order.

[0102] Figure 4 is a schematic diagram of a work order delivery process according to an embodiment of the present application, such as Figure 4 As shown, when PV n >1, it means that the delivery time of the first n work orders is within the scheduled time range. n >0, it means that the current nth work order has PV n The probability is that the door will be on time. If it is lower than 0.5, there is a risk of overtime and an early warning is needed. At the same time, this probability PV n Bring in Calculate the next work order timeout probability. n <0, it means that the delivery time of the nth work order must exceed the appointment time range, and an alarm should be issued at this time that the work order has a timeout risk.

[0103] Figure 5 This is a schematic diagram comparing the differences in business capabilities of a method for predicting work order delivery time according to an embodiment of the present application. Figure 5 It can be seen that the difference between this application and the existing business process lies in: analysis and processing of key data of historical work orders; construction of a time-consuming model for the delivery of historical work orders; recursive prediction of the delivery of in-transit work orders; and risk assessment of overtime delivery of in-transit work orders.

[0104] Among them, the analysis and processing of key data of historical work orders is the first step in building a predictive model. It involves collecting and analyzing data generated during the processing of past work orders, including but not limited to business type, delivery type, appointment time, pickup time, door-to-door time, completion time, installation and maintenance personnel information, physical grid units of customer addresses, average construction time, etc. These data are preprocessed and cleaned to remove any invalid or abnormal work order records and converted into a format that can be understood by the machine learning model. Preprocessing also includes feature engineering, such as calculating the processing time of work orders, encoding the skill level of installation and maintenance personnel, and standardizing timestamps. By deeply mining historical work order data, the rules of work order processing time can be found, providing a data basis for building a predictive model.

[0105] The construction of the historical work order delivery time consumption model is to use the preprocessed historical data to train a model to predict the delivery time consumption of work orders. At this stage, advanced machine learning or deep learning technologies, such as the Transformer model, are usually used. The model uses the self-attention mechanism to effectively capture the temporal and spatial dependencies in the work order sequence and understand how the processing of a work order affects the delivery of subsequent work orders. By learning the delivery time consumption of historical work orders, the model can predict the processing time of future work orders under given conditions, which is the key to the subsequent overtime risk prediction.

[0106] Recursive prediction of in-transit work order delivery is a dynamic prediction method used to predict the probability of timeout of subsequent work orders in a sequence of multiple work orders being processed. Recursive prediction is based on the concept that the processing result of the current work order (whether it is completed on time) will affect the processing time of the next work order, and thus affect the timeout risk. Specifically, the model predicts the processing time of the next work order based on the information of the processed work order and the real-time status of the installation and maintenance personnel, and then uses this prediction result to update the timeout risk of the work order. This process will be repeated as the installation and maintenance personnel process each work order, forming a continuous prediction chain, which improves the accuracy and timeliness of the prediction.

[0107] The risk assessment of overtime delivery of in-transit work orders is based on the results of recursive prediction, and the probability of overtime risk of the work orders being processed is assessed. This step uses the delivery timeliness rate PV(n) as a key indicator. PV(n) reflects the probability of on-time completion of the nth work order within the predicted time range. By comparing PV(n) with the preset overtime risk threshold, it is possible to determine in real time whether the work order has a risk of overtime. If PV(n) is lower than the threshold, it indicates that the work order has a high probability of overtime. The system will automatically issue an early warning to prompt the installation and maintenance personnel or the dispatch center to take measures, such as adjusting the order of work orders, rescheduling installation and maintenance personnel, or communicating with customers to adjust the appointment time. Through real-time overtime risk assessment, it is possible to respond and optimize the work order processing process in a timely manner, reduce the occurrence of overtime events, and improve customer satisfaction and the work efficiency of installation and maintenance personnel.

[0108] By predicting the risk of work order timeouts, this application can take measures in advance to optimize the work order delivery process, avoid delays caused by timeouts, and improve the overall work efficiency of installation and maintenance delivery personnel. By reducing the incidence of work order timeouts, predicting and reducing work order timeouts can help improve customer experience, ensure that customer issues can be resolved in a timely manner, and enhance customer satisfaction and loyalty. By identifying potential risk factors, delivery personnel can take measures before problems occur, prioritize high-risk work orders, and ensure that critical work orders are handled in a timely manner, thereby reducing the occurrence of actual timeout events.

[0109] Figure 6 is a structural diagram of a model training device according to an embodiment of the present application, such as Figure 6 As shown, the device comprises:

[0110] The acquisition module 62 is used to acquire multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequence at least includes: the delivery time, construction time, installation appointment time, work order start time and delivery time of the work order.

[0111] The prediction module 64 is used to input multiple feature sequences into the neural network model in chronological order, predict the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation, and recursively update the target probability.

[0112] The training module 66 is used to complete the training of the neural network model when the loss function between the target probability and the true probability meets the preset convergence condition.

