A performance appraisal management method for a mine workshop

Abstract

The present invention relates to the technical field of performance appraisal management, and particularly to a performance appraisal management method for a mine workshop, including data collection and preprocessing, dynamic performance evaluation, task difficulty adjustment, incentive mechanism, performance feedback and optimization; optimizing factor weights and predicting employee performance based on a multi-factor weighted model combined with neural network and reinforcement learning, and calculating performance scores in real time; using the analytic hierarchy process and fuzzy comprehensive evaluation to quantify task difficulty, and adapting and optimizing in real time in combination with employee ability analysis; adjusting integral value and customized solutions by using market supply and demand and collaborative filtering algorithms, and optimizing reward standards according to cost-benefit and performance sensitivity analysis; generating reports by means of a visualization library and generating suggestions by using text mining, warning through prediction and causal analysis, and adjusting and feedback optimizing the management strategy according to the recommendation of reinforcement learning; the present invention can comprehensively, accurately and dynamically assess and manage the performance of the mine workshop, improve operation efficiency and employee enthusiasm, and promote the sustainable development of the enterprise.

Description

Technical Field

[0001] The present invention relates to the technical field of performance appraisal management, and particularly to a performance appraisal management method for a mine workshop. Background Technique

[0002] The performance appraisal management of a mine workshop is crucial for enterprise operation, but traditional methods have many drawbacks; in terms of data collection and processing, the data source is single, mostly relying on simple equipment parameter records, lacking the collection of multi-dimensional states of equipment and detailed information on employees' work, and the data processing is insufficient, with noise and missing values, and it is not effectively integrated and analyzed, unable to provide an accurate data basis for appraisal; in terms of performance evaluation, it focuses on single production indicators, ignoring key factors such as task diversity, complexity, and employees' teamwork and innovation abilities, with one-sided evaluation, affecting employees' enthusiasm and enterprise efficiency; when adjusting task difficulty, the task assignment lacks scientificity, does not consider the adaptation of tasks to employees' skills, and there is no real-time monitoring and dynamic adjustment mechanism, affecting employees' performance and production efficiency; the incentive mechanism is single and lacks personalization, not closely related to performance appraisal, with insufficient long-term incentives, unable to meet employees' diverse needs, and not conducive to the long-term development of the enterprise; in the performance feedback and optimization link, the feedback lacks analysis and suggestions, and the adjustment of management strategies has no data-driven basis, mostly being empirical responses, unable to achieve systematic continuous optimization; in short, traditional methods cannot meet the modern management needs of mining enterprises, and there is an urgent need for innovative performance appraisal management methods;

[0003] Therefore, a performance appraisal management method for a mine workshop is proposed to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a performance appraisal management method for a mine workshop to solve the problems raised in the above background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A performance appraisal management method for a mine workshop includes the following steps:

[0007] Step S1, data collection and preprocessing: Collect work data through sensor fusion, text analysis, and determination of the self-evaluation weight of employees, and store it after multiple imputation and robust standardization processing;

[0008] Step S2, dynamic performance evaluation: Based on a multi-factor weighted model, use the combination of neural network and reinforcement learning to optimize factor weights and predict employees' future performance, and calculate and output performance scores in real time accordingly;

[0009] Step S3, task difficulty adjustment: Use the analytic hierarchy process and fuzzy comprehensive evaluation to quantify task difficulty, combine with analysis of employees' abilities, and rely on linear programming and gradient descent algorithms to adapt and optimize task difficulty in real time;

[0010] Step S4, Incentive Mechanism: The integral value and the customized integral plan are adjusted respectively by using the market supply and demand and the collaborative filtering algorithm. The reward criteria are optimized according to the cost-benefit and performance sensitivity analysis to realize integral redemption and data feedback;

[0011] Step S5, Performance Feedback and Optimization: Generate a performance report by means of a visualization library and generate improvement suggestions by using text mining. Predict and causally analyze the performance changes to warn of performance changes, and use reinforcement learning to recommend a management strategy adjustment plan and feedback for optimization.

[0012] As an optimal solution, in data collection and preprocessing, data collection includes:

[0013] Install sensors on the key equipment in the mine workshop, use the Kalman filter algorithm to fuse the sensor data to obtain the basic data related to the equipment operation and production. At the same time, use the natural language processing NLP technology to analyze the text of the employees' work records, quantify the key information of the text and supplement it to the work data. Also, based on the linear regression algorithm, determine the weight of the employees' self-evaluation in the overall data collection according to the coincidence degree between the employees' past self-evaluation and the actual performance, obtain the employees' work data, and then integrate the employees' work data in a unified format and transmit it to the data preprocessing server by means of the big data message queue technology; among them, build a linear regression model , where, is the set of actual performance data, is the set of employees' self-evaluation data, is the intercept, is the regression coefficient, is the error term, estimated by the least squares method and to determine the weight of the employees' self-evaluation;

[0014] Data preprocessing includes:

[0015] On the data preprocessing server, use the Multiple Imputation by Chained Equations (MICE) method to fill in the missing values in the data. Establish multiple prediction models based on the data correlation relationship for multiple fillings to ensure data integrity. Use the RobustScaler method to standardize the data, and its calculation formula is , where, is the original data, is the standardized data, represents the data of the median, represents the data of the interquartile range. The preprocessed data is stored in the distributed database.

[0016] As an optimal solution, in dynamic performance evaluation, the construction and optimization process of the multi-factor weighted model includes:

[0017] Determine the task type, difficulty, employee's historical performance, team collaboration effect, and innovation achievements as the key factors for performance evaluation. Let the set of these factors be , and initially set the weights of each factor based on expert experience and historical data analysis , and represents the total number of key factors for performance evaluation, is the index variable used to traverse all key factors to construct the multi-factor weighted model; the performance score is calculated by the formula:

[0018] , where is the optimized weight, is the value of each factor corresponding to the current work data;

[0019] After that, use the neural network algorithm. Take the employee performance fluctuation and the changes in the key points of the workshop business as the input, and according to the conventional training and weight adjustment mechanism of the neural network, adjust the weights of each factor in the weighted model in real time through the backpropagation method; at the same time, use the XGBoost algorithm combined with reinforcement learning to construct a prediction model. Regard the employee's work tasks as "actions" and the performance score as "rewards", and based on the exploration learning mechanism of reinforcement learning, combined with the processing ability of XGBoost for historical performance data, optimize the prediction model to accurately predict the future performance of employees;

[0020] The implementation method of performance score calculation and output is: input the preprocessed employee work data into the optimized multi-factor weighted model, combine the results of the prediction model, calculate the employee performance score in real time according to the established calculation logic, and output it to the subsequent module through the API interface to provide data support for relevant operations.

[0021] As an optimal solution, in task difficulty adjustment, the analytic hierarchy process includes:

[0022] Construct a hierarchical structure model. Let the goal layer be the task difficulty , the criterion layer contains relevant factors, and construct a judgment matrix by pairwise comparison of the factors in the criterion layer , where represents the degree of importance of factor relative to , and use the scale method for assignment;

[0023] Solve the judgment matrix to obtain the maximum eigenvalue and its corresponding eigenvector , Calculate the consistency index , where is the order of the judgment matrix. To test the consistency, when is less than the average random consistency index , it is considered that the consistency passes. The normalized eigenvector is the weight vector of each factor in the criterion layer ;

[0024] The fuzzy comprehensive evaluation method includes:

[0025] Let the evaluation set be . For the sub-factors under each factor in the criterion layer, conduct fuzzy evaluations respectively to construct a fuzzy relation matrix , where represents the membership degree of factor for evaluation . Combining with the weight vector of the factors in the criterion layer, using the fuzzy comprehensive evaluation formula , obtain the fuzzy comprehensive evaluation result vector of the task difficulty, where represents the fuzzy composition operator. Then, through the defuzzification method, convert it into a specific task difficulty value. The defuzzification method is the centroid method, and the calculation formula is , where is the -th element in , is the quantization value corresponding to the -th evaluation in the evaluation set . is the final task difficulty value, which is stored in the task management database as a reference for task difficulty adjustment; is the number of evaluations in the evaluation set ;

[0026] Use the clustering analysis algorithm to divide employees into different ability level categories based on their past performance and skill mastery. By calculating the similarity between employees and applying the corresponding clustering algorithm for classification. For each category of employees, use the Bayesian network model to analyze the probability relationship between tasks of different difficulties and the positive work efficiency of employees. Learn the network structure and conditional probability based on historical data to provide a reference basis for reasonable task allocation;

[0027] The specific task difficulty real-time adjustment strategy is:

[0028] According to the employee performance scores and task completion status obtained from the performance evaluation module, combined with the task difficulty assessment model and the analysis results of employee ability adaptation, through the linear programming algorithm, with a suitable objective function and meeting the corresponding constraint conditions, calculate the task difficulty adjustment range suitable for the current stage of the employee, and accordingly adjust the task assignment;

[0029] During the process of employees executing tasks, combine the real-time emotional state of employees obtained from wearable devices and workshop video monitoring, and the work efficiency tracking curve depicted by time series analysis to judge the effect of task difficulty adjustment. If it does not meet the expectations, based on the gradient descent algorithm, quickly adjust the task difficulty according to its iterative update mechanism to ensure that employees maintain a positive working state under tasks of appropriate difficulty, and timely feedback and update relevant information.

