Prediction time correction method and device for part change task
By correcting the prediction to library time of the part change task using the time correction model, the problems of low prediction efficiency and low accuracy in the prior art are solved, and more efficient and accurate prediction of library time is achieved.
Patent Information
- Application Number
- CN202311000153.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-09
- Publication Date
- 2025-02-25
AI Technical Summary
In the prior art, the prediction of the in-store time of part change tasks is low and the accuracy is low. Experts rely on experience to predict the prediction time, and model prediction depends on complex training data and the amount of data is small, which makes it difficult to meet the model training requirements.
By obtaining the target task data of the target part change task, input it into the time correction model, using the time correction model obtained by training based on the sample part change task and the residual label training, the target time residual is calculated and the prediction time is corrected to obtain the final target time.
It improves the accuracy of prediction of part change tasks to the library time, reduces the accuracy requirements for the initial prediction time, and is suitable for the actual situation of vehicle production.
Smart Images
Figure CN119515594A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of vehicle manufacturing technology, and more particularly to a method for correcting the predicted time of a part change task. The present invention also relates to a device for correcting the predicted time of a part change task, a computing device, and a computer-readable storage medium. Background Art
[0002] The vehicle manufacturing process involves a large number of parts change tasks, and for these parts change tasks, it is necessary to predict the arrival time of parts in order to receive parts or arrange vehicle manufacturing plans. Currently, experts often predict the arrival time of parts in part change tasks based on production line engineering personnel and business requirements; or train prediction models to automatically predict the arrival time of parts.
[0003] However, based on the expert prediction of the arrival time of parts, experts need to spend a lot of time analyzing the part change tasks and make time predictions based on their own experience. This is highly dependent on the experts' own experience and is unable to make differentiated predictions for different change tasks. This is inefficient and has low accuracy, which greatly wastes manpower. Automatic prediction of the arrival time of parts based on models requires complex training data, and the prediction accuracy depends on the training effect of the model, which makes it difficult to guarantee the accuracy of the predicted arrival time of parts. Summary of the invention
[0004] In view of this, the embodiment of this specification provides a method for correcting the predicted time of a part change task. This specification also relates to a device for correcting the predicted time of a part change task, a computing device, and a computer-readable storage medium to solve the technical defects existing in the prior art.
[0005] According to a first aspect of an embodiment of this specification, a method for correcting the predicted time of a part change task is provided, the method comprising:
[0006] Acquire target task data of the target part change task;
[0007] Input the target task data into the time correction model to obtain the target time residual corresponding to the target part change task, wherein the time correction model is trained based on the sample task data and residual label of the sample part change task, and the residual label is the difference between the sample real arrival time and the sample predicted arrival time of the sample part change task;
[0008] The predicted arrival time of the target part change task is corrected according to the target time residual to obtain the target arrival time of the target part change task.
[0009] According to a second aspect of an embodiment of this specification, there is provided a predicted time correction device for a part change task, comprising:
[0010] An acquisition module is configured to acquire target task data of the target part change task;
[0011] An input module is configured to input the target task data into a time correction model to obtain a target time residual corresponding to the target part change task, wherein the time correction model is trained based on sample task data and residual labels of the sample part change task, and the residual label is the difference between the sample actual arrival time and the sample predicted arrival time of the sample part change task;
[0012] The correction module is configured to correct the predicted arrival time of the target part change task according to the target time residual to obtain the target arrival time of the target part change task.
[0013] According to a third aspect of an embodiment of this specification, a computing device is provided, including:
[0014] Memory and processor;
[0015] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the steps of the method for correcting the predicted time of the part change task.
[0016] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the method for correcting the predicted time of a part change task are implemented.
[0017] The predicted time correction method for a part change task provided in an embodiment of the present specification obtains target task data of the target part change task; inputs the target task data into a time correction model to obtain a target time residual corresponding to the target part change task, wherein the time correction model is trained based on sample task data and residual labels of sample part change tasks, and the residual label is the difference between the sample actual arrival time and the sample predicted arrival time of the sample part change task; the predicted arrival time of the target part change task is corrected according to the target time residual to obtain the target arrival time of the target part change task.
[0018] An embodiment of the present specification realizes that the difference between the sample actual arrival time and the sample predicted arrival time of the sample part change task is used as a residual label, and a time correction model is obtained based on the sample task data and residual label of the sample part change task. When it is necessary to predict the target arrival time of the target part change task, the target task data of the target part change task is input into the pre-trained time correction model to obtain the target time residual corresponding to the target part change task, and then the predicted arrival time of the target part change task is corrected according to the target time residual to obtain the final predicted target arrival time of the target part change task. In this way, the initial predicted arrival time can be automatically corrected based on the trained time correction model to obtain the final predicted target arrival time, thereby ensuring the accuracy of the final predicted target arrival time, and further correcting the initial predicted arrival time using the time correction model, thereby reducing the accuracy requirement of the initial predicted arrival time, which is more suitable for the actual production situation. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a flow chart of a method for correcting the predicted time of a part change task provided in an embodiment of this specification;
[0020] Figure 2 It is a schematic diagram of the training process of a time correction model in a prediction time correction method for a part change task provided in an embodiment of this specification;
[0021] Figure 3 It is a processing flow chart of a method for correcting the predicted time of a part change task applied to a vehicle production scenario provided in an embodiment of this specification;
[0022] Figure 4 It is a structural schematic diagram of a predicted time correction device for a part change task provided in an embodiment of this specification;
[0023] Figure 5 It is a structural block diagram of a computing device provided in an embodiment of this specification. DETAILED DESCRIPTION
[0024] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.
[0025] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.
[0026] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0027] First, the terms involved in one or more embodiments of this specification are explained.