[0113] Optionally, the target probability of the delivery time of the historical work order corresponding to the feature sequence is predicted by forward propagation within the installation appointment time, and the target probability is recursively updated, including: determining the target probability of the first historical work order according to the installation appointment time of the first historical work order, the starting time in the installation appointment time of the first historical work order, the preset start time, and the delivery time of the first historical work order; determining the target probability of the first historical work order according to the installation appointment time of the nth historical work order, the end time in the installation appointment time of the nth historical work order, and the starting time in the installation appointment time of the n-1th historical work order, The target probability of the nth historical work order is determined based on the delivery time of the nth historical work order and the construction time of the n-1th historical work order, where n is a positive integer greater than 1; the target probability of the n+1th historical work order is determined based on the deadline in the installation appointment time of the n+1th historical work order, the start time in the installation appointment time of the nth historical work order, the delivery time of the n+1th historical work order, the construction time of the nth historical work order, the installation appointment time of the nth historical work order, the target probability of the nth historical work order, and the installation appointment time of the n+1th historical work order.

[0114] Optionally, the target probability of the first historical work order is determined using the following formula: Among them, PV1 is the target probability of the first historical work order, Y1 is the installation appointment time of the first historical work order, and Y end(1) is the starting time of the installation appointment time of the first historical work order, CurT is the preset start time, and p1 is the delivery time of the first historical work order.

[0115] Optionally, when the target probability of the n-1th historical work order is greater than 1, the target probability of the nth historical work order is determined using the following formula: Among them, PV n is the target probability of the nth historical work order, Y n The installation appointment time for the nth historical work order, Y end(n) Y is the deadline of the installation appointment time of the nth historical work order. star(n-1) is the starting time of the installation appointment time of the n-1th historical work order, P n is the delivery time of the nth historical work order, M (n-1) It is the construction time of the n-1th historical work order.

[0116] Optionally, when the target probability of the nth historical work order is in the interval (0,1), the target probability of the n+1th historical work order is determined using the following formula: Among them, PV n+1 is the target probability of the n+1th historical work order, Y end(n+1)Y is the deadline of the installation appointment time of the n+1th historical work order. star(n) is the starting time of the installation appointment time of the nth historical work order, P n+1 is the delivery time of the n+1th historical work order, M (n) is the construction time of the nth historical work order, Y (n) The installation appointment time for the nth historical work order, PV (n) is the target probability of the nth historical work order, Y n+1 Schedule the installation time for the n+1th historical work order.

[0117] Optionally, before obtaining the characteristic sequence of historical work orders, the following steps can also be performed: determine the delivery time between physical grid units based on the physical grid to which the work order address in the historical work order belongs and the continuous delivery information of multiple work orders; determine the first physical grid unit to which the start delivery position of the historical work order belongs based on the work order delivery track of the delivery object in the historical work order, and determine the second physical grid unit to which the end delivery position of the historical work order belongs; determine the delivery time of the historical work order based on the delivery time between the first physical grid unit, the second physical grid unit and the physical grid units.

[0118] It should be noted that the above Figure 6 The modules in the embodiment may be program modules (e.g., a set of program instructions for implementing a specific function) or hardware modules. For the latter, they may be expressed in the following forms, but are not limited thereto: the expression form of the above modules is a processor, or the functions of the above modules are implemented by a processor.

[0119] It should be noted that Figure 6 The preferred implementation of the illustrated embodiment can be found in Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0120] Figure 7 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a model training method. Figure 7 As shown, the computer terminal 70 may include one or more (shown as 702a, 702b, ..., 702n in the figure) processors 702 (the processor 702 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 704 for storing data, and a transmission module 706 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 7The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 7 More or fewer components as shown, or with Figure 7 Different configurations are shown.

[0121] It should be noted that the one or more processors 702 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 70. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0122] The memory 704 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the model training method in the embodiment of the present application. The processor 702 executes various functional applications and data processing by running the software programs and modules stored in the memory 704, that is, the above-mentioned model training method is realized. The memory 704 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 704 may further include a memory remotely arranged relative to the processor 702, and these remote memories may be connected to the computer terminal 70 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0123] The transmission module 706 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 70. In one example, the transmission module 706 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 706 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0124] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 70 .

[0125] It should be noted that, in some optional embodiments, the above Figure 7The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 7 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0126] It should be noted that Figure 7 The computer terminal shown is used to execute Figure 1 The model training method shown, therefore the relevant explanations in the execution method of the above commands are also applicable to the electronic device and will not be repeated here.

[0127] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above model training method.

[0128] A program for a non-volatile storage medium to perform the following functions: obtaining multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequences include at least: the delivery time, construction time, installation appointment time, work order start time, and delivery time of the work order; inputting the multiple feature sequences into a neural network model in chronological order, predicting the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation, and recursively updating the target probability; and completing the training of the neural network model when the loss function between the target probability and the true probability meets the preset convergence condition.

[0129] An embodiment of the present application also provides an electronic device, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the above model training method is executed when the program is running.

[0130] The processor is used to run a program that performs the following functions: obtaining multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequences include at least: the delivery time, construction time, installation appointment time, work order start time and delivery time of the work order; inputting the multiple feature sequences into the neural network model in chronological order, predicting the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation, and recursively updating the target probability; completing the training of the neural network model when the loss function between the target probability and the true probability meets the preset convergence condition.