[0030] As an optimal solution, in the incentive mechanism, the collaborative filtering algorithm includes:

[0031] Let the user set be , the item set be , and the user rating matrix for items be ( represents the rating of user on item . If the user does not rate, it is recorded as a missing value); based on the user rating matrix , calculate the similarity between users. The commonly used Pearson correlation coefficient method is used, and the calculation formula is , where and are different users, is the set of items that both user and have rated, and are the average ratings of users and respectively; represents the similarity between user and ; represents the rating of user on item ;

[0032] Find the most similar users to form the neighbor set , and predict the rating of the target user on the unrated item , the collaborative filtering algorithm is used to analyze the personalized characteristics of employees, referring to the past performance improvement of similar employees, and customized personalized integral acquisition rules and distribution plans for employees according to the above algorithm. After employees complete tasks, they obtain integral according to the plan based on their performance and store it in the integral account database;

[0033] A simulation algorithm based on the market supply and demand theory is adopted. According to the popularity of different rewards being concerned and exchanged by employees, the corresponding integral value is dynamically adjusted according to the basic principle of the algorithm;

[0034] The dynamic adjustment of the reward standard includes:

[0035] From the perspective of the enterprise's cost - benefit, the cost - benefit analysis model and the dynamic programming algorithm are used. According to the enterprise's business objectives and financial status, the reward standard is optimized in real - time according to the corresponding algorithm logic. Using statistical methods such as variance analysis and elasticity analysis, the sensitivity between the performance score and the change of the reward standard is analyzed, so as to finely adjust the reward standard corresponding to different performance intervals. Employees choose to exchange rewards on the reward exchange platform according to the integral and the standard, and the incentive mechanism module feeds back the exchange records to other relevant modules for subsequent optimization analysis;

[0036] By comprehensively considering the cost and benefit of rewards and analyzing the correlation between performance and rewards, the reward standard is set scientifically and reasonably to motivate employees to improve their performance, and at the same time, the synergy optimization between the incentive mechanism and the entire performance appraisal management system is realized.

[0037] As an optimal solution, in performance feedback and optimization, the adjustment and optimization of management strategies include:

[0038] The time - series prediction algorithm is used to predict the trends of team and individual performance data, and the causal analysis algorithm is combined to mine the potential reasons leading to performance changes. When the probability of performance decline or potential problems is predicted, the early - warning mechanism is triggered according to the set threshold, and the early - warning information is pushed to the workshop management personnel;

[0039] Based on the reinforcement learning algorithm, by simulating the performance changes after the implementation of different management strategies, the optimal management strategy adjustment plan is recommended for the management personnel according to the strategy recommendation mechanism of reinforcement learning. The management personnel make corresponding adjustments in the workshop management system accordingly, and the performance change data after adjustment is fed back to the system for further optimizing the parameters and models of each algorithm module, realizing the continuous iterative optimization of the entire performance appraisal management system and improving the overall operation efficiency of the mine workshop.

[0040] It can be seen from the technical solution provided by the present invention above that a performance appraisal management method for a mine workshop provided by the present invention has the following beneficial effects:

[0041] 1. Data collection and processing are more scientific and comprehensive:

[0042] Multi-source data fusion to ensure accuracy: By installing a variety of sensors on key equipment in the mine workshop, such as sensors for output, pressure, running duration, speed, etc., and using the Kalman filter algorithm to fuse the data, more accurate basic data related to equipment operation and production can be obtained. For example, the fusion of output and pressure sensor data on excavation equipment can accurately reflect the relationship between the actual working state of the equipment and production efficiency, avoiding errors of a single sensor and providing a reliable basis for subsequent assessment.

[0043] Text analysis to mine potential information: Using natural language processing (NLP) technology to analyze the text of employees' work records, and quantifying information such as task progress and quality feedback to supplement the work data, making the assessment data more rich and comprehensive. For example, the quality problem descriptions mined from employees' records can timely detect production hidden dangers and help with targeted improvements.

[0044] Optimizing data quality with the weight of employees' self-evaluation: Based on the linear regression algorithm to determine the weight of employees' self-evaluation, and adjusting it according to the coincidence degree between past self-evaluation and actual performance, which improves the effectiveness of self-evaluation data in the overall data collection. If an employee's self-evaluation highly matches the actual performance, a higher weight is given, making the overall data more able to truly reflect the employee's work situation.

[0045] 2. More accurate and dynamic performance evaluation:

[0046] Comprehensive consideration with a multi-factor weighted model: Constructing a weighted model by integrating multi-factors such as task type, difficulty, employees' historical performance, team collaboration effect, and innovation achievements, avoiding the one-sidedness of single-factor evaluation. For example, when evaluating employees' performance, both the task completion volume (related to task type and difficulty) and the contribution in team collaboration (team collaboration effect) are considered, making the evaluation more comprehensive and objective.

[0047] Neural network and reinforcement learning to optimize weights and predict the future: The neural network adjusts the factor weights in real time by combining employees' performance fluctuations and changes in the key points of workshop operations. Reinforcement learning uses work tasks as "actions" and performance scores as "rewards" to optimize the prediction model and accurately predict employees' future performance. For example, if the workshop recently focuses on product quality (changes in key operations), the neural network will correspondingly increase the weights of quality-related factors. Reinforcement learning simulates task scenarios, allowing employees to optimize performance strategies in different task selections, achieving accurate evaluation and prediction.

[0048] 3. Higher task difficulty adaptability:

[0049] Analytic Hierarchy Process and Fuzzy Comprehensive Evaluation for Quantifying Science: The Analytic Hierarchy Process is used to construct a task difficulty assessment model to determine the weights of criterion-level factors such as skill level, time urgency, and resource dependence. The Fuzzy Comprehensive Evaluation method combines the factor weights and the fuzzy relation matrix, and through defuzzification, an accurate task difficulty value is obtained, providing a scientific quantitative basis for task allocation. For example, for a complex, time-consuming, and resource-intensive task, this method can accurately assess its difficulty and reasonably allocate it to employees with matching capabilities.

[0050] Employee Ability Analysis for Achieving Personalized Adaptation: Cluster analysis classifies employees based on their past performance and skill mastery. The Bayesian network model analyzes the adaptation relationship between different categories of employees and task difficulty, realizing the personalized adaptation of tasks and employee abilities, and improving employee work efficiency and overall production efficiency. For example, employees are divided into high, medium, and low ability levels, and tasks of appropriate difficulty are assigned to different categories of employees. High-ability employees undertake high-difficulty tasks to give full play to their potential, while low-ability employees gradually improve from tasks of appropriate difficulty, avoiding inefficiencies caused by tasks being too difficult or too easy.

[0051] 4. The incentive mechanism is more flexible and effective:

[0052] Dynamic Adjustment of Integral Value to Meet Requirements: Based on the market supply and demand theory, the integral value is dynamically adjusted according to the attention and redemption popularity of different rewards, making the integral system more in line with employees' needs and improving the incentive effect. For example, the integral value of popular training opportunities increases with the increase in demand, guiding employees to work hard to obtain integral points, and at the same time reasonably allocating enterprise resources.

[0053] Collaborative Filtering for Customizing Personalized Solutions: The collaborative filtering algorithm is used to analyze employees' personalized characteristics, and the integral acquisition rules and distribution plans are customized with reference to the performance improvement of similar employees, making the incentive more targeted. For example, for employees with strong technical improvement needs, a plan is customized to obtain specific training reward points by completing technology-related tasks, meeting the personalized development needs of employees and stimulating their work enthusiasm.

[0054] Cost-Benefit and Performance Sensitivity Analysis for Optimizing Reward Standards: From the perspective of cost-benefit, relevant algorithms are used to optimize the reward standards in real time. Through performance sensitivity analysis, the rewards in different performance intervals are finely adjusted, and the reward standards are set scientifically and reasonably to achieve the coordinated optimization of incentives and enterprise benefits. For example, when it is found through analysis that performance improvement is sensitive to a certain reward, the intensity of that reward is appropriately increased to maximize the incentive for employees to improve performance under the premise of controlling costs, while ensuring the balance of the enterprise's input-output benefits.