[0028] It should be noted that the production and manufacturing process of vehicles involves a large number of parts change tasks. For the parts change tasks, it is necessary to predict the arrival time of the parts in order to receive the parts or arrange the production and manufacturing plan of the vehicle.
[0029] At present, experts often predict the time for part change tasks based on production line engineering personnel and business requirements. Experts need to spend a lot of time analyzing part change tasks and make time predictions based on their own experience. They rely heavily on the experts' own experience and cannot make differentiated predictions for different change tasks. This results in low efficiency and accuracy, and is extremely labor-intensive.
[0030] Alternatively, it is also possible to train a machine learning / deep learning time prediction model based on the industrial production time from a large-scale data source of the vehicle production process, and use the trained time prediction model to predict the arrival time of the part change task.
[0031] However, the time prediction model of machine learning / deep learning often requires a large amount of training data, and has high requirements for training data. It is a big data model and is more suitable for time prediction in the Internet big data scenario. In the vehicle production and manufacturing scenario, the time information is relatively simple and the amount of data is small, which makes it difficult to meet the data requirements for model training. Complex time prediction models such as machine learning / deep learning often have high requirements for data volume and data items, and require long-term, large-scale industrial production data. The training process is slow, inefficient and costly, and it is difficult to use directly in industry, and the scalability is poor. In addition, when the data source and data items change, the time prediction model of machine learning / deep learning needs to be retrained, which greatly increases the cost, and the scalability and compatibility are poor. Furthermore, the different dimensions of the input data may bring about the problem of unbalanced information, resulting in deviations in the trained model and reduced prediction accuracy.
[0032] Therefore, in order to solve the above technical problems, a method for correcting the predicted time of a part change task is provided in the embodiment of the specification, which realizes that the difference between the sample real arrival time and the sample predicted arrival time of the sample part change task is used as a residual label, and a time correction model is obtained based on the sample task data and the residual label of the sample part change task. When it is necessary to predict the target arrival time of the target part change task, the target task data of the target part change task is input into the pre-trained time correction model to obtain the target time residual corresponding to the target part change task, and then the predicted arrival time of the target part change task is corrected according to the target time residual to obtain the final predicted target arrival time of the target part change task. In this way, the initial predicted arrival time can be automatically corrected based on the trained time correction model to obtain the final predicted target arrival time, thereby ensuring the accuracy of the final predicted target arrival time, and further correcting the initial predicted arrival time using the time correction model, thereby reducing the accuracy requirement of the initial predicted arrival time, which is more suitable for the actual production situation.
[0033] In this specification, a method for correcting the predicted time of a part change task is provided. This specification also relates to a device for correcting the predicted time of a part change task, a computing device, and a computer-readable storage medium, which are described in detail one by one in the following embodiments.
[0034] Figure 1 102 is a flowchart of a method for correcting the predicted time of a part change task provided in an embodiment of the present specification, which specifically includes the following steps 102-106:
[0035] Step 102: Obtain target task data of the target part change task.
[0036] Specifically, a part change task refers to a task of obtaining and changing vehicle-related parts, which are parts required for the production process of the vehicle. The target part change task is any part change task initiated by the staff, such as the task of obtaining part A of the vehicle.
[0037] In addition, task data refers to at least one data item related to the part change task, such as task data including but not limited to part model, associated factory, associated vehicle model, process owner (i.e., process responsible person), change start time, etc., where change start time refers to the start time of the part change task, i.e., the application trigger time of the part change task. Target task data refers to the task data of the target part change task.
[0038] It should be noted that when making a part change, you need to first submit a request for the part change task. After approval, the corresponding parts will be issued. There is often a certain interval between the change start time and the final part arrival time. In order to receive parts or arrange vehicle production and manufacturing plans, etc., it is necessary to predict the part arrival time of the part change task.
[0039] In actual implementation, the prediction platform can receive a prediction request for a target part change task. After receiving the prediction request, the target task data of the target part change task can be obtained. The target task data can be carried in the prediction request or stored in a database. The corresponding target task data can be queried in the database based on the identification information carried in the prediction request.
[0040] In an optional implementation of this embodiment, the target task data includes a target change start time; after obtaining the target task data of the target part change task, the following is also included:
[0041] Determine a target time interval according to a time prediction model, wherein the target time interval is used to indicate the interval between the change start time and the time when the parts arrive at the warehouse;
[0042] According to the target time interval and the target change start time, the predicted arrival time of the target part change task is determined.
[0043] Specifically, the time prediction model refers to a pre-trained model for predicting the time interval between the change start time and the part arrival time in the warehouse. The time prediction model may be a deep learning model, a machine learning model, or a distribution function, etc.
[0044] In specific implementation, the target time interval can be determined based on the time prediction model, and then the target time interval can be added to the target change start time to determine the predicted arrival time of the target part change task.
[0045] In a possible implementation, the time prediction model is a distribution function, which is obtained based on the prediction sample fitting. The specific implementation process can be as follows:
[0046] Acquire multiple prediction samples, wherein the prediction samples include the prediction sample change start time and the prediction sample part arrival time;
[0047] Determine the time difference between the prediction sample change start time of the first prediction sample and the prediction sample part arrival time, and use the time difference as the first sample change time difference corresponding to the first prediction sample, wherein the first prediction sample is any one of the multiple prediction samples;
[0048] Fit the distribution function according to the sample change time difference corresponding to each predicted sample.
[0049] Among them, the distribution function is an important function in probability statistics. Through the distribution function, the random variable can be studied by mathematical analysis. The distribution function is the most important probability feature of the random variable. The distribution function can fully describe the statistical law of the random variable and determine all other probability features of the random variable. The distribution function can be a Poisson distribution, a binomial distribution, a normal distribution, etc. The distribution function obtained by fitting in the embodiment of this specification can be used to represent the probability of the time interval between the change opening time and the time when the parts arrive at the warehouse.
