Distributed quality inspection task dynamic scheduling method based on Internet and artificial intelligence

Abstract

The invention relates to the technical field of quality inspection tasks, and provides a distributed quality inspection task dynamic scheduling method based on Internet and artificial intelligence, and the method comprises the steps: S1, carrying out the preprocessing of core features through a normalization algorithm and a one-hot coding algorithm; s2, calculating a matching probability P1 and a local matching probability P2 of the quality inspector and the task, and calculating a final matching probability corresponding to the P1 and the P2; s3, setting a genetic algorithm; s4, allocating the quality inspection task to a quality inspector based on the final matching probability in the step S2 and the genetic algorithm in the step S3, and performing quality inspection processing on the quality inspection task by the quality inspector to generate a quality inspection task result; and S5, based on the real busy and idle index and the personnel ability portrait model, carrying out feedback adjustment on the distribution schemes in the step S2 and the step S3. According to the method, accurate matching in a structured data extraction quality detection task allocation scene is realized, and the system interpretability is enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of quality inspection tasks, and specifically to a method for dynamic scheduling of distributed quality inspection tasks based on the Internet and artificial intelligence. Background Art

[0002] In the task allocation scenario of structured data extraction quality inspection, the current mainstream methods mainly include static automatic allocation based on personnel idle status / level and autonomous claiming of task pools. However, both methods have significant drawbacks: 1. Common defects: Both methods lack a closed-loop feedback mechanism and cannot dynamically optimize the allocation strategy based on historical task execution quality data, resulting in repeated similar errors; 2. Limitations of static allocation: The automatic allocation model that relies on fixed rules is prone to capacity mismatch and low resource utilization because it does not quantitatively evaluate the matching relationship between personnel skills and task complexity; 3. Insufficient claims model: The autonomous claiming mechanism is prone to uneven task distribution (such as the retention of complex tasks), and requires additional manpower to verify the integrity of tasks, resulting in increased management costs. Summary of the invention

[0003] In order to help solve the above technical problems, this application provides a distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence, which adopts the following technical solutions: A distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence, wherein the distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence includes: Step S1: The input features are screened by the LASSO regression algorithm to obtain the core features, and the core features are preprocessed by the normalization algorithm and the one-hot encoding algorithm; Step S2: Calculate the probability P1 of the quality inspector matching with the task and the local matching probability P2 based on the core features in step S1 through the logistic regression model and the KNN model, and calculate the final matching probabilities corresponding to P1 and P2 through the SHAP algorithm; Step S3: Setting a genetic algorithm, the genetic algorithm includes an objective function, the core features include skill matching, urgency and load pressure, and the objective function is Maximize∑(first objective function weight • skill matching + second objective function weight • urgency − third objective function weight • load pressure); Step S4: Based on the final matching probability of step S2 and the genetic algorithm of step S3, the quality inspection task is assigned to the quality inspector, and the quality inspector performs quality inspection on the quality inspection task and generates the quality inspection task result; Step S5: Feedback adjustment is performed on the allocation schemes of steps S2 and S3 based on the real busy-idle index and the personnel capability portrait model, including: The real busy-idle index is calculated based on the terminal behavior analysis of the quality inspector. The terminal behavior analysis includes active time and task switching frequency. The real busy-idle index = the first real busy-idle index weight × (active time / total working time) + the second real busy-idle index weight × (1-task switching frequency / 10). The real busy-idle index is used to indicate the workload. Set up a personnel capability profile model, including introducing comprehensive skill labels, historical task quality scores, and real-time monitoring response speed and matching them with core features. Comprehensive skill labels, historical task quality scores, and real-time monitoring response speed are automatically updated through quality inspection task results. Based on the real busy-idle index, the personnel capability portrait model, the final matching probability of step S2 and the genetic algorithm of step S3, the quality inspection task is assigned to the quality inspector. The quality inspector processes the quality inspection task, generates the quality inspection task result, and performs acceptance on the quality inspection task result to obtain the task acceptance result.