[0055] 5. Performance feedback and optimization form a virtuous cycle:

[0056] Visual reports and improvement suggestions to help employees improve: By using a visualization library to generate intuitive and interactive performance reports, employees can clearly understand their own performance advantages and disadvantages and their comparison with the team; Using text mining technology to generate targeted improvement suggestions to assist employees in clarifying the improvement direction and promoting personal performance improvement; For example, through a radar chart comparison, an employee finds that their work efficiency is lower than the team average, and combines the improvement suggestions to reasonably plan the work process, refer to excellent practices, and improve work efficiency;

[0057] Prediction and analysis early warning to assist management decision-making: Using time series prediction and causal analysis algorithms to predict performance changes and mine potential causes, and recommending management strategy adjustment plans based on reinforcement learning to help managers respond to problems in advance and optimize management decisions; For example, when it is predicted that the performance of a certain team has declined, and after analysis, it is found that it is due to increased task difficulty and insufficient training, managers can organize training and adjust task allocation according to the recommended plan to avoid the deterioration of performance problems;

[0058] Feedback optimization loop to continuously improve system effectiveness: The performance data after the implementation of the adjustment plan is fed back to the system, and each module optimizes its own algorithms and functions accordingly, realizing the continuous iteration and optimization of the performance appraisal management system, and improving the overall operation efficiency of the mine workshop; For example, the performance evaluation module optimizes the scoring model according to the impact of rewards on performance, and the task difficulty adjustment module more accurately adjusts the task difficulty based on the improvement of employees' abilities, so that the entire system continuously adapts to the actual operation needs of the workshop and continuously improves the management level and production efficiency;

[0059] 6. Overall improvement of the management level and competitiveness of the mine workshop:

[0060] Optimize resource allocation: Through accurate performance evaluation, reasonable task allocation and effective incentive mechanisms, employees and tasks are better matched, and resources such as equipment are more reasonably utilized, avoiding resource idleness or overconsumption, and improving the resource allocation efficiency of the mine workshop; For example, high-skilled employees are assigned to key and complex tasks to ensure the efficient completion of tasks, while avoiding low-skilled employees wasting time and resources on high-difficulty tasks;

[0061] Enhance employees' enthusiasm and sense of belonging: A fair, scientific and personalized performance appraisal management system enables employees' efforts to be accurately evaluated and reasonably rewarded, and the incentive mechanism meets the different needs of employees, enhancing employees' work enthusiasm and sense of belonging to the enterprise, and reducing personnel turnover; For example, when employees see the actual rewards and career development opportunities brought by the improvement of their own performance, they will be more actively involved in their work and their loyalty to the enterprise will increase;

[0062] Adapting to Market Changes and Enterprise Development: The system can adjust the assessment management strategy in a timely manner according to changes in the internal and external environment of the workshop (such as adjustments in business priorities, changes in market demand, etc.), continuously optimize the operation process and employee performance, and help mining enterprises maintain competitiveness in the market competition and achieve sustainable development. For example, when the market's requirements for product quality increase, the system can adjust the weights of performance evaluation factors to guide employees to focus on quality improvement, so that the enterprise's products meet market demands and gain a competitive advantage.

[0063] Through various innovations and optimizations, the performance assessment management method for mining workshops of the present invention comprehensively improves the performance assessment management level of mining workshops, brings significant advantages to mining enterprises in production operation, employee management, market competition, etc., and has high practical value and promotion significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic diagram of the steps of a performance assessment management method for a mining workshop of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0066] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the accompanying drawings of the specification and the specific embodiments.

[0067] As Figure 1 shown, an embodiment of the present invention provides a performance assessment management method for a mining workshop, including the following steps:

[0068] Step S1, data collection and preprocessing: Collect work data through sensor fusion, text analysis and determination of employee self-evaluation weights, and store it after multiple imputation and robust standardization processing.

[0069] Step S2, dynamic performance assessment: Based on a multi-factor weighted model, use the combination of neural network and reinforcement learning to optimize factor weights and predict employees' future performance, and calculate the performance score output in real time accordingly.

[0070] Step S3, task difficulty adjustment: Use the analytic hierarchy process and fuzzy comprehensive evaluation to quantify the task difficulty, combine with the analysis of employees' abilities, and rely on linear programming and gradient descent algorithms to adapt and optimize the task difficulty in real time.

[0071] Step S4, incentive mechanism: Use market supply and demand and collaborative filtering algorithms to adjust the integral value and customize the integral plan respectively, optimize the reward standard according to cost-benefit and performance sensitivity analysis, and realize integral exchange and data feedback.

[0072] Step S5, Performance Feedback and Optimization: Generate a performance report with the help of a visualization library and use text mining to generate improvement suggestions, predict and analyze the causes of performance changes through causal analysis, and feedback and optimize the management strategy adjustment plan recommended by reinforcement learning.

[0073] In this embodiment, the specific operation steps of Step S1 include:

[0074] Step S1-1, Data Collection:

[0075] Step S1-1-1, Sensor Data Collection and Fusion:

[0076] Install a variety of sensors on the key equipment in the mine workshop. For example, install a production sensor and a pressure sensor on the mining equipment, and install an operation duration sensor and a speed sensor on the transportation equipment. Each sensor collects the corresponding data in real time to form multiple sensor data sets, which are respectively recorded as ;

[0077] Use the Kalman filter algorithm to fuse the data collected by each sensor to improve the accuracy and reliability of the data. Taking two sensors as an example (which can actually be extended to multiple sensors), assume that the data collected by sensor 1 at time is , and the data collected by sensor 2 at time is . The fusion is completed through the following steps:

[0078] First, determine the observation matrix (set according to the characteristics of the sensor data and the fusion requirements according to the requirements of the Kalman filter algorithm);

[0079] Calculate the Kalman gain (calculate according to the recursive formula of the Kalman filter standard, combined with parameters such as the covariance of the sensor data);

[0080] Finally, the fused data is calculated according to the following formula: ;

[0081] Combine the fused data at all times to form a set as the basic data related to equipment operation and production;

[0082] Step S1-1-2, Text Data Collection and Quantification:

[0083] Employees input work-related text records through the workshop work record system. These records cover content such as task progress description, quality feedback, and problem records;

[0084] Analyze using natural language processing (NLP) technology, and the specific operations are as follows:

[0085] Use a word vector model (for example, select the Word2Vec algorithm) to convert the words in the text record into vector representations. For each text record , generate the corresponding vector ;

[0086] Quantitatively convert the vectors according to the predefined keyword - value mapping rules. For example, if a keyword combination related to high - quality task completion appears in the text, map it to the value 8 according to the rules; if there are expressions related to general completion situations, map it to the value 5, etc. Thus, convert the text record into quantified text data, which, together with the previous sensor fusion data, serves as the data content to be synthesized;

[0087] Step S1 - 1 - 3: Determine the weight of the employee's self - evaluation and synthesize the data:

[0088] Collect the employee's past self - evaluation data and the corresponding actual performance data, and set them as the employee self - evaluation data set and the actual performance data set ;

[0089] Construct a linear regression model (where is the intercept, is the regression coefficient, is the error term), and use the least - squares method to estimate and

[0090] Through the calculation formula of the least - squares method (here is the design matrix containing the independent variable , is the dependent variable vector), calculate the estimated values of and ;

[0091] According to value and indicators such as the goodness of fit of the model (such as value) to determine the weight of the employee's self - evaluation If is close to 1 and is relatively high, then takes a relatively large value, and vice versa;

[0092] According to the determined , the employee self-assessment data is combined with sensor fusion data and quantified text data according to weights to form a complete employee work data set covering task completion, work efficiency, work quality, equipment operation status, etc. ;

[0093] Step S1-1-4, data integration and transmission:

[0094] Integrate all types of collected data in a unified data format (such as JSON format) to ensure that the data structure is standardized and unified for subsequent processing;

[0095] With the help of big data message queue (such as Kafka) technology, the integrated data is transmitted to the data preprocessing server in real time to ensure the timeliness of the data so that it can enter the next step of preprocessing process in time;

[0096] Step S1-2, data preprocessing:

[0097] Step S1-2-1: Multiple imputation method (MICE) to fill missing values:

[0098] After the data arrives at the data preprocessing server, for the data matrix with missing values (This includes the employee work data set collected and integrated earlier , there may be some missing data) for processing;

[0099] For columns with missing values Every missing value in ( Indicates row index), multiple prediction models are established based on the correlation between data (for example, linear regression models and decision tree models are constructed using other non-missing columns as independent variables):

[0100] For each prediction model , use the training data (i.e. the non-missing part of the data) to train the model and obtain the missing values The predicted value of ;

[0101] go through After filling, missing values The final filling value Calculated by the following formula:

[0102] ;

[0103] In this way, all missing values ​​in the data are filled to improve the completeness of the data;

[0104] Step S1-2-2, Process data using RobustScaler:

[0105] Use the RobustScaler method to standardize the data with missing values filled, making the data conform to the characteristics of a normal distribution and insensitive to outliers;

[0106] For the original data , the standardized data is calculated according to the following formula:

[0107] ;

[0108] where is the median of the data , and is the interquartile range of the data (the difference between the upper quartile and the lower quartile). This calculation is performed for each data element in the data set to obtain the standardized employee work data set ;

[0109] Step S1-2-3, Store the preprocessed data:

[0110] Store the employee work data set after multiple imputation and RobustScaler processing in a distributed database (such as HBase) so that subsequent modules such as dynamic performance evaluation, task difficulty adjustment, incentive mechanisms, and performance feedback and optimization can smoothly call this data as the basic data source to carry out corresponding work;

[0111] Through the above detailed operations of Step S1 and its sub-steps, the whole process from data collection to preprocessing is completed, providing a high-quality and standardized data foundation for the entire mine workshop performance appraisal management system, ensuring that subsequent links can conduct corresponding analysis and decision-making based on accurate and reliable data.