[0050] It should be noted that it can be assumed that the sample change time difference corresponding to each predicted sample follows the corresponding distribution, and the sample change time difference is used as the only variable to fit the distribution function. Assuming that the sample change time difference corresponding to each predicted sample follows the Poisson distribution, in specific implementation, the variance of the sample change time difference corresponding to each predicted sample can be determined, and the variance is used as the parameter λ of the Poisson distribution. The sample change time difference is used as the only variable to fit the Poisson distribution. The sample change time difference is said to follow the Poisson distribution with parameter λ. The mean and variance of the Poisson distribution are equal, both of which are parameters λ. The variance is used as an explanation in the embodiments of this specification. After fitting the distribution function, the optimal distribution function of the fit can be determined to determine the target time interval, and then the predicted target time interval is added to the target change start time to obtain the predicted arrival time of the target part change task.
[0051] In another possible implementation, the time prediction model may also be a deep learning model or a machine learning model, which may be obtained based on prediction sample training, and the prediction sample carries an interval label, which is the time interval between the change start time of the prediction sample and the time when the part arrives at the warehouse. After the training is completed, the target change start time may be input into the trained deep learning model or machine learning model to obtain the target time interval corresponding to the target part change task, and then the predicted target time interval is added to the target change start time to obtain the predicted arrival time corresponding to the target part change task.
[0052] For example, assuming that the target change start time is June 1, and the target time interval determined based on the time prediction model is 25 days, then the predicted arrival time corresponding to the time-stamped part change task is June 26.
[0053] In an embodiment of the present specification, the target time interval between the change start time and the part arrival time can be determined based on the time prediction model. The target time interval refers to the time interval with the highest probability of occurrence, which facilitates the subsequent prediction of the target part change task's arrival time based on the target time interval, automatically realizing the prediction of the arrival time, and improving the prediction efficiency and accuracy.
[0054] Step 104: Input the target task data into the time correction model to obtain the target time residual corresponding to the target part change task, wherein the time correction model is trained based on the sample task data and residual label of the sample part change task, and the residual label is the difference between the sample actual arrival time and the sample predicted arrival time of the sample part change task.
[0055] Specifically, the time correction model refers to a pre-trained model for outputting the time difference between the actual arrival time and the predicted arrival time. That is, the time correction model is a model for determining the residual between the actual value and the predicted value. The time correction model can be a deep learning model, a machine learning model, or other models that can output time residuals.
[0056] In practical applications, the difference between the actual arrival time of the sample part change task and the predicted arrival time of the sample can be used as the residual label. The time correction model is trained based on the sample task data and the residual label of the sample part change task. Then, the acquired target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task output by the time correction model. The target time residual can indicate the time difference between the predicted arrival time and the actual arrival time of the target part change task, thereby facilitating the subsequent correction of the predicted arrival time based on the target time residual to obtain a more accurate target arrival time, so that the final predicted part arrival time is closer to the actual situation and more accurate.
[0057] In an optional implementation of this embodiment, the time correction model is obtained by training using the following method:
[0058] Obtain sample task data, where the sample task data carries a residual label;
[0059] Input the sample task data into the initial time correction model to obtain the predicted time residual corresponding to the sample task data;
[0060] According to the predicted time residual and the residual label of the sample task data, the loss value of the initial time correction model is calculated, and the model parameters of the initial time correction model are reversely adjusted according to the loss value. Return to execute the step of obtaining the sample task data until the training stop condition is reached to obtain the trained time correction model.
[0061] It should be noted that the residual label may refer to the difference between the actual arrival time of the sample in the sample task data and the predicted arrival time of the sample. The residual label may be manually labeled or automatically calculated based on the sample task data. This specification does not impose any limitation on this.
[0062] In one possible implementation, it is possible to determine whether the training stop condition is met based only on the relationship between the loss value and the loss threshold. Specifically, if the loss value is greater than or equal to the loss value threshold, it means that the difference between the predicted time residual corresponding to the sample task data and the residual label of the sample task data is large, and the analysis and processing capability of the initial time correction model is poor. At this time, it can be determined that the training stop condition is not met, and the model parameters of the initial time correction model are adjusted in reverse, and the operation steps of obtaining the sample task data are returned to continue training the initial time correction model until the loss value is less than the loss value threshold, indicating that the difference between the predicted time residual corresponding to the sample task data and the residual label of the sample task data is small, and it is determined that the training stop condition is met, and the training is stopped to obtain the trained time correction model.
[0063] It should be noted that the loss value can be calculated based on the predicted time residual and the residual label of the sample task data. The residual label of the sample task data refers to the result that the initial time correction model actually wants to output, that is, the residual label of the sample task data is the true result, and the sample task data is input into the initial time correction model. The predicted time residual corresponding to the output sample task data is the predicted result of the initial time correction model. When the difference between the predicted result and the true result is small enough, it means that the predicted result is close enough to the true result. At this time, the initial time correction model training is completed, and the trained time correction model is obtained.
[0064] The loss value threshold is the critical value of the loss value. When the loss value is greater than or equal to the loss value threshold, it means that there is still a certain deviation between the prediction result of the initial time correction model and the actual result, and the model parameters of the initial time correction model still need to be adjusted. At this time, it is determined that the training stop condition has not been met; when the loss value is less than the loss value threshold, it means that the prediction result of the initial time correction model is close enough to the actual result, and the training can be stopped. At this time, it can be determined that the training stop condition has been met. The value of the loss value threshold is selected according to the actual situation, and the embodiments of this specification do not impose any restrictions on this.