[0004] Preferably, the distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence further includes step S6: based on the task acceptance result, dynamically adjusting the objective function weight parameter of step S3 through the RL reinforcement learning algorithm, RL reinforcement learning algorithm includes state design, action design and reward design. State design includes: The current task queue, including the list of quality inspection tasks to be assigned and the core features corresponding to each quality inspection task; Personnel load, including the current workload of each quality inspector. The current workload includes the number of assigned quality inspection tasks and the estimated completion time. Skill matching distribution, which indicates the matching degree between the quality inspectors and the quality inspection tasks. Action design involves adjusting the objective function weights, The reward design includes calculating the reward based on the task acceptance results. The task acceptance results include the task completion time score and the error rate. The reward = task completion time score × the first reward weight + (1-error rate) × the second reward weight.

[0005] Preferably, the distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence also includes step S7: based on the task acceptance result, if the error rate > the preset error rate threshold, the score of the comprehensive skill label of the personnel capability portrait model in step S5 corresponding to the quality inspector is reduced.

[0006] Preferably, the distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence also includes step S8: setting a task dashboard, which is used to display the task progress and personnel load heat map in real time, and generate a reason explanation for the allocation of each task. The task dashboard includes: Global view, used to display the load of each workstation through real-time heat map, The details panel is designed to click on the quality inspection task to view the assignment reason.

[0007] Preferably, the step S1 comprises: Task characteristics, personnel characteristics and real-time status are taken as input features. LASSO regression is used to screen the input features to obtain core features. The core features include continuous features and categorical features. The normalization algorithm is used to convert continuous features of different dimensions to the same dimension, and the one-hot encoding algorithm is used to convert categorical features into numerical features.

[0008] Preferably, step S2 comprises: The preprocessed core features are input into the logistic regression model, and the probability P1 of the quality inspector matching the task is output. The KNN model is used to find n historical quality inspection tasks that match the current core features based on the historical core features, and the local matching probability P2 of the quality inspector in the n historical quality inspection tasks is calculated. The probability P1 of the quality inspector matching the task and the local matching probability P2 of the quality inspector are used as input, and the contribution of P1 and P2 to the final prediction result are calculated through the SHAP algorithm, and weights are assigned according to the contribution. The final matching probabilities are generated according to the weights, and n is a positive integer.

[0009] Preferably, step S3 comprises: Genetic coding design: Each chromosome represents a distribution scheme, Genetic manipulation design: The crossover operation includes selecting two parent solutions and exchanging some of the task allocation segments of the two parent solutions. Mutation operations include randomly replacing the inspectors of a certain quality inspection task. The constraints include requiring the same quality inspector to handle the same number of tasks at the same time ≤ the threshold of the number of tasks to be handled at the same time, and tasks with an urgency greater than the threshold of the urgency are allocated within a preset time.

[0010] Preferably, the step S5 includes: the weight of the first real busy-idle index is 0.6, the weight of the second real busy-idle index is 0.4, and the total working time is 8 hours.

[0011] Preferably, step S7 includes: based on the task acceptance result, if task rework is triggered due to skill mismatch, triggering allocation strategy correction; if task rework is triggered due to detection process defects, notifying the process engineer to optimize the SOP.

[0012] Preferably, step S8 includes: when a quality inspector continuously handles a urgent tasks, he / she is automatically marked as having an overload risk; when a sudden batch quality inspection task comes in, the skill label that needs to be assigned first is highlighted, and a is a positive integer.

[0013] In summary, the method of the present invention achieves accurate matching in the task allocation scenario of structured data extraction quality inspection through a three-layer architecture of data-driven modeling, multi-objective dynamic optimization, and human-machine collaborative decision-making, while reducing mismatches and resource waste, and can flexibly cope with complex dynamic environments, enhancing system interpretability and user trust. This technical effect is universal in the fields of data quality inspection, data operation and maintenance management, and provides a standardized and scalable solution to the problem of multi-node task allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a schematic block diagram of an embodiment of a distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence of the present application. DETAILED DESCRIPTION

[0015] The present application is further described below in conjunction with the accompanying drawings, and the structure and principle of the present application are very clear to people in the field. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0016] Figure 1 This is a schematic block diagram of an embodiment of a distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence of the present application. Figure 1 "Task reception" corresponds to step S1, "algorithm" corresponds to step S2 and step S3, "task distribution", "task processing" and "task result" correspond to step S4, "personnel capability profile", "workload situation" and "result acceptance" correspond to step S5, the connection between "result acceptance" and "algorithm" corresponds to step S6, the connection between "result acceptance" and "personnel capability profile" corresponds to step S7, and the "manpower load heat map" and "task progress display" correspond to step S8.