[0112] In this embodiment, Step S2: The specific operation steps of dynamic performance evaluation include:

[0113] Step S2-1, Construction and initialization of the multi-factor weighted model:

[0114] Step S2-1-1, Determine the key factors for performance evaluation: Analyze the work characteristics of the mine workshop and the performance appraisal requirements, and determine factors such as task type, difficulty, employee historical performance, team collaboration effect, and innovation achievements as the key factors for performance evaluation. These factors are grouped into a set, denoted as ;

[0115] Step S2-1-2, Preliminary setting of factor weights: Based on expert experience and the analysis of the past historical performance data of the mine workshop, preliminary weights are set for each key factor to form a weight set , while satisfying conditions, and thus an initial multi-factor weighted model is constructed. The performance score is calculated according to the following formula in the initial stage: , is an index variable used to traverse all key factors, starting from 1 to ends. Among them, represents the total number of key factors for performance evaluation;

[0116] Step S2-2, Optimizing factor weights using neural network:

[0117] Step S2-2-1, Data preparation and input determination: Collect data on the recent performance fluctuations of employees (such as the change trend and fluctuation range of performance scores over a certain period of time) and information on the changes in the workshop business focus (such as the production indicators and quality improvement directions that the workshop has been focusing on recently), and organize these data into a format suitable for neural network input to form an input vector ;

[0118] Step S2-2-2, Construction of neural network structure:

[0119] Construct a neural network in deep learning, including an input layer, a hidden layer (determine the number of hidden layers and the number of nodes in each layer according to the actual situation. Let the number of nodes in the hidden layer be , select a suitable activation function, such as the ReLU function, etc., denoted as and an output layer:

[0120] For the calculation from the input layer to the hidden layer, let the weight matrix from the input layer to the hidden layer be and the bias vector be , and the output of the hidden layer is calculated according to the following formula:

[0121] ;

[0122] For the calculation from the hidden layer to the output layer, let the weight matrix from the hidden layer to the output layer be and the bias vector be , and the output of the output layer (i.e., the adjusted factor weight vector is obtained through the following formula:

[0123] ;

[0124] Step S2-2-3, Weight adjustment training process:

[0125] Define an appropriate loss function (such as the mean squared error loss function, denoted as , where is the true factor weight label, is the weight value predicted by the neural network, is the number of samples, is the sample index variable), and through the backpropagation algorithm, calculate the gradient based on the loss function, and continuously adjust the weights of the neural network based on optimization algorithms such as gradient descent , so that the predicted factor weight constantly approaches the true optimal weight, and realizes real-time adjustment of the weights of each factor in the weighted model based on the fluctuations of employee performance and changes in the key points of workshop operations;

[0126] Step S2-3, construct a prediction model by combining reinforcement learning:

[0127] Step S2-3-1, prepare training data and define environmental elements:

[0128] Collect a large amount of historical performance data of the mine workshop to construct a training data set,

[0129] Set it as , where is the employee feature vector (including various features such as employee basic information, past task completion data, and skill mastery), is the corresponding performance score. At the same time, define the relevant elements in reinforcement learning;

[0130] State space: Define the various comprehensive states before each task execution by the employee as the state space , for example, including state factors such as the current employee skill level, recent performance, and task difficulty faced;

[0131] Action space: Regard each work task executed by the employee as an "action" in reinforcement learning to form an action space , and different tasks can be regarded as different action selections;

[0132] Transition probability: Determine the transition probability , indicating the probability of transitioning from state to state by taking action . This probability can be obtained through statistical analysis of historical data and reflects the possibility of state changes after task execution;

[0133] Reward function: Use the performance score as the "reward" to construct a reward function , that is, when the employee executes action in state The performance score obtained after (completing the corresponding tasks) is used as reward feedback to measure the quality of behavior;

[0134] Step S2-3-2, constructing the XGBoost model foundation:

[0135] Use the XGBoost algorithm to construct a machine learning prediction model, and its objective function is composed of the loss function (such as the commonly used logarithmic loss function, etc., used to measure the predicted performance score and the true performance score The difference between them) and the regularization term (used to prevent overfitting and ensure the generalization ability of the model), and the specific formula is as follows: , where is the sample index;

[0136] Based on the training data set , train the XGBoost model so that it can make a preliminary prediction of employee performance;

[0137] Step S2-3-3, the reinforcement learning optimization process:

[0138] Combined with the Markov decision process in reinforcement learning, let the intelligent agent follow a certain policy (indicating the probability of taking action in state ) to explore and learn in the past task scenarios (i.e., the environment such as the state space and action space constructed by historical data):

[0139] Use common reinforcement learning algorithms such as Q-learning to update the value function (indicating the expected cumulative reward of taking action in state ), the intelligent agent continuously tries different actions (performing different tasks), and learns the strategy of choosing the optimal action in different states according to the obtained reward (performance score) feedback, so as to discover the optimal performance improvement strategy of employees in different situations, and further optimize the prediction model constructed by XGBoost, so that it can more accurately predict the future performance of employees;

[0140] Step S2-4, performance score calculation and output:

[0141] Step S2-4-1, data input and score calculation:

[0142] Input the current work data of employees (covering various aspects such as task completion, work efficiency, and work quality) obtained from the previous data collection and preprocessing steps into the multi-factor weighted model optimized through the above steps. At the same time, combine the results output by the prediction model optimized by XGBoost combined with reinforcement learning. According to the calculation logic of the multi-factor weighted model (i.e., At this time, is the optimized weight, is the value of each factor corresponding to the current work data, represents the total number of key factors for performance evaluation, is the index variable used to traverse all key factors), and calculate the specific performance score of each employee in real time;

[0143] Step S2-4-2, Data Output and Transmission:

[0144] Output the calculated performance score and the score situation of each factor (i.e., each value) and other relevant data to the subsequent task difficulty adjustment module and incentive mechanism module in the form of an API interface, providing data support for subsequent operations such as task difficulty adjustment and incentive measure formulation, and ensuring the smooth connection and continuous progress of the entire performance appraisal management process;

[0145] Through the above detailed steps S2 and its sub-steps, based on the multi-factor weighted model, combined with neural networks and reinforcement learning to optimize the factor weights, accurately predict the future performance of employees, and finally calculate and output the performance score in real time, providing a scientific and dynamic evaluation basis for the performance appraisal management of the mine workshop.

[0146] In this embodiment, Step S3: The specific operation steps of task difficulty adjustment include:

[0147] Step S3-1, Task Difficulty Quantification and Analysis:

[0148] Step S3-1-1, Construct a comprehensive task difficulty evaluation model using the Analytic Hierarchy Process (AHP):

[0149] Construct a hierarchical structure model:

[0150] Define the goal layer as task difficulty , representing the final task difficulty result to be evaluated;

[0151] Determine the factors included in the criterion layer, such as skill level (covering operation skill complexity, knowledge skill requirements, etc.), time urgency , and resource dependence And other relevant factors incorporated according to the actual situation of the mine workshop (such as the degree of safety risk, etc.) are used to construct a hierarchical structure to clearly show the hierarchical relationship among various factors;

[0152] Construct a judgment matrix:

[0153] For pairwise comparison of each factor in the criterion layer, a judgment matrix is constructed based on their relative importance degree , where represents the importance degree of factor relative to , and the 1-9 scale method is used for assignment; for example, if is equally important as , then ; if is slightly more important than , then ; and so on until the construction of the entire judgment matrix is completed;

[0154] Calculate the weight vector and consistency test:

[0155] Solve the judgment matrix to obtain the maximum eigenvalue and its corresponding eigenvector Common calculation methods such as the power method, etc. (which can be implemented with the help of mathematical calculation software);

[0156] Calculate the consistency index ( is the order of the judgment matrix), and compare it with the average random consistency index (which can be obtained by querying the corresponding standard table for the value of the corresponding value) for consistency test. When is less than , it is considered that the consistency of the judgment matrix passes. At this time, the eigenvector is normalized to obtain the weight vector of each factor in the criterion layer, and this weight vector reflects the relative importance proportion of each factor in the criterion layer when measuring the task difficulty;

[0157] Step S3-1-2, use the fuzzy comprehensive evaluation method to determine the task difficulty value:

[0158] Determine the comment set:

[0159] According to the common evaluation grades of the task difficulty in the mine workshop, set the comment set , represents the th comment element in the comment set. For example , which is used to describe different degrees of task difficulty;

[0160] Construct the fuzzy relation matrix:

[0161] For the sub-factors under each factor at the criterion level, conduct fuzzy evaluations respectively to determine the membership degrees of each sub-factor to the evaluation comments in the evaluation comment set and construct the fuzzy relation matrix , where represents the membership degree of factor to the evaluation comment . The determination of the membership degree can be completed through methods such as expert scoring and historical data statistical analysis. For example, invite multiple experts familiar with the business of the mine workshop to evaluate the task difficulty levels corresponding to each sub-factor, and count the frequencies of each evaluation comment as the membership degree values;