[0065] In the embodiments of the present specification, the difference between the predicted result and the actual result of the initial time correction model can be intuitively shown by calculating the loss value. Subsequently, the initial time correction model can be trained in a targeted manner based on the difference, and the parameters of the initial time correction model can be adjusted, thereby effectively improving the training rate and training effect of the initial time correction model.
[0066] In another possible implementation, in addition to comparing the relationship between the loss value and the loss value threshold, the number of iterations can also be combined to determine whether the training stop condition is met. Specifically, if the loss value is greater than the loss value threshold, it can be further determined whether the number of iterations at this moment has reached the preset number of iterations. If the number of iterations at this moment has not reached the preset number of iterations, it can be determined that the training stop condition has not been met, and the model parameters of the initial time correction model can be adjusted, and the step of obtaining sample task data can be returned to continue training the initial time correction model until the preset number of iterations is reached. It is determined that the training stop condition is met, and the iteration is stopped to obtain the trained time correction model.
[0067] Among them, the preset number of iterations is set according to the actual situation, and the embodiments of this specification do not impose any restrictions on this. When the number of training times reaches the preset number of iterations, it means that the number of training times of the initial time correction model is sufficient. At this time, the prediction result of the initial time correction model is already very close to the actual result, and the training can be stopped.
[0068] In practical applications, there are many loss functions for calculating loss values, such as cross entropy loss function, L1 norm loss function, maximum loss function, mean square error loss function, logarithmic loss function, etc. The loss function can be used to evaluate the degree of difference between the predicted results of the model and the actual results. The better the loss function, the better the performance of the model.
[0069] In the embodiments of the present specification, the specific training situation of the initial time correction model can be judged according to the loss value, and if the training fails, the model parameters of the initial time correction model can be reversely adjusted according to the loss value to improve the analysis ability of the initial time correction model, with high training rate and good training effect.
[0070] In an optional implementation of this embodiment, a training sample set may be pre-constructed to obtain sample task data therefrom, that is, before obtaining the sample task data, the following steps may be further included:
[0071] Acquire sample task data of at least one sample part change task, wherein the sample part change task is a part change task that has been completed before the current time;
[0072] Obtain the sample actual arrival time and the sample predicted arrival time corresponding to the first sample task data, wherein the first sample task data corresponds to the sample task data of the first sample part change task, and the first sample part change task is any sample part change task;
[0073] Determine the time difference between the actual arrival time of the sample and the predicted arrival time of the sample, and use the time difference as the residual label corresponding to the first sample task data;
[0074] According to each sample task data and the corresponding residual label, a training sample set is constructed;
[0075] Accordingly, sample task data is obtained, including:
[0076] Get sample task data from the training sample set.
[0077] Specifically, the sample part change task refers to the part change task that was completed before the current time, that is, the historical part change task. For the sample part change task, the sample change start time and the sample real arrival time can be accurately known. That is, in the training stage, in addition to the part model, associated factory, associated vehicle model, process owner, and sample change start time, the sample task data can also include the sample real arrival time.
[0078] In actual implementation, after obtaining the sample task data of at least one sample part change task, the sample real arrival time and the sample predicted arrival time corresponding to the first sample task data can be obtained, and the time difference between the sample real arrival time and the sample predicted arrival time can be used as the residual label corresponding to the first sample task data. Similarly, the sample task data of each sample part change task can be used as the first sample task data, and the corresponding residual label is determined. Each sample task data and the corresponding residual label are obtained to construct a training sample set, which includes the multiple sample task data, and each sample task data carries a corresponding residual label. Then, the sample task data can be randomly obtained from the training sample set and input into the initial time correction model to train the initial time correction model.
[0079] It should be noted that in order to determine the time difference between the actual arrival time of the sample and the predicted arrival time of the sample, so as to obtain the residual label corresponding to the sample task data, it is also necessary to obtain the predicted arrival time of the sample corresponding to the sample task data. The predicted arrival time of the sample can be directly carried in the sample task data or obtained through a trained time prediction model.
[0080] In an embodiment of the present specification, the time difference between the actual arrival time of the sample and the predicted arrival time of the sample corresponding to the sample task data can be used as a residual label to construct a training sample set, so as to facilitate the training of the initial time correction model, so that the trained time correction model can predict the target time residual corresponding to the target part change task based on the input target task data, so as to correct the predicted arrival time of the target part change task and obtain a more accurate target arrival time, so that the final predicted part arrival time is closer to the actual situation and more accurate.
[0081] In an optional implementation of this embodiment, the first sample task data includes a first change start time of the first sample part change task; obtaining the sample predicted arrival time corresponding to the first sample task data includes:
[0082] Determine a sample time interval according to a time prediction model, wherein the sample time interval is used to indicate the interval between a sample change start time and a sample part arrival time in the warehouse;
[0083] According to the sample time interval and the first change start time, the sample predicted arrival time corresponding to the first sample part change task is determined.
[0084] In actual applications, the first sample task data may not include the sample predicted arrival time in the warehouse, but only include the first change start time of the sample part change task. At this time, the trained time prediction model can be used to determine the corresponding sample time interval, and then the sample time interval is increased based on the first change start time to obtain the sample predicted arrival time corresponding to the first sample part change task.
[0085] It should be noted that the specific implementation process of determining the predicted arrival time of samples based on the time prediction model in the training phase is similar to the specific implementation process of determining the predicted arrival time corresponding to the target part change task based on the time prediction model in the above-mentioned application phase, and the embodiments of this specification will not be repeated here.
[0086] In an optional implementation of this embodiment, the trained time correction model may also be tested based on the test task data, that is, after obtaining the trained time correction model, the following step may be further included:
[0087] Acquire test task data, wherein the test task data carries a test residual;
[0088] The trained time correction model is tested according to the test task data, and the trained time correction model is obtained if the test passes.