[0017] Reinforcement Learning (RL) is a paradigm of machine learning. Its core idea is to enable the intelligent agent to gradually learn and optimize decision-making strategies in the interaction with the environment through trial and error and reward mechanism to achieve specific goals.

[0018] Genetic Algorithm (GA) is an optimization algorithm that simulates natural selection and genetic mechanisms. It searches for the optimal solution or approximate optimal solution in the solution space by simulating operations such as natural selection, crossover and mutation.

[0019] A distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence in this application includes: Step S1: The input features are screened by the LASSO regression algorithm to obtain the core features, and the core features are preprocessed by the normalization algorithm and the one-hot encoding algorithm. In step S1, the task features, personnel features and real-time status are used as input features, and the input features are screened by LASSO regression to obtain the core features. The core features include continuous features and categorical features. The continuous features of different dimensions are converted to the same dimension by the normalization algorithm, and the categorical features are converted to numerical features by the one-hot encoding algorithm.

[0020] Step S2: The probability P1 of the quality inspector matching with the task and the local matching probability P2 are calculated based on the core features in step S1 through the logistic regression model and the KNN model, and the final matching probabilities corresponding to P1 and P2 are calculated through the SHAP algorithm. In step S2, the preprocessed core features are input into the logistic regression model, and the probability P1 of the quality inspector matching with the task is output. The n historical quality inspection tasks matching the current core features are found based on the historical core features through the KNN model, and the local matching probability P2 of the quality inspector in the n historical quality inspection tasks is calculated. The probability P1 of the quality inspector matching with the task and the local matching probability P2 of the quality inspector are used as inputs, and the contribution of P1 and P2 to the final prediction result is calculated through the SHAP algorithm, and weights are assigned according to the contribution, and the final matching probabilities are generated according to the weights, where n is a positive integer.

[0021] Step S3: Set up a genetic algorithm. The genetic algorithm includes an objective function. The core features include skill matching, urgency and load pressure. The objective function is Maximize∑(weight of the first objective function • skill matching + weight of the second objective function • urgency − weight of the third objective function • load pressure). In step S3, it also includes genetic coding design, genetic operation design and constraints. Genetic coding design: each chromosome represents an allocation scheme. The genetic operation design includes crossover operation and mutation operation. The crossover operation includes selecting two parent schemes and exchanging some task allocation segments of the two parent schemes. The mutation operation includes randomly replacing the quality inspector of a certain quality inspection task. The constraints include requiring the same quality inspector to handle the same number of tasks at the same time ≤ the threshold of the number of tasks to be handled at the same time, and tasks with an urgency greater than the threshold of the urgency are allocated within a preset time.

[0022] Step S4: Based on the final matching probability of step S2 and the genetic algorithm of step S3, the quality inspection task is assigned to the quality inspector, and the quality inspector performs quality inspection on the quality inspection task and generates a quality inspection task result.

[0023] Step S5: Feedback adjustment is performed on the allocation schemes of steps S2 and S3 based on the real busy-idle index and the personnel capability portrait model, including: The real busy-idle index is calculated based on the terminal behavior analysis of the quality inspectors. The terminal behavior analysis includes active time and task switching frequency. The real busy-idle index = the first real busy-idle index weight × (active time / total working time) + the second real busy-idle index weight × (1-task switching frequency / 10). The real busy-idle index is used to indicate the workload situation.

[0024] Set up a personnel capability portrait model, including introducing comprehensive skill labels, historical task quality scores, and real-time monitoring response speed and matching them with core features. Comprehensive skill labels, historical task quality scores, and real-time monitoring response speed are automatically updated through quality inspection task results.

[0025] Based on the real busy-idle index, the personnel capability portrait model, the final matching probability of step S2 and the genetic algorithm of step S3, the quality inspection task is assigned to the quality inspector, who processes the quality inspection task, generates the quality inspection task result, and performs the result acceptance on the quality inspection task result to obtain the task acceptance result. In step S5, the weight of the first real busy-idle index is 0.6, the weight of the second real busy-idle index is 0.4, and the total working time is 8 hours.