[0162] Fuzzy comprehensive evaluation calculation:

[0163] Combine the weight vector of the factors at the criterion level with the fuzzy relation matrix , and use the fuzzy comprehensive evaluation formula to calculate ( represents the fuzzy composition operator, and commonly used ones include the dominant factor type , the weighted average type , etc. represents the maximum operation, represents the minimum operation), and obtain the fuzzy comprehensive evaluation result vector of the task difficulty, where represents the factor index in the factor set, and

[0164] Defuzzification processing:

[0165] Convert the fuzzy comprehensive evaluation result vector into a specific task difficulty value through defuzzification methods. For example, use the centroid method, and the calculation formula is (where is the th element in , is the quantization value corresponding to the th evaluation comment in the evaluation comment set , is the final task difficulty value; is the number of evaluation comments in the evaluation comment set ), and store the calculated task difficulty value in the task management database to provide an accurate reference standard for subsequent task difficulty adjustment;

[0166] Step S3-1-3, Employee ability analysis:

[0167] Employee Classification (Cluster Analysis):

[0168] Collect the historical performance data of employees (such as performance scores at each stage, task completion rates, etc.) and skill mastery data (including proficiency levels of various skills, acquisition of skill certificates, etc.), and form data sets respectively. Let the performance data set be and the skill mastery data set be ( is the number of employees);

[0169] Calculate the similarity between employees. Commonly used similarity calculation methods such as Euclidean distance (where represents the th eigenvalue of the th employee, is the feature dimension, and the feature dimension here covers data related to performance and skill mastery and other related dimensions), or other appropriate similarity measurement methods can also be selected according to the actual situation;

[0170] Apply the cluster analysis algorithm (such as the K-Means algorithm). Based on the calculated similarity between employees, divide the employees into different ability level categories. By continuously iteratively updating the cluster centers (in the K-Means algorithm, initially randomly select cluster centers, and then reassign the employees' cluster membership according to the distance between the employees and the cluster centers, and update the cluster centers. Iterate repeatedly until the cluster centers no longer change or reach the preset stop conditions such as the number of iterations), so that the similarity within the same cluster is high and the similarity between different clusters is low, and finally obtain employee ability categories;

[0171] Analyze the Adaptation Relationship between Tasks and Employee Abilities (Bayesian Network Model):

[0172] For each category of employees, use the Bayesian network model to analyze the probability relationship between different difficulty tasks and the positive work efficiency of employees. Let the task difficulty variable be and the employee efficiency variable be ;

[0173] Based on the historical data of the mine workshop (including the work efficiency data generated after different difficulty tasks are assigned to different categories of employees in the past, such as the improvement of work efficiency, the quality of task completion, etc.), learn the structure of the Bayesian network (i.e., the dependency relationship graph between variables) and the conditional probability table , and through the inference mechanism of the Bayesian network, analyze the probability of positive work efficiency that different difficulty tasks may bring for a certain category of employees, providing a scientific reference basis for subsequent reasonable task allocation;

[0174] Step S3-2, Task Difficulty Real-time Adjustment Strategy:

[0175] Step S3-2-1, Determine the Task Difficulty Adjustment Range (Linear Programming):

[0176] Objective Function Setting:

[0177] According to the employee performance score obtained from the performance evaluation module and the task completion situation, combined with the task difficulty evaluation model and the analysis result of the adaptation between employee ability and task, set an appropriate objective function. For example, the objective is to maximize performance improvement or employee satisfaction. Let the employee performance score be , the current task difficulty be , the expected interval of task difficulty suitable for employee ability be (obtained through Bayesian network analysis, etc.), the objective function can be set as (where is the expected performance score of the employee after the task difficulty is adjusted, is the estimated value of employee satisfaction, and are the weight coefficients set according to the enterprise's emphasis level);

[0178] Constraint Conditions Determination:

[0179] Define the constraint of the task difficulty adjustment range , ensure that the adjusted task difficulty is within the reasonable range suitable for employee ability. At the same time, consider other actual constraint conditions, such as the limitation of task required resources (for example, some high-difficulty tasks require specific equipment resources. If the resources in the current workshop are limited, the allocation of such high-difficulty tasks needs to be restricted), the matching situation between task time requirements and the available time of employees, etc. Add these constraint conditions to the linear programming model in the form of mathematical expressions;

[0180] Solve for the Task Difficulty Adjustment Value:

[0181] Use a linear programming solver (such as common algorithms like the simplex method) to solve the constructed linear programming model and calculate the optimal task difficulty adjustment value . This value is the specific numerical value of the task difficulty adjustment range most suitable for the employee at the current stage. Based on this, adjust the task allocation. For example, for employees with continuously high performance scores and whose ability adaptation analysis shows that they are more suitable for high-difficulty tasks, appropriately increase the difficulty of the tasks they will undertake in the future. On the contrary, for employees with poor performance, appropriately reduce the task difficulty;

[0182] Step S3-2-2, Task Difficulty Adjustment Effect Monitoring and Iterative Optimization (Gradient Descent Algorithm):

[0183] Monitoring data collection:

[0184] During the process of employees performing tasks, by combining wearable devices (such as smart bracelets, etc., which can obtain physiological data of employees such as heart rate and blood pressure, indirectly reflecting the emotional state and work pressure of employees) and workshop video monitoring (analyzing employees' facial expressions, body movements, etc. through image recognition technology to judge the emotional state of employees), the real-time emotional state of employees is obtained. With the help of emotion recognition algorithms (such as classification models based on deep learning convolutional neural networks, etc.), the emotional state of employees is classified and judged into categories such as positive, neutral, and negative;

[0185] At the same time, the time series analysis method is used to depict the work efficiency tracking curve of employees, and collect data related to work efficiency during the task execution process of employees (such as output per unit time, product qualification rate, etc.). Let the time series data be , where represents the work efficiency data collected at time . It can be smoothed by methods such as moving average and exponential smoothing, and a suitable time series model (such as autoregressive model : (where is the autoregressive coefficient, is the error term, represents the time series 's past lagged values, ) is used to analyze the trend of efficiency change;

[0186] Effect judgment and adjustment decision:

[0187] According to the data such as the real-time emotional state of employees and the work efficiency tracking curve collected, judge whether the adjustment of task difficulty has achieved the expected effect. For example, if the emotional state of employees remains negative and the work efficiency shows a downward trend, it can be considered that the adjustment of task difficulty has not achieved the expected, and further optimization is needed;

[0188] Gradient descent iterative adjustment:

[0189] If it is judged that the adjustment of task difficulty has not achieved the expected effect, based on the gradient descent algorithm, quickly iterate and adjust the task difficulty. Let the task difficulty adjustment amount be , define a suitable loss function (such as using the sum of squares of the difference between employees' work efficiency and expected efficiency as the loss measurement index), and according to the update formula of the gradient descent algorithm (where is the current task difficulty, is the learning rate, which needs to be reasonably set according to the actual situation, is the loss function at (the gradient at ), iteratively update the task difficulty to ensure that employees maintain a positive working state under tasks of appropriate difficulty, and timely feedback the adjusted task assignment information to the employee work terminals and update the relevant task information in the task management database for continuous tracking and further optimization of the task difficulty adjustment;

[0190] Through the above detailed steps S3 and its sub-steps, the task difficulty is quantified based on the analytic hierarchy process and fuzzy comprehensive evaluation. Combining the analysis of employees' abilities, relying on linear programming and gradient descent algorithms to adapt and optimize the task difficulty in real time, making the task assignment in the mine workshop more scientific and reasonable, which helps to improve employees' work efficiency and overall production efficiency.