[0089] It should be noted that historical part change tasks can be obtained, and a set proportion of the historical part change tasks can be used as sample part change tasks, and the task data of the sample part change tasks can be used as sample task data to train the time correction model; then the remaining tasks in the historical part change tasks can be used as test tasks, and the task data of the test tasks can be used as test task data to test the trained time correction model to evaluate the effect of the trained time correction model. For example, 80% of the historical part change tasks can be used as sample part change tasks, and the remaining 20% can be used as test tasks.
[0090] In practical applications, the test task data can carry test residuals, which can be calculated based on the actual arrival time and predicted arrival time corresponding to the test task data. During specific testing, the test task data can be input into the trained time correction model, and the time correction model can output the prediction residual, which is the predicted value; the test residual carried by the test task data is the true value, and the test error between the test residual and the prediction residual can be calculated at this time, and the average value of the test error corresponding to each test task data can be determined, and whether the average value meets the regression evaluation index MAE (mean absolute error). The regression evaluation index MAE can be set in advance, such as setting MAE to 3.9 days. If the calculated average value of the test error is greater than the regression evaluation index MAE, it means that the prediction accuracy of the trained time correction model is poor, and the time correction model needs to be further trained, and the test fails; if the calculated average value of the test error is not greater than the regression evaluation index MAE, it means that the prediction accuracy of the trained time correction model is good, and there is no need to continue training the time correction model. The test passes, and it can be determined that the final trained time correction model is obtained.
[0091] In an embodiment of the present specification, other tasks in the historical part change tasks except the sample part change tasks can be used as test tasks. Based on the test task data of the test tasks that have been executed, the time correction model after preliminary training is tested to verify the prediction accuracy of the time correction model, thereby testing the effect of the time correction model. If the test fails, the time correction model can continue to be trained. If the test passes, the final trained time correction model is obtained, which improves the accuracy of the time correction model in determining the time residual, ensures the accuracy of the subsequent correction of the predicted arrival time, and further ensures the accuracy of the target arrival time obtained by the final prediction.
[0092] In an optional implementation of this embodiment, the target task data includes at least one data item, and the time correction model is a machine learning model. In this case, the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task, including:
[0093] Encode each data item of the target task data to obtain an encoding vector of the target task data;
[0094] The encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
[0095] It should be noted that the target task data includes at least one data item, which can be an associated data item in multiple different stages of the production part change task, such as part model, associated factory, associated vehicle model, process owner, change start time, planned time, etc. The dimensions of different data items can be the same or different. For the machine learning model, after receiving the input data, it will splice the various data items of the input data and divide them into multiple data segments for analysis and identification. In this way, the data dimensions of the data items will not affect the results of the recognition analysis, and there is no need to unify the dimensions.
[0096] Therefore, in practical applications, if the time correction model is a machine learning model, each data item of the target task data can be directly encoded to obtain the encoding vector of the target task data, which can be encoded using one-hot encoding or other encoding methods. Then, the encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
[0097] In the embodiment of the present specification, if the time correction model is a machine learning model, each data item of the target task data is directly encoded and input into the time correction model without the need for redundant normalization processing, thereby saving processing resources and ensuring the accuracy of the model's analysis results.
[0098] In an optional implementation of this embodiment, the target task data includes at least one data item, the data dimension of the at least one data item is different, and the time correction model is a deep learning model;
[0099] Before inputting the target task data into the time correction model to obtain the target time residual corresponding to the target part change task, it also includes:
[0100] determining a data dimension of at least one data item;
[0101] Normalize the data dimensions of each data item to obtain update task data;
[0102] Accordingly, the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task, including:
[0103] Encode each data item of the update task data to obtain an encoding vector of the target task data;
[0104] The encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
[0105] In practical applications, for deep learning models, different data items will be encoded separately. When the data items of the target task data are spliced, the dimension of the data item will affect the recognition of the model. The model will focus more on data items with large data dimensions, that is, it will focus more on data items with more information. Unbalanced information may affect the results of the model. For example, if the data dimension of data item A is 128 and the dimension of data item B is 64, then the recognition analysis of the model will focus more on data item A.
[0106] Therefore, in a specific implementation, if the time correction model is a deep learning model and the data dimension of at least one data item is different, then the data dimension of at least one data item can be determined, and the data dimensions of each data item are first normalized to obtain update task data, in which the dimensions of each data item are the same. Then, each data item of the update task data is encoded to obtain the encoding vector of the target task data, and the encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
[0107] It should be noted that for deep learning models, the dimensions of each data item of the target task data can be normalized first to ensure that the amount of information in the input data input into the time correction model is balanced, thereby ensuring the recognition and analysis results of the time correction model.
[0108] In the embodiments of this specification, for different types of time correction models, it is possible to determine whether the target task data needs to be normalized based on the data processing rules of the model of that type, which can be applicable to various application scenarios.
[0109] Step 106: Correct the predicted arrival time of the target part change task according to the target time residual to obtain the target arrival time of the target part change task.
[0110] Specifically, the predicted arrival time may be information carried in the target task data, or may be a time predicted based on a time prediction model, or may be an arrival time predicted by other means, and this specification does not limit this.
[0111] It should be noted that after the target time residual is determined based on the time correction model, the target time residual is used to indicate the difference between the actual arrival time and the predicted arrival time. The difference can be positive or negative. The sum of the target time residual and the predicted arrival time is the target arrival time of the target part change task.
[0112] For example, if the target time residual is 3, it means that the actual arrival time minus the predicted arrival time is 3 days, that is, the actual arrival time is 3 days later than the predicted arrival time. Assuming that the predicted arrival time is June 21, then June 21 plus 3 days, the target arrival time of the target part change task can be obtained as June 24; the target time residual is -3, which means that the actual arrival time minus the predicted arrival time is -3 days, that is, the actual arrival time is 3 days earlier than the predicted arrival time. Assuming that the predicted arrival time is June 21, then June 21 plus -3 days (that is, minus 3 days) can be obtained as June 18.