[0026] Step S6: Based on the task acceptance results, the objective function weight parameters of step S3 are dynamically adjusted through the RL reinforcement learning algorithm. The RL reinforcement learning algorithm includes state design, action design and reward design. The state design includes the current queue and personnel load. The current task queue includes the list of quality inspection tasks to be assigned and the core features corresponding to each quality inspection task; the personnel load includes the current workload of each quality inspector. The current workload includes the number of assigned quality inspection tasks and the estimated completion time. The skill matching distribution indicates the skill matching degree between the quality inspector and the quality inspection task. The action design includes adjusting the objective function weight. The reward design includes calculating the reward according to the task acceptance results. The task acceptance results include the task completion timeliness score and the error rate. The reward = task completion timeliness score × the first reward weight + (1-error rate) × the second reward weight.

[0027] Step S7: Based on the task acceptance result, if the error rate > the preset error rate threshold, the score of the comprehensive skill label of the personnel capability portrait model in step S5 corresponding to the quality inspector is reduced.

[0028] In step S7, based on the task acceptance result, if the task rework is triggered by skill mismatch, the allocation strategy correction is triggered; if the task rework is triggered by detection process defects, the process engineer is notified to optimize the SOP. SOP refers to Standard Operating Procedure, which is a detailed guidance document that specifies the standard steps, methods and requirements for specific activities within an organization.

[0029] Step S8: Set up a task dashboard, which is used to display the task progress and personnel load heat map in real time, and generate a reason for the assignment of each task. The task dashboard includes a global view and a detail panel design. The global view is used to display the load of each workstation through a real-time heat map, and the detail panel design is used to click on the quality inspection task to view the assignment reason.

[0030] In step S8, when a quality inspector handles a urgent tasks continuously, he / she is automatically marked as having an overload risk. When a sudden batch quality inspection task comes in, the skill label that needs to be assigned first is highlighted, and a is a positive integer.

[0031] Specifically, the scenario background: A structured data extraction quality inspection system needs to complete multi-link quality inspection of more than 2,000 tasks every day, and the existing pain points are as follows: 1. Uneven distribution: Highly skilled quality inspectors are frequently assigned simple tasks, resulting in a backlog of complex tasks.

[0032] 2. Delay in urgent tasks: The quality inspection response to sudden urgent tasks is slow, affecting the delivery timeliness.

[0033] 3. Load imbalance: Some quality inspectors are fully employed, while others are idle for a long time due to skill mismatch.

[0034] 4. High error rate: Manual allocation relies on experience, and skill mismatch leads to a missed detection rate of up to 8%.

[0035] This application includes: 1. Data preprocessing is a key step in machine learning tasks, which involves cleaning, transforming and normalizing raw data to provide high-quality data for subsequent model training. In the above scheme, the data preprocessing in step 1 mainly includes feature selection using LASSO regression, normalization of continuous features, and unique-hot encoding of categorical features.

[0036] This application uses LASSO regression for feature selection: LASSO (Least Absolute Shrinkage and Selection Operator) regression is a linear regression method that adds an L1 regularization term to the loss function to make the coefficients of some unimportant features become 0, thereby achieving the purpose of feature selection.

[0037] In this solution, task features (such as complexity, urgency, business type), personnel features (such as skill tags, historical quality inspection accuracy, average response speed) and real-time status (such as the number of tasks assigned on the day, terminal operation logs) are used as input features. Through LASSO regression, the most critical features for task allocation can be screened out, such as task complexity, urgency, personnel skill matching, historical accuracy, etc. Through feature selection, the interference of noise features can be reduced, the complexity of the model can be reduced, and the training efficiency and generalization ability of the model can be improved.

[0038] This application normalizes continuous features: Normalization is to convert features of different dimensions to the same dimension so that they are comparable. For continuous features, such as task complexity and urgency, their value ranges may be different, and directly inputting them into the model may affect the performance of the model. The value range of continuous features is scaled to [0,1] using the linear transformation method. Normalization can make the model more stable and improve the convergence speed and accuracy of the model.

[0039] This application performs one-hot encoding on categorical features: Categorical features, such as business types (temporary announcements, prospectuses, interest-bearing bonds), are non-numeric features and cannot be directly input into the model. One-hot encoding is a method of converting categorical features into numerical features. For each categorical feature, create a binary vector equal to the number of its categories. For example, for the business type feature, you can create a 3-dimensional binary vector corresponding to temporary announcements, prospectuses, and interest-bearing bonds. If the business type of a task is a temporary announcement, the corresponding vector is represented as [1, 0, 0]. One-hot encoding allows categorical features to be correctly processed by the model, improving the accuracy of the model.