[0191] In this embodiment, step S4: The specific operation steps of the incentive mechanism include:

[0192] Step S4-1, Design and management of the integral system:

[0193] Step S4-1-1, Adjust the integral value based on the market supply and demand theory:

[0194] Data collection and analysis:

[0195] Collect the popularity data and redemption frequency data of different rewards (such as cash rewards, training opportunities, promotion opportunities, physical prizes, etc.) among the employees in the mine workshop. These data can be obtained through channels such as the enterprise internal reward redemption platform records and employee questionnaire feedback; for example, count the number of times each reward is browsed by employees and the number of times of applying for redemption within a certain period of time to reflect the demand degree of employees for different rewards;

[0196] Simulate the market supply and demand relationship:

[0197] According to the collected data, simulate the principle of market supply and demand to dynamically adjust the integral value corresponding to the reward; for rewards with strong demand (that is, high popularity and high redemption frequency), increase their corresponding integral value according to certain rules to make them relatively "more expensive"; on the contrary, for those rewards with low demand, reduce their integral value to make it easier for employees to redeem them by accumulating fewer points; for example, if a certain training opportunity is favored by many employees recently and the number of applicants for redemption increases, increase its corresponding integral value by a pre-set ratio (such as 10%) every certain period (such as weekly or monthly); for a certain type of physical prize that has been neglected for a long time, reduce its integral value by a certain ratio accordingly;

[0198] Step S4-1-2, Customize the integral plan using the collaborative filtering algorithm:

[0199] Construct a user-item rating matrix:

[0200] Let the set of employees be representing users, and the set of rewards be representing items, construct a rating matrix of users for items , where represents user 's rating of item . The rating can be customarily quantified and assigned based on factors such as the employees' preference for rewards and their satisfaction after past exchanges. If an employee has not rated a certain reward, it is recorded as a missing value. For example, an employee can give a high score (such as 5 points) to a reward they are very looking forward to and a low score (such as 1 point) to a reward they are less interested in. Initially, some rating data can be collected through methods such as questionnaires or the employees' initial evaluations on the reward redemption platform to construct the matrix;

[0201] Calculate user similarity:

[0202] Based on the rating matrix of users , use a suitable similarity calculation method to calculate the similarity between users. Commonly used methods such as the Pearson correlation coefficient method, and its calculation formula is: ;

[0203] where and are two different users, is the set of items that both user and have rated, and are the average ratings of users and respectively; represents the similarity between user and ; represents user 's rating of item . By calculating the similarity values between each pair of users through this formula, a user similarity matrix is formed to measure the similarity degree among employees in terms of reward preferences;

[0204] Find neighboring users and predict ratings:

[0205] For each target user , find the users most similar to them (i.e., neighboring users) from all users, and form a neighboring set ; Usually, they can be sorted in descending order according to the similarity values, and the top users are selected as neighbors;

[0206] Predict the target user by using the rating data of neighboring users through methods such as weighted average For unrated items Rating , for example, the prediction formula based on weighted average of user similarity is:

[0207] ;

[0208] According to the predicted rating situation, analyze the potential preference degree of employees for different rewards, and combine factors such as employees' work performance (such as performance ratings, skill improvement situations, etc.) and job requirements to customize personalized point acquisition rules and distribution plans for each employee. For example, for employees with excellent performance and a predicted high interest in a certain professional skill training reward, formulate a plan that can quickly accumulate the points required for this training reward by completing difficult tasks; for newly recruited employees, design a point acquisition method and distribution ratio that focuses more on helping them familiarize themselves with the work environment and obtain basic skill improvement opportunities;

[0209] Employee point acquisition and storage:

[0210] After employees complete various work tasks, they obtain corresponding points according to their performance ratings, task difficulty completion coefficients and other factors according to the personalized point plan customized for them. For example, completing a difficult and high-quality task can obtain more points, while regular tasks obtain relatively fewer points;

[0211] Store the point data obtained by employees in the employee point account database, which is convenient for employees to query their point balances and point acquisition details at any time, and is also convenient for subsequent point redemption operations and data statistical analysis;

[0212] Step S4-2, Dynamic adjustment of reward standards:

[0213] Step S4-2-1, Cost-benefit analysis:

[0214] Cost accounting:

[0215] For each reward, carefully calculate the costs involved, including direct costs (such as the amount of cash rewards, the cost of purchasing physical prizes, etc.) and indirect costs (such as the teacher fees and venue rental fees involved in providing training opportunities, the future salary growth costs corresponding to promotion opportunities, etc.), and summarize the costs of each item to form the total cost data of the reward;

[0216] Continuously track and record the cost changes of various rewards in different periods. For example, as the market price fluctuates, the procurement cost of physical prizes may rise or fall; as the training teacher level improves, the training cost may also increase accordingly, etc., and update the cost data in a timely manner to ensure the accuracy of cost accounting;

[0217] Benefit evaluation:

[0218] Analyze the benefits brought to the enterprise after the issuance of rewards, mainly measured from the perspective of the improvement of employee performance. For example, compare the changes in performance scores, work efficiency improvement, and loyalty changes to the enterprise before and after employees receive rewards. Establish a quantitative evaluation index system (such as setting indicators like performance improvement rate, work efficiency growth rate, etc.) to specifically evaluate the benefits brought by the rewards;

[0219] At the same time, consider the positive impacts of rewards on aspects such as the team cooperation atmosphere and employees' innovation enthusiasm, and comprehensively evaluate the overall benefits of the rewards. For example, observe whether the internal cooperation projects of the team are more smooth and efficient under the incentive of the rewards, whether employees put forward more innovative working methods or suggestions, etc., and try to convert these qualitative influencing factors into quantifiable data and incorporate them into the scope of benefit evaluation;

[0220] Step S4-2-2, Optimize the reward standard using the dynamic programming algorithm:

[0221] Construct a decision model:

[0222] Based on the data of cost-benefit analysis, combined with the business objectives of the enterprise at different stages (such as increasing production in the short term, enhancing the innovation ability of the enterprise in the long term, etc.) and financial status (such as budget constraints, cash flow conditions, etc.), construct an optimization decision model for the reward standard based on dynamic programming, define state variables (such as reward status corresponding to different performance intervals, financial status of the enterprise at different time periods, etc.), decision variables (such as specific reward amounts, selection of reward types, etc.), and state transition equations (mathematical relationships describing how the reward standard changes over time, with changes in performance, etc.);

[0223] Solve for the optimal reward standard:

[0224] Solve the constructed model through the dynamic programming algorithm (such as using conventional solution methods like the backward induction method), calculate the optimal reward standards in different situations such as at each stage and each performance interval under the current business conditions and objectives of the enterprise. For example, determine which combination of reward amount and reward type can maximize employee performance and overall benefits while meeting the enterprise's cost budget within a certain performance interval, record and update these optimal reward standards as the actual implemented reward standards;

[0225] Step S4-2-3, Performance sensitivity analysis and fine-tuning:

[0226] Analyze the sensitivity of the performance score to changes in the reward standard:

[0227] Using statistical analysis methods (such as variance analysis, elasticity analysis, etc.), deeply study the relationship between the changes in performance scores and the changes in reward criteria, calculate the sensitivity coefficient of performance scores to changes in reward criteria (for example, by methods such as regression analysis, establish a functional relationship between performance scores and reward criteria, and derive the sensitivity coefficient), and understand the sensitivity of employees' performance to reward adjustments in different performance intervals. For example, it is found that employees in the medium to high performance intervals are more sensitive to the increase in reward amount with the improvement of their performance; while for employees with lower performance, they may be more concerned about the diversification of reward types rather than simply the increase in amount;

[0228] Fine-tune the reward criteria:

[0229] According to the results of sensitivity analysis, fine-tune the reward criteria corresponding to different performance intervals. For performance intervals with high sensitivity, appropriately increase the reward intensity (such as increasing the reward amount, adding more attractive reward types, etc.); for intervals with low sensitivity, optimize the reward structure (such as adjusting the combination of reward types, setting hierarchical reward conditions, etc.), so that the reward criteria are more scientific, reasonable and targeted, and can better motivate employees to improve their performance;

[0230] Step S4-3, Points Exchange and Data Feedback:

[0231] Step S4-3-1, Points Exchange Operation:

[0232] Based on the points they have accumulated and the current reward criteria, employees independently select and exchange corresponding reward items on the reward exchange platform established within the enterprise. The exchange process needs to follow the procedures set by the platform, such as first logging in to the personal account, viewing the list of available rewards, deducting the corresponding points after confirmation, and generating an exchange record, etc., to ensure the standardization and traceability of the points exchange operation;

[0233] Step S4-3-2, Data Feedback Mechanism:

[0234] The incentive mechanism module timely feeds back the employees' points exchange records (including detailed information such as exchange time, exchanged rewards, consumed points, etc.) and the data on the changes in employees' performance after the rewards are issued (such as the fluctuations in performance scores and the changes in work efficiency within a certain period after the rewards are exchanged) to other relevant modules such as the performance evaluation module and the task difficulty adjustment module, so that each module can conduct corresponding optimization analysis based on this feedback information. For example, the performance evaluation module can further optimize the performance scoring model according to the impact of rewards on performance; the task difficulty adjustment module can refer to the improvement of employees' abilities under the incentive of rewards and more accurately adjust the task difficulty range, etc., to achieve the collaborative optimization and continuous improvement of the entire performance appraisal and management system;

[0235] Through the above detailed steps S4 and its sub - steps, a reasonable design of the points system is achieved through market supply - demand and collaborative filtering algorithms, the reward criteria are scientifically optimized based on cost - benefit and performance - sensitivity analysis, and points redemption and data feedback are effectively completed, thus constructing a complete and flexible incentive mechanism to better motivate employees in the mine workshop and improve the overall operation efficiency and performance level.