[0113] The predicted time correction method for the part change task provided in the embodiment of this specification realizes that the difference between the sample real arrival time and the sample predicted arrival time of the sample part change task is used as the residual label, and the time correction model is obtained based on the sample task data and the residual label of the sample part change task. When it is necessary to predict the target arrival time of the target part change task, the target task data of the target part change task is input into the pre-trained time correction model to obtain the target time residual corresponding to the target part change task, and then the predicted arrival time of the target part change task is corrected according to the target time residual to obtain the final predicted target arrival time of the target part change task. In this way, the initial predicted arrival time can be automatically corrected based on the trained time correction model to obtain the final predicted target arrival time, thereby ensuring the accuracy of the final predicted target arrival time, and further correcting the initial predicted arrival time using the time correction model, thereby reducing the accuracy requirement of the initial predicted arrival time, which is more suitable for the actual production situation.
[0114] Figure 2 FIG. 1 is a schematic diagram of a training process of a time correction model in a method for correcting the predicted time of a part change task provided in an embodiment of the present specification, such as Figure 2As shown, the sample task data of the sample part change task is input into the time prediction model, and the time prediction model can be a Poisson distribution obtained by fitting. The sample time interval is determined based on the Poisson distribution. The sample predicted arrival time of the sample task data is determined based on the sample time interval and the sample change start time. The residual label of the sample task data is determined based on the sample actual arrival time of the sample task data and the sample predicted arrival time of the sample task data. The sample task data of the sample part change task is input into the encoding layer to encode the sample task data to obtain the sample encoding feature, and the sample encoding feature is input into the initial time correction model. The initial time correction model can be a machine learning model or a deep learning model. The initial time correction model outputs the predicted time residual. Based on the predicted time residual and the residual label of the sample task data, the loss value of the initial time correction model is calculated. According to the loss value, the model parameters of the initial time correction model are reversely adjusted, and the step of obtaining the sample task data is returned until the training stop condition is reached to obtain the trained time correction model.
[0115] It should be noted that the difference between the actual arrival time of the sample of the sample part change task and the predicted arrival time of the sample is used as the residual label, and the time correction model is trained based on the sample task data and the residual label of the sample part change task, so that the target time residual corresponding to the target part change task to be tested can be determined through the trained time correction model, and then the predicted arrival time of the target part change task is corrected according to the target time residual, thereby ensuring the accuracy of the final predicted target arrival time.
[0116] The following combination Figure 3 Taking the application of the prediction time correction method for the part change task provided in this specification in the vehicle production scenario as an example, the prediction time correction method for the part change task is further explained. Figure 3 This is a processing flow chart of a method for correcting the predicted time of a part change task applied to a vehicle production scenario provided in an embodiment of this specification, which specifically includes the following steps:
[0117] Step 302: Obtain sample task data of at least one sample part change task in vehicle production, wherein the sample part change task is a vehicle part change task that has been completed before the current time.
[0118] Step 304: Obtain the sample actual arrival time and sample predicted arrival time corresponding to the first sample task data, wherein the first sample task data is sample task data corresponding to a first sample part change task, and the first sample part change task is any sample part change task.
[0119] Step 306: Determine the time difference between the actual arrival time of the sample and the predicted arrival time of the sample, and use the time difference as the residual label corresponding to the first sample task data; and construct a training sample set based on each sample task data and the corresponding residual label.
[0120] Step 308: Obtain sample task data from the training sample set, where the sample task data carries a residual label.
[0121] Step 310: Input the sample task data into the initial time correction model to obtain the predicted time residual corresponding to the sample task data; calculate the loss value of the initial time correction model based on the predicted time residual and the residual label of the sample task data, and reversely adjust the model parameters of the initial time correction model based on the loss value.
[0122] Return to execute the above step 308 until the training stop condition is reached to obtain the trained time correction model.
[0123] Step 312: Acquire test task data, wherein the test task data carries test residuals; test the trained time correction model according to the test task data, and obtain the trained time correction model if the test passes.
[0124] Step 314: Obtain target task data of the target part change task.
[0125] Step 316: When the target task data includes at least one data item, at least one data item has different data dimensions, and the time correction model is a deep learning model, determine the data dimension of at least one data item, normalize the data dimensions of each data item, and obtain the updated task data; encode each data item of the updated task data to obtain the encoding vector of the target task data, input the encoding vector of the target task data into the time correction model, and obtain the target time residual corresponding to the target part change task.
[0126] Step 318: When the target task data includes at least one data item and the time correction model is a machine learning model, encode each data item of the target task data to obtain an encoding vector of the target task data, input the encoding vector of the target task data into the time correction model, and obtain the target time residual corresponding to the target part change task.
[0127] Step 320: When the target task data includes the target change start time, determine the target time interval according to the time prediction model, and determine the predicted arrival time of the target part change task according to the target time interval and the target change start time.
[0128] Step 322: Correct the predicted arrival time of the target part change task according to the target time residual to obtain the target arrival time of the target part change task.
[0129] The predicted time correction method for the part change task provided in the embodiment of this specification realizes that the difference between the sample real arrival time and the sample predicted arrival time of the sample part change task is used as the residual label, and the time correction model is obtained based on the sample task data and the residual label of the sample part change task. When it is necessary to predict the target arrival time of the target part change task, the target task data of the target part change task is input into the pre-trained time correction model to obtain the target time residual corresponding to the target part change task, and then the predicted arrival time of the target part change task is corrected according to the target time residual to obtain the final predicted target arrival time of the target part change task. In this way, the initial predicted arrival time can be automatically corrected based on the trained time correction model to obtain the final predicted target arrival time, thereby ensuring the accuracy of the final predicted target arrival time, and further correcting the initial predicted arrival time using the time correction model, thereby reducing the accuracy requirement of the initial predicted arrival time, which is more suitable for the actual production situation.