[0040] Step 1 compresses the feature dimension from 15 dimensions to 8 dimensions, reduces noise interference, and improves model training efficiency.

[0041] 2. Logistic regression is a widely used linear model that is suitable for linearly separable tasks. In this scenario, logistic regression is used to predict whether a quality inspector is suitable for handling a specific type of task.

[0042] The input features include preprocessed task features (such as complexity and urgency) and personnel features (such as skill matching and historical accuracy). The model output is the probability P1 of the quality inspector matching the task, which indicates the suitability of the quality inspector to handle the task. KNN (K-Nearest Neighbors) is a non-parametric model used to capture the local similarity between data points. In this scheme, KNN is used to find the past tasks that are most similar to the current task requirements to evaluate the local matching ability of the quality inspector. Select k=5, that is, find the five past tasks that are most similar to the current task requirements based on historical data. Calculate the local matching probability P2 of the quality inspector in these five tasks, which indicates the performance of the quality inspector when handling similar tasks. SHAP (SHapleyAdditive exPlanations) is a method for explaining the predictions of machine learning models. In this scheme, the SHAP explanation layer is used to fuse the prediction results of logistic regression and KNN to generate an interpretable weight distribution. The probability P1 of the quality inspector matching the task and the local matching probability P2 are used as input, and the contribution of each feature to the final prediction result is calculated by the SHAP method. Assign weights according to contribution, such as 60% for logistic regression and 40% for KNN, to generate the final matching probability.

[0043] The personnel capability portrait is used to comprehensively describe the capability characteristics of quality inspectors, including skill labels, historical task quality scores, response speed and other dimensions. This application sets comprehensive skill labels, such as "text parsing expert", "numerical verification expert", etc., which are automatically updated according to the historical performance of quality inspectors. Introduce historical task quality scores to reflect the performance of quality inspectors in past tasks, monitor real-time status characteristics such as response speed in real time, and update personnel capability portraits. The real busy-idle index is used to reflect the real-time working status of quality inspectors to avoid overload or idleness. This application calculates the busy-idle index through terminal behavior analysis, such as active time, task switching frequency, etc. Busy-idle index = 0.6× (active time / 8 hours) + 0.4× (1-task switching frequency / 10), the higher the value, the busier it is.

[0044] Step 2 increased the task matching accuracy from 70% to 92%, and reduced the rework rate caused by mismatch by 40%.

[0045] 3. Through genetic algorithms, better task allocation plans are gradually generated to maximize resource utilization, improve the response speed of emergency tasks, and ensure the rationality of task allocation.

[0046] Objective function: Maximize ∑(w1•skill matching + w2•urgency - w3•load pressure).

[0047] In this application, the objective function is concretized as Maximize ∑(0.5•skill matching + 0.3•urgency - 0.2•load pressure). Here, w1, w2, and w3 represent the weights of skill matching, urgency, and load pressure, respectively. By adjusting these weights, different aspects of the allocation scheme can be optimized. The higher the skill matching, the more the task matches the inspector's skills; the higher the urgency, the faster the task needs to be processed; the smaller the load pressure, the smaller the current workload of the inspector.

[0048] Encoding method: Each chromosome represents an allocation scheme, that is, a mapping relationship between a task and a quality inspector. For example, [Task 1 → Personnel A, Task 2 → Personnel C, ...] represents a possible allocation scheme.

[0049] Crossover: Select two parent solutions and exchange some of their task allocation segments.

[0050] Mutation: Randomly replace the assigned personnel for a task, giving priority to inspectors with high skill matching and low load.

[0051] Constraints: The number of tasks that the same quality inspector can handle at the same time is ≤ 3: to avoid overloading the quality inspector and ensure the rationality of task allocation. Tasks with an urgency greater than 3 must be assigned within 30 minutes: to ensure that urgent tasks can be handled in a timely manner.

[0052] Step 3 increased resource utilization from 65% to 85% and shortened the average response time for emergency tasks to 15 minutes.

[0053] 4. Use RL reinforcement learning method to dynamically adjust the allocation strategy weight parameters.