[0236] In this embodiment, step S5: The specific operations of performance feedback and optimization include:

[0237] Step S5 - 1: Generation of performance reports and provision of improvement suggestions:

[0238] Step S5 - 1 - 1: Generate performance reports with the help of a visualization library:

[0239] Data collection and collation:

[0240] Collect multi - faceted performance information of employees from various relevant data sources (such as the employee performance scores and task completion data output by the performance evaluation module, the task assignment and difficulty data of the task difficulty adjustment module, the points acquisition and redemption records of the incentive mechanism module, etc.). Organize these data according to certain rules and formats to ensure the integrity and accuracy of the data. For example, map different types of performance data of each employee to corresponding data fields to form a structured data set for subsequent visualization processing;

[0241] Select the type of visualization chart:

[0242] According to the characteristics of the performance information to be presented and the analysis purpose, select a suitable type of visualization chart and use the D3.js visualization library to generate performance reports; for example:

[0243] If you want to show the inter - relationships between various performance factors of employees (such as task completion quality, work efficiency, teamwork, etc.) and their proportions in the overall performance, you can choose a Sankey diagram, and intuitively reflect the associations and weight situations between factors through lines of different widths and flow directions;

[0244] If you want to compare the comprehensive performance of employees at different stages (such as monthly, quarterly) and the differences from the team average level, it is more appropriate to use a radar chart. Map the performance indicator values at different stages to the coordinate axes of the radar chart to clearly show the change trend and relative position of the performance;

[0245] Generation and display of visualization reports:

[0246] Using the relevant functions and syntax of the D3.js library, according to the requirements of the selected chart type, visualize and render the organized data to generate an intuitive and interactive performance report. For example, by writing JavaScript code to configure parameters such as the axes, data binding, and style settings of the chart, the performance report can present a good visual effect.

[0247] Deploy the generated performance report to the management platform inside the mine workshop. Employees and managers can log in to their respective accounts and conveniently view the performance reports they are concerned about on the web page. Employees can intuitively understand their own strengths and weaknesses in the work process and the comparison with the overall team level, providing a reference basis for self-improvement.

[0248] Step S5-1-2: Generate improvement suggestions using text mining:

[0249] Data preprocessing:

[0250] Collect performance data and employees' daily work records (such as text content like task summaries, problem feedback, experience sharing, etc.) as the data source for text mining. Clean these text data to remove noise information (such as redundant punctuation marks, spaces, repetitive meaningless content, etc.). At the same time, perform basic natural language processing operations such as lexical analysis and part-of-speech tagging to convert the text into a format convenient for subsequent analysis. For example, split the text into meaningful words and mark their part of speech.

[0251] Topic extraction and key information mining:

[0252] Use text mining techniques (such as topic model algorithms, for example, the LDA algorithm) to deeply analyze the preprocessed text data, and automatically extract the key topics and problem points. For example, analyze the text of employees' work records through the LDA algorithm to mine topic information such as "equipment failures affecting task progress" and "room for improvement in communication and collaboration efficiency" that reflect common problems and key aspects of concern in the work.

[0253] Combined with predefined business rules and key indicators, further screen and refine the key information closely related to performance improvement. For example, for text related to task completion, focus on key descriptive content that indicates task timeouts, substandard quality, etc. that affect performance.

[0254] Improvement suggestion generation:

[0255] Based on the mined key themes and information, combined with the rules set by the expert system (these rules are summarized from the business experience and best practices of the mine workshop. For example, if the situation of "equipment failure affecting task progress" occurs, it is recommended to strengthen the daily maintenance of equipment and formulate a contingency plan for equipment failure) and the pre-trained natural language generation model (which can be trained based on a large number of existing performance improvement case texts to enable it to generate suggestion content that conforms to language logic and actual business scenarios), targeted and highly operable improvement suggestions are automatically generated for employees. For example, for the problem of low work efficiency in employees' performance, specific suggestion content such as "It is recommended to reasonably plan the work process, refer to the operation methods of excellent employees in the same position, and make preparations for tasks in advance to avoid unnecessary time waste" is generated, and these improvement suggestions are associated and displayed with the corresponding performance reports to assist employees in better improving their performance levels;

[0256] Step S5-2, Performance Change Early Warning and Management Strategy Recommendation:

[0257] Step S5-2-1, Performance Change Prediction:

[0258] Data Preparation and Selection of Prediction Model:

[0259] Collect the past performance data of the team and individuals (including performance scores in different time periods, detailed data of each performance factor, etc.), and organize these historical performance data into a time series data format in chronological order. For example, for the performance score data of a single employee, a time series similar to is formed, where represents the performance score value at time ;

[0260] According to the characteristics of the performance data (such as whether there are seasonal fluctuations, trend changes, etc.), select an appropriate time series prediction algorithm. Common ones include the ARIMA model (suitable for stationary time series, which makes the data stationary through operations such as differencing and then makes predictions), the LSTM model in deep learning (which has a good prediction effect on time series with complex non-linear characteristics and long-term dependence relationships), etc., and construct the corresponding prediction model;

[0261] Model Training and Prediction:

[0262] Use the historical performance data to train the selected prediction model. For example, in the ARIMA model, by estimating parameters such as the autoregressive order, differencing order, and moving average order of the model, and using methods such as maximum likelihood estimation to determine the optimal parameter values to enable the model to fit the historical data; in the LSTM model, configure an appropriate network structure (such as the number of hidden layer nodes, the number of layers, etc.), and train and optimize the model through algorithms such as backpropagation;

[0263] Apply the trained model to the prediction of future performance data. Input the corresponding feature data according to the current time point (such as recent performance, external environmental factors, etc., which can be appropriately selected and preprocessed according to the model requirements), and output the predicted values of performance changes within a certain period in the future (such as the next month or quarter). For example, predict that the performance score of a certain team may show a downward trend in the next quarter;

[0264] Step S5-2-2, Causal analysis to mine the reasons for performance changes:

[0265] Select a causal analysis algorithm:

[0266] Apply a suitable causal analysis algorithm (such as Granger causality test, etc.). This algorithm judges whether there is a causal relationship between variables based on the lag relationship between time series data, and is suitable for analyzing the causal relationship between different factors (such as task difficulty changes, incentive measure adjustments, employee skill improvements, etc.) and performance changes in performance data;

[0267] Causal relationship mining:

[0268] Use performance data and relevant factor data that may affect performance (such as task difficulty indicators, incentive points distribution, employee training participation, etc.) as input, conduct causal analysis operations such as Granger causality test, and mine the potential reasons for performance changes. For example, through analysis, it is found that the recent sudden increase in task difficulty and the lack of corresponding skill training support for employees are important reasons for the decline in performance scores, or it is found that there is a significant causal relationship between the newly implemented incentive measures and performance improvement, etc., providing a basis for formulating management strategy adjustment plans in the future;

[0269] Step S5-2-3, Recommend management strategy adjustment plans based on reinforcement learning:

[0270] Define the elements of the reinforcement learning environment:

[0271] Clarify the key elements such as the state space, action space, and reward function in reinforcement learning; the state space can be defined as a comprehensive state description composed of multiple factors including the current performance status (such as performance score, performance change trend, etc.), employee status (such as skill level, work attitude, etc.), and enterprise operation environment (such as market demand changes, resource allocation, etc.); the action space is the different management strategy adjustment behaviors available for managers to choose (such as adjusting the task allocation method, changing the incentive mechanism, organizing training activities, etc.); the reward function is set according to the performance changes after the implementation of the management strategy. For example, if the performance is significantly improved, a higher reward value is given, otherwise a lower reward value is given to measure the effectiveness of the management strategy adjustment;

[0272] Construct a reinforcement learning model and train it:

[0273] Select a suitable reinforcement learning algorithm (such as Q-learning algorithm, deep Q-network algorithm, etc.) to build a reinforcement learning model. Use the historical management strategy adjustment records and the corresponding performance change feedback data of the mine workshop as training samples. Let the intelligent agent continuously explore and learn in the simulated management decision-making environment. By repeatedly trying different management strategy adjustment actions, update its own strategy value function (such as Q-value function) according to the obtained reward feedback, and learn the ability to select the optimal management strategy adjustment plan in different states;

[0274] Management strategy recommendation:

[0275] When it is predicted that the performance may decline or potential problems are found, based on the current actual state (determine the current performance state, employee state, etc. by collecting and analyzing relevant data in real time), use the trained reinforcement learning model to recommend the optimal management strategy adjustment plan for managers. For example, recommend specific adjustment suggestions such as "carry out special skill training for employees with poor performance, adjust their task assignments, reduce the task difficulty, and at the same time increase corresponding incentive measures to encourage them to improve their performance". Managers can perform corresponding management strategy adjustment operations in the workshop management system according to the recommended plan;

[0276] Step S5-3, feedback optimization loop:

[0277] Step S5-3-1, implementation of the adjustment plan and performance data feedback:

[0278] Managers perform corresponding adjustment operations in the actual management work of the mine workshop according to the recommended management strategy adjustment plan. For example, adjust the task assignments of employees, organize relevant training activities, modify the incentive mechanism, etc. according to the suggestions to ensure that the adjustment plan can be implemented;

[0279] After the implementation of the adjustment plan, continuously collect and record performance change data (such as changes in employee performance scores, improvements in task completion, etc.) and other relevant feedback information (such as employees' satisfaction with the adjustment plan, changes in the team cooperation atmosphere, etc.) to form a new performance data set, providing a basis for subsequent optimization analysis;

[0280] Step S5-3-2, system optimization and iterative improvement:

[0281] The new performance data collected after implementing the adjustment plan is fed back into the entire performance appraisal management system. Each relevant module (such as the performance evaluation module, task difficulty adjustment module, incentive mechanism module, etc.) uses the corresponding algorithms and models (such as adjusting the weights in the dynamic adjustment multi-factor weighted model, updating the parameters of the task difficulty evaluation model, optimizing the reward criteria, etc.) for its own optimization analysis based on this feedback information, further improving its respective functions and algorithm logics, and realizing the continuous iterative optimization of the entire performance appraisal management system, continuously enhancing the performance appraisal management level and overall operation efficiency of the mine workshop;

[0282] Through the above detailed steps of S5 and its sub-steps, the performance feedback and optimization function is fully realized. With the help of visualization means and intelligent analysis technology, it not only provides clear and intuitive performance information and improvement suggestions for employees and managers, but also can recommend reasonable management strategy adjustment plans through scientific prediction, causal analysis, and reinforcement learning, and continuously improve the effectiveness and adaptability of the entire performance appraisal management system of the mine workshop through the feedback optimization loop, helping the mine workshop to operate efficiently and stably.