[0130] Corresponding to the above method embodiment, this specification also provides an embodiment of a predicted time correction device for a part change task, Figure 4 1 is a schematic diagram of a predicted time correction device for a part change task provided in an embodiment of the present specification. Figure 4 As shown, the device comprises:
[0131] An acquisition module 402 is configured to acquire target task data of a target part change task;
[0132] An input module 404 is configured to input the target task data into a time correction model to obtain a target time residual corresponding to the target part change task, wherein the time correction model is trained based on sample task data and residual labels of the sample part change task, and the residual label is the difference between the sample actual arrival time and the sample predicted arrival time of the sample part change task;
[0133] The correction module 406 is configured to correct the predicted arrival time of the target part change task according to the target time residual to obtain the target arrival time of the target part change task.
[0134] Optionally, the device further includes a first determining module configured to:
[0135] Determine a target time interval according to a time prediction model, wherein the target time interval is used to indicate the interval between the change start time and the time when the parts arrive at the warehouse;
[0136] According to the target time interval and the target change start time, the predicted arrival time of the target part change task is determined.
[0137] Optionally, the device further comprises a training module configured to:
[0138] Obtain sample task data, where the sample task data carries a residual label;
[0139] Input the sample task data into the initial time correction model to obtain the predicted time residual corresponding to the sample task data;
[0140] According to the predicted time residual and the residual label of the sample task data, the loss value of the initial time correction model is calculated, and the model parameters of the initial time correction model are reversely adjusted according to the loss value. Return to execute the step of obtaining the sample task data until the training stop condition is reached to obtain the trained time correction model.
[0141] Optionally, the training module is further configured to:
[0142] Acquire sample task data of at least one sample part change task, wherein the sample part change task is a part change task that has been completed before the current time;
[0143] Obtain the sample actual arrival time and the sample predicted arrival time corresponding to the first sample task data, wherein the first sample task data corresponds to the sample task data of the first sample part change task, and the first sample part change task is any sample part change task;
[0144] Determine the time difference between the actual arrival time of the sample and the predicted arrival time of the sample, and use the time difference as the residual label corresponding to the first sample task data;
[0145] According to each sample task data and the corresponding residual label, a training sample set is constructed;
[0146] Get sample task data from the training sample set.
[0147] Optionally, the device further comprises a testing module configured to:
[0148] Acquire test task data, wherein the test task data carries a test residual;
[0149] The trained time correction model is tested according to the test task data, and the trained time correction model is obtained if the test passes.
[0150] Optionally, the target task data includes at least one data item, the data dimension of the at least one data item is different, and the time correction model is a deep learning model;
[0151] The apparatus further includes a normalization module configured to:
[0152] determining a data dimension of at least one data item;
[0153] Normalize the data dimensions of each data item to obtain update task data;
[0154] Accordingly, the input module 404 is further configured to:
[0155] Encode each data item of the update task data to obtain an encoding vector of the target task data;
[0156] The encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
[0157] Optionally, the target task data includes at least one data item, and the time correction model is a machine learning model;
[0158] The input module 404 is further configured to:
[0159] Encode each data item of the target task data to obtain an encoding vector of the target task data;
[0160] The encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
[0161] Optionally, the target task data includes a target change start time; the device further includes a second determination module configured to:
[0162] Determine a target time interval according to a time prediction model, wherein the target time interval is used to indicate the interval between the change start time and the time when the parts arrive at the warehouse;
[0163] According to the target time interval and the target change start time, the predicted arrival time of the target part change task is determined.
[0164] The predicted time correction device for the part change task provided in the embodiment of the present specification realizes that the difference between the sample real arrival time and the sample predicted arrival time of the sample part change task is used as the residual label, and the time correction model is obtained based on the sample task data and the residual label of the sample part change task. When it is necessary to predict the target arrival time of the target part change task, the target task data of the target part change task is input into the pre-trained time correction model to obtain the target time residual corresponding to the target part change task, and then the predicted arrival time of the target part change task is corrected according to the target time residual to obtain the final predicted target arrival time of the target part change task. In this way, the initial predicted arrival time can be automatically corrected based on the trained time correction model to obtain the final predicted target arrival time, thereby ensuring the accuracy of the final predicted target arrival time, and further correcting the initial predicted arrival time using the time correction model, thereby reducing the accuracy requirement of the initial predicted arrival time, which is more suitable for the actual production situation.
[0165] The above is a schematic scheme of a prediction time correction device for a part change task of this embodiment. It should be noted that the technical scheme of the prediction time correction device for a part change task and the technical scheme of the prediction time correction method for the above-mentioned part change task belong to the same concept, and the details not described in detail in the technical scheme of the prediction time correction device for a part change task can be referred to the description of the technical scheme of the prediction time correction method for the above-mentioned part change task.
[0166] Figure 5 The structure block diagram of a computing device provided according to an embodiment of the present specification is shown. The components of the computing device 500 include but are not limited to a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and the database 550 is used to store data.
[0167] The computing device 500 also includes an access device 540 that enables the computing device 500 to communicate via one or more networks 560. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.
[0168] In one embodiment of the present specification, the above components of the computing device 500 and Figure 5 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 5 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.
[0169] The computing device 500 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or PC. The computing device 500 may also be a mobile or stationary server.
[0170] The processor 520 is used to execute the following computer executable instructions to implement the steps of the above-mentioned method for correcting the predicted time of the part change task.