[0131] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0132] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0133] In the above-mentioned embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary protection measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0134] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0135] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0136] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0137] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0138] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A model training method, characterized in that: include: Acquire multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequences at least include: the delivery time, construction time, installation appointment time, work order start time, and delivery time of the work order; Inputting the plurality of feature sequences into the neural network model in chronological order, predicting the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation, and recursively updating the target probability; When the loss function between the target probability and the true probability satisfies a preset convergence condition, the training of the neural network model is completed.

2. The method according to claim 1, characterized in that Predicting the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation, and recursively updating the target probability, including: Determine the target probability of the first historical work order according to the installation appointment time of the first historical work order, the starting time of the installation appointment time of the first historical work order, the preset start time, and the delivery time of the first historical work order; According to the installation appointment time of the nth historical work order, the deadline in the installation appointment time of the nth historical work order, the start time in the installation appointment time of the n-1th historical work order, the delivery time of the nth historical work order, and the construction time of the n-1th historical work order, the target probability of the nth historical work order is determined, where n is a positive integer greater than 1; Determine the target probability of the n+1th historical work order based on the deadline in the installation appointment time of the n+1th historical work order, the start time in the installation appointment time of the nth historical work order, the delivery time of the n+1th historical work order, the construction time of the nth historical work order, the installation appointment time of the nth historical work order, the target probability of the n+1th historical work order, and the installation appointment time of the n+1th historical work order.

3. The method according to claim 2, characterized in that The target probability of the first historical work order is determined using the following formula: Wherein, PV1 is the target probability of the first historical work order, Y1 is the installation appointment time of the first historical work order, and Y end(1) is the starting time of the installation appointment time of the first historical work order, CurT is the preset start timing time, and p1 is the delivery time of the first historical work order.

4. The method according to claim 2, characterized in that: When the target probability of the n-1th historical work order is greater than 1, the target probability of the nth historical work order is determined using the following formula: Among them, PV n is the target probability of the nth historical work order, Y n is the installation appointment time of the nth historical work order, Y end(n) is the deadline of the installation appointment time of the nth historical work order, Y star(n-1) is the starting time of the installation appointment time of the n-1th historical work order, P n is the delivery time of the nth historical work order, M (n-1) It is the construction time of the n-1th historical work order.

5. The method according to claim 2, characterized in that: When the target probability of the nth historical work order is in the interval (0,1), the target probability of the n+1th historical work order is determined using the following formula: Among them, PV n+1 is the target probability of the n+1th historical work order, Y end(n+1) is the deadline of the installation appointment time of the n+1th historical work order, Y star(n) is the starting time of the installation appointment time of the nth historical work order, P n+1 is the delivery time of the n+1th historical work order, M (n) is the construction time of the nth historical work order, Y (n) The installation appointment time of the nth historical work order, PV (n) is the target probability of the nth historical work order, Y n+1 The installation appointment time for the n+1th historical work order.

6. The method according to claim 1, characterized in that Before obtaining the feature sequence of the historical work order, the method further includes: Determine the delivery time between physical grid units according to the physical grid to which the work order address in the historical work order belongs and the continuous delivery information of multiple work orders; According to the work order delivery track of the delivery object in the historical work order, determine the first physical grid unit to which the start delivery position of the historical work order belongs, and determine the second physical grid unit to which the end delivery position of the historical work order belongs; The delivery time of the historical work order is determined according to the delivery time of the first physical grid unit, the second physical grid unit, and the delivery time between the physical grid units.

7. A method for predicting work order delivery time, characterized in that: include: Get multiple work orders to be predicted; In chronological order, the plurality of work orders to be predicted are input into the neural network model in sequence to obtain the target probability that the delivery time of each work order to be predicted output by the neural network model is within the installation appointment time, wherein the neural network model is obtained by training using the model training method described in any one of claims 1 to 6.

8. The method according to claim 7, characterized in that After obtaining the target probability that the delivery time of each of the to-be-predicted work orders output by the neural network model is within the installation appointment time, the method further includes: If the target probability of any work order is within the interval of (0,1), and the target probability of the work order is less than a preset threshold, a first alarm message is generated, and the first alarm message is sent to the delivery object; If the target probability of any work order is less than 0, a second alarm message is generated, and the work order information of the work order and the second alarm message are sent to the management platform and the delivery object.

9. A model training device, characterized in that: include: An acquisition module is used to acquire multiple feature sequences corresponding to multiple historical work orders, wherein the feature sequences at least include: the delivery time, construction time, installation appointment time, work order start time and delivery time of the work order; A prediction module, used to input the plurality of feature sequences into a neural network model in chronological order, predict the target probability that the delivery time of the historical work order corresponding to each feature sequence is within the installation appointment time through forward propagation, and recursively update the target probability; The training module is used to complete the training of the neural network model when the loss function between the target probability and the true probability meets a preset convergence condition.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the model training method described in any one of claims 1 to 6.

11. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the program executes the model training method described in any one of claims 1 to 6 when running.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the model training method described in any one of claims 1 to 6 is implemented.

Images