[0054] Status: Current task queue: includes the list of tasks to be assigned, and the characteristics of each task (such as complexity, urgency, business type, etc.). Personnel load: The current workload of each quality inspector, including the number of assigned tasks, estimated completion time, etc. Skill matching distribution: The skill matching degree between the quality inspector and the task, calculated based on the personnel capability profile and skill label.

[0055] Action: Adjust objective function weights: Dynamically adjust weight parameters (such as w1, w2, w3) in the objective function according to the current state to optimize the allocation strategy.

[0056] Reward: Calculate rewards based on the timeliness and error rate of task completion to encourage the model to make better allocation decisions. The reward function can be designed as: Reward = Timeliness score × 0.7 + (1 - Error rate) × 0.3, where the timeliness score and error rate are calculated based on the actual task acceptance results.

[0057] Closed-loop feedback mechanism: Quality feedback: After task acceptance, if the error rate exceeds the set threshold (such as 5%), the score of the quality inspector's corresponding skill label will be reduced to reflect his performance in the actual task. Abnormal backtracking: Automatically analyze tasks that are frequently reworked, identify the root cause and take corresponding measures. If rework is caused by skill mismatch, such as assigning non-"temporary announcement experts" to handle difficult temporary announcements, the allocation strategy will be modified to optimize the skill matching calculation. If rework is caused by defects in the inspection process, such as the lack of specific instruments or steps, the process engineer will be notified to optimize the standard operating procedure (SOP).

[0058] Step 4 shortens the system's adaptive adjustment cycle from 2 weeks to 2 days, and the accuracy of process defect identification reaches 90%.

[0059] 5. Task board design: Global view: Real-time heat map showing the load of each workstation (red for high load, green for idle).

[0060] Details panel: Click on a task to view the reason for the assignment (e.g. "Choose Zhang Gong: skill matching 95% + historical accuracy 98%").

[0061] Early warning system: When a quality inspector handles three urgent tasks in succession, he or she is automatically marked as "overload risk" and managers are advised to intervene.

[0062] When sudden batch tasks come in, the dashboard highlights the skill tags that need to be assigned first (such as "urgent need for temporary announcement experts").

[0063] Effect: Manager intervention efficiency increased by 50%, and employee satisfaction with distribution fairness increased from 60% to 88%.

[0064] In summary, the implementation of the technical solution of this application in a certain structured data extraction quality detection system has achieved: 1. Accurate dynamic matching: multi-model fusion + GA optimization to reduce skill mismatch and load imbalance.

[0065] 2. Agile response: RL dynamic parameter adjustment + visual dashboard to quickly respond to urgent orders and abnormal events.

[0066] 3. Continuous self-optimization: The closed-loop feedback mechanism promotes system iteration and gradually approaches Pareto optimality.

Claims

1. A distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence, characterized in that: The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence includes: Step S1: The input features are screened by the LASSO regression algorithm to obtain the core features, and the core features are preprocessed by the normalization algorithm and the one-hot encoding algorithm; Step S2: Calculate the probability P1 of the quality inspector matching with the task and the local matching probability P2 based on the core features in step S1 through the logistic regression model and the KNN model, and calculate the final matching probabilities corresponding to P1 and P2 through the SHAP algorithm; Step S3: Setting a genetic algorithm, the genetic algorithm includes an objective function, the core features include skill matching, urgency and load pressure, and the objective function is Maximize∑(first objective function weight • skill matching + second objective function weight • urgency − third objective function weight • load pressure); Step S4: Based on the final matching probability of step S2 and the genetic algorithm of step S3, the quality inspection task is assigned to the quality inspector, and the quality inspector performs quality inspection on the quality inspection task and generates the quality inspection task result; Step S5: Feedback adjustment is performed on the allocation schemes of steps S2 and S3 based on the real busy-idle index and the personnel capability portrait model, including: The real busy-idle index is calculated based on the terminal behavior analysis of the quality inspector. The terminal behavior analysis includes active time and task switching frequency. The real busy-idle index = the first real busy-idle index weight × (active time / total working time) + the second real busy-idle index weight × (1-task switching frequency / 10). The real busy-idle index is used to indicate the workload. Set up a personnel capability portrait model, including introducing comprehensive skill labels, historical task quality scores, and real-time monitoring response speed and matching them with core features. Comprehensive skill labels, historical task quality scores, and real-time monitoring response speed are automatically updated through quality inspection task results. Based on the real busy-idle index, the personnel capability portrait model, the final matching probability of step S2 and the genetic algorithm of step S3, the quality inspection task is assigned to the quality inspector. The quality inspector processes the quality inspection task, generates the quality inspection task result, and performs acceptance on the quality inspection task result to obtain the task acceptance result.