[0171] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the predicted time correction method of the above-mentioned part change task belong to the same concept, and the details not described in detail in the technical scheme of the computing device can be referred to the description of the technical scheme of the predicted time correction method of the above-mentioned part change task.
[0172] An embodiment of the present specification also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, are used to implement the steps of the above-mentioned method for correcting the predicted time of a part change task.
[0173] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the predicted time correction method of the above-mentioned part change task belong to the same concept, and the details not described in detail in the technical scheme of the storage medium can be referred to the description of the technical scheme of the predicted time correction method of the above-mentioned part change task.
[0174] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0175] Computer instructions include computer program codes, which may be in source code form, object code form, executable files, or some intermediate form, etc. Computer readable media may include: any entity or device capable of carrying computer program codes, recording media, USB flash drives, mobile hard disks, magnetic disks, optical disks, computer memories, read-only memories (ROM), random access memories (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.
[0176] It should be noted that, for the convenience of description, the aforementioned method embodiments are all described as a series of action combinations, but those skilled in the art should be aware that this specification is not limited by the order of the actions described, because according to this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this specification.
[0177] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0178] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the above-mentioned specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is limited only by the claims and their full scope and equivalents.
Claims
1. A method for correcting the predicted time of a part change task, characterized in that: The method comprises: Obtain target task data of the target part change task; Input the target task data into the time correction model to obtain the target time residual corresponding to the target part change task, wherein the time correction model is trained based on the sample task data and residual label of the sample part change task, and the residual label is the difference between the sample real arrival time and the sample predicted arrival time of the sample part change task; The predicted arrival time of the target part change task is corrected according to the target time residual to obtain the target arrival time of the target part change task.
2. The method for correcting the predicted time of a part change task according to claim 1, characterized in that: The target task data includes the target change start time; After obtaining the target task data of the target part change task, the method further includes: Determine a target time interval according to a time prediction model, wherein the target time interval is used to indicate the interval between the change start time and the time when the parts arrive at the warehouse; The predicted arrival time of the target part change task is determined according to the target time interval and the target change start time.
3. The method for correcting the predicted time of a part change task according to claim 1, characterized in that: The time correction model is obtained by training using the following method: Acquire sample task data, wherein the sample task data carries a residual label; Inputting the sample task data into an initial time correction model to obtain a predicted time residual corresponding to the sample task data; According to the predicted time residual and the residual label of the sample task data, the loss value of the initial time correction model is calculated, the model parameters of the initial time correction model are reversely adjusted according to the loss value, and the step of obtaining the sample task data is returned to execute until the training stop condition is reached to obtain the trained time correction model.
4. The method for correcting the predicted time of a part change task according to claim 3, characterized in that: Before obtaining the sample task data, the method further includes: Acquire sample task data of at least one sample part change task, wherein the sample part change task is a part change task that has been completed before the current time; Obtain the sample actual arrival time and the sample predicted arrival time corresponding to the first sample task data, wherein the first sample task data is sample task data corresponding to a first sample part change task, and the first sample part change task is any sample part change task; Determine the time difference between the actual arrival time of the sample and the predicted arrival time of the sample, and use the time difference as the residual label corresponding to the first sample task data; According to each sample task data and the corresponding residual label, a training sample set is constructed; Accordingly, the obtaining of sample task data includes: The sample task data is obtained from the training sample set.
5. The method for correcting the predicted time of a part change task according to claim 3, characterized in that: After obtaining the trained time correction model, the method further includes: Acquire test task data, wherein the test task data carries a test residual; The trained time correction model is tested according to the test task data, and the trained time correction model is obtained if the test passes.
6. The method for correcting the predicted time of a part change task according to any one of claims 1 to 5, characterized in that: The target task data includes at least one data item, the data dimensions of the at least one data item are different, and the time correction model is a deep learning model; Before inputting the target task data into the time correction model to obtain the target time residual corresponding to the target part change task, the method further includes: determining a data dimension of the at least one data item; Normalize the data dimensions of each data item to obtain update task data; Accordingly, the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task, including: Encoding each data item of the update task data to obtain an encoding vector of the target task data; The encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
7. The method for correcting the predicted time of a part change task according to any one of claims 1 to 5, characterized in that: The target task data includes at least one data item, and the time correction model is a machine learning model; The step of inputting the target task data into a time correction model to obtain a target time residual corresponding to the target part change task includes: Encoding each data item of the target task data to obtain an encoding vector of the target task data; The encoding vector of the target task data is input into the time correction model to obtain the target time residual corresponding to the target part change task.
8. The method for correcting the predicted time of a part change task according to any one of claims 1 to 5, characterized in that: The target task data includes the target change start time; After obtaining the target task data of the target part change task, the method further includes: Determine a target time interval according to a time prediction model, wherein the target time interval is used to indicate the interval between the change start time and the time when the parts arrive at the warehouse; The predicted arrival time of the target part change task is determined according to the target time interval and the target change start time.
9. A predicted time correction device for a part change task, characterized in that: The device comprises: An acquisition module is configured to acquire target task data of a target part change task; An input module is configured to input the target task data into a time correction model to obtain a target time residual corresponding to the target part change task, wherein the time correction model is trained based on sample task data and residual labels of the sample part change task, and the residual label is the difference between the sample actual arrival time and the sample predicted arrival time of the sample part change task; The correction module is configured to correct the predicted arrival time of the target part change task according to the target time residual to obtain the target arrival time of the target part change task.
10. A computing device, characterized in that: include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the steps of the method for correcting the predicted time of a part change task as described in any one of claims 1-8.
11. A computer-readable storage medium, characterized in that: It stores computer instructions, which, when executed by a processor, implement the steps of the method for correcting the predicted time of a part change task as described in any one of claims 1-8.