2. The method for dynamic scheduling of distributed quality inspection tasks based on the Internet and artificial intelligence according to claim 1 is characterized in that: The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence also includes step S6: based on the task acceptance result, dynamically adjusting the objective function weight parameter of step S3 through the RL reinforcement learning algorithm, RL reinforcement learning algorithm includes state design, action design and reward design. State design includes: The current task queue, including the list of quality inspection tasks to be assigned and the core features corresponding to each quality inspection task; Personnel load, including the current workload of each quality inspector. The current workload includes the number of assigned quality inspection tasks and the estimated completion time. Skill matching distribution, which indicates the matching degree between the quality inspectors and the quality inspection tasks. Action design involves adjusting the objective function weights, The reward design includes calculating the reward based on the task acceptance results. The task acceptance results include the task completion time score and the error rate. The reward = task completion time score × the first reward weight + (1-error rate) × the second reward weight.

3. The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence according to claim 2 is characterized in that: The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence also includes step S7: based on the task acceptance result, if the error rate is greater than the preset error rate threshold, the score of the comprehensive skill label of the personnel capability portrait model in step S5 corresponding to the quality inspector is reduced.

4. The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence according to claim 3 is characterized in that: The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence also includes step S8: setting a task board, which is used to display the task progress and personnel load heat map in real time, and generate a reason explanation for the allocation of each task. The task board includes: Global view, used to display the load of each workstation through real-time heat map, The details panel is designed to click on the quality inspection task to view the assignment reason.

5. The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence according to claim 1 is characterized in that: The step S1 comprises: Task characteristics, personnel characteristics and real-time status are taken as input features. LASSO regression is used to screen the input features to obtain core features. The core features include continuous features and categorical features. The normalization algorithm is used to convert continuous features of different dimensions to the same dimension, and the one-hot encoding algorithm is used to convert categorical features into numerical features.

6. The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence according to claim 1 is characterized in that: The step S2 comprises: The preprocessed core features are input into the logistic regression model, and the probability P1 of the quality inspector matching the task is output. The KNN model is used to find n historical quality inspection tasks that match the current core features based on the historical core features, and the local matching probability P2 of the quality inspector in the n historical quality inspection tasks is calculated. The probability P1 of the quality inspector matching the task and the local matching probability P2 of the quality inspector are used as input, and the contribution of P1 and P2 to the final prediction result are calculated through the SHAP algorithm, and weights are assigned according to the contribution. The final matching probabilities are generated according to the weights, and n is a positive integer.

7. The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence according to claim 1 is characterized in that: The step S3 comprises: Genetic coding design: Each chromosome represents a distribution scheme, Genetic manipulation design: The crossover operation includes selecting two parent solutions and exchanging some of the task allocation segments of the two parent solutions. Mutation operations include randomly replacing the inspectors of a certain quality inspection task. The constraints include requiring the same quality inspector to handle the same number of tasks at the same time ≤ the threshold of the number of tasks to be handled at the same time, and tasks with an urgency greater than the threshold of the urgency are allocated within a preset time.

8. The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence according to claim 1 is characterized in that: The step S5 includes: the weight of the first real busy-idle index is 0.6, the weight of the second real busy-idle index is 0.4, and the total working time is 8 hours.

9. The distributed quality inspection task dynamic scheduling method based on the Internet and artificial intelligence according to claim 3 is characterized in that: The step S7 includes: based on the task acceptance result, if the task rework is triggered due to skill mismatch, triggering the allocation strategy correction; if the task rework is triggered due to detection process defects, notifying the process engineer to optimize the SOP.

10. The method for dynamic scheduling of distributed quality inspection tasks based on the Internet and artificial intelligence according to claim 4 is characterized in that: The step S8 includes: when a quality inspector continuously handles a urgent tasks, he is automatically marked as having an overload risk; when a sudden batch quality inspection task comes in, the skill label that needs to be assigned first is highlighted, and a is a positive integer.

Images