[0283] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A mine workshop performance assessment management method, characterized by: The following steps are involved: Step S1, data collection and preprocessing: collect work data through sensor fusion, text analysis and employee self-assessment weights, and store it after multiple filling and robust standardization; Step S2, dynamic performance evaluation: Based on a multi-factor weighted model, neural networks and reinforcement learning are combined to optimize factor weights and predict employees' future performance, and the performance score output is calculated in real time; Step S3, task difficulty adjustment: using analytic hierarchy process and fuzzy comprehensive evaluation to quantify task difficulty, combined with employee capability analysis, relying on linear programming and gradient descent algorithm to adapt and optimize task difficulty in real time; Step S4, incentive mechanism: using market supply and demand and collaborative filtering algorithms to adjust the points value and customize the points plan, optimizing the reward standards based on cost-effectiveness and performance sensitivity analysis, and realizing points redemption and data feedback; Step S5, performance feedback and optimization: Generate performance reports with the help of visualization libraries and generate improvement suggestions using text mining, warn of performance changes through prediction and causal analysis, and use reinforcement learning to recommend management strategy adjustment plans and provide feedback for optimization; In the task difficulty adjustment, the hierarchical analysis method includes: Construct a hierarchical model and set the target layer as the task difficulty ,The criterion layer contains relevant factors, and the judgment matrix is ​​constructed by comparing the factors in the criterion layer pairwise. ,in, Indication factors Relative to The importance of Scaling method assignment; Solving the Judgment Matrix by Eigenvector Method The largest characteristic root and its corresponding eigenvector , calculate the consistency index ,in, To determine the matrix order and check consistency, Less than average random consistency index When , the consistency is considered to be passed, and the normalized feature vector That is, the weight vector of each criterion layer factor ; The fuzzy comprehensive evaluation method includes: Set of comments , fuzzy evaluation is performed on the sub-factors under each criterion layer, and a fuzzy relationship matrix is ​​constructed. ,in, Indication factors For the comment collection Middle The membership degree of each comment is combined with the weight vector of the criterion layer factors , using the fuzzy comprehensive evaluation formula , and obtain the fuzzy comprehensive evaluation result vector of task difficulty ,in, represents the fuzzy synthesis operator, which is then converted into a specific task difficulty value through the defuzzification method. The defuzzification method is the centroid method, and the calculation formula is: ,in, for Middle elements, For comment collection Middle The quantitative value corresponding to each comment, The final task difficulty value is stored in the task management database as a reference for task difficulty adjustment; For comment collection The number of comments; Cluster analysis algorithms are used to classify employees into different ability levels based on their past performance and skill mastery. The similarity between employees is calculated and the corresponding clustering algorithms are used to achieve classification. For each type of employee, the Bayesian network model is used to analyze the probability relationship between tasks of different difficulty levels and employees' positive work efficiency. The network structure and conditional probability are learned based on historical data to provide a reference for task allocation. The specific strategy for real-time adjustment of task difficulty is as follows: Based on the employee performance scores and task completion status obtained from the performance evaluation module, combined with the task difficulty evaluation model and the employee capability adaptation analysis results, the linear programming algorithm is used to calculate the task difficulty adjustment range suitable for the employee's current stage with the objective function and satisfy the corresponding constraints, and the task allocation is adjusted accordingly; When employees are performing tasks, the real-time emotional state of the employees obtained through wearable devices and workshop video monitoring, as well as the work efficiency tracking curve drawn using time series analysis, are combined to judge the effect of task difficulty adjustment. If it does not meet expectations, the task difficulty is adjusted based on the iterative update mechanism of the gradient descent algorithm, and relevant information is promptly fed back and updated.

2. A mine workshop performance assessment management method according to claim 1, characterized in that: In the data collection and preprocessing, data collection includes: Sensors are installed on key equipment in the mine workshop, and the Kalman filter algorithm is used to fuse sensor data to obtain basic data related to equipment operation and production. At the same time, natural language processing (NLP) technology is used to analyze employee work record texts, and the key information in the texts is quantified and added to the work data. Based on the linear regression algorithm, the weight of employee self-evaluation in the overall data collection is determined according to the consistency between the employee's past self-evaluation and actual performance, and the employee work data is obtained. The employee work data is then integrated in a unified format and transmitted to the data preprocessing server with the help of big data message queue technology; among them, a linear regression model is constructed ,in, is the actual performance data set, A collection of employee self-evaluation data. is the intercept, is the regression coefficient, is the error term, and the least squares method is used to estimate and To determine the weight of employee self-assessment; Data preprocessing includes: On the data preprocessing server, the multiple imputation method MICE is used to fill in the data with missing values. Multiple prediction models are established based on the data association relationship to ensure data integrity. The robust standardization RobustScaler method is used to standardize the data. The calculation formula is: ,in, is the original data, is the standardized data, Representation data the median of Representation data The interquartile range of the preprocessed data is stored in a distributed database.

3. A mine workshop performance assessment management method according to claim 1, characterized in that: In the dynamic performance evaluation, the multi-factor weighted model construction and optimization process includes: Determine the task type, difficulty, employee historical performance, teamwork effect, and innovation results as the key factors for performance evaluation. Let these factors be , based on expert experience and historical data analysis, the weights of each factor are initially set ,and , represents the total number of key factors for performance evaluation, It is an index variable used to traverse all key factors and build a multi-factor weighted model; performance score The calculation formula is: ； Then, using the neural network algorithm, the fluctuation of employee performance and the changes in the focus of the workshop business are taken as inputs. According to the conventional training and weight adjustment mechanism of the neural network, the weights of each factor in the weighted model are adjusted in real time through back propagation. At the same time, the XGBoost algorithm is combined with reinforcement learning to build a prediction model, taking the employee's work task as an "action" and the performance score as a "reward". Based on the exploration learning mechanism of reinforcement learning and combined with XGBoost's ability to process historical performance data, the prediction model is optimized to predict the future performance of employees. The implementation method of performance score calculation and output is as follows: input the pre-processed employee work data into the optimized multi-factor weighted model, combine it with the prediction model results, calculate the employee performance score in real time according to the established calculation logic, and output it to the subsequent modules through the API interface to provide data support for related operations.

4. A mine workshop performance assessment management method according to claim 1, characterized in that: In the incentive mechanism, the collaborative filtering algorithm includes: Assume the user set is , the item set is , the user's rating matrix for items is , Indicates user For items Ratings of users, if the user has not rated, it is recorded as a missing value; based on the user's rating matrix , calculate the similarity between users, using the Pearson correlation coefficient method, the calculation formula is ,in, and are different users, Is a user and A collection of items that have been rated. and The users are and Average rating of Indicates user and The similarity between Indicates user For items Ratings; Find target users Most similar Users form a neighbor set , predict the target user by weighted average For unrated items Rating , use collaborative filtering algorithm to analyze employee-related personalized characteristics, refer to the past performance improvement of similar employees, customize personalized points acquisition rules and distribution plans for employees according to the above algorithm, and employees will obtain points according to their performance after completing the task, which will be stored in the points account database; Adopt a simulation algorithm based on market supply and demand theory to dynamically adjust the corresponding points value according to the popularity of different rewards among employees and their redemption popularity; Dynamic adjustments to reward standards include: Use the cost-benefit analysis model in combination with the dynamic programming algorithm, optimize the reward standards in real time according to the corresponding algorithm logic based on the company's business objectives and financial status, and use the statistical methods of difference analysis and elasticity analysis to analyze the sensitivity between performance scores and changes in reward standards. Adjust the reward standards corresponding to different performance ranges. Employees choose to redeem rewards on the reward redemption platform based on points and standards. The incentive mechanism module will feed back the redemption records to other related modules for subsequent optimization analysis.

5. A mine workshop performance assessment management method according to claim 1, characterized in that: In the performance feedback and optimization, management strategy adjustment and optimization include: Use time series prediction algorithms to predict trends in team and individual performance data, and combine causal analysis algorithms to explore potential causes of performance changes. When the probability of performance decline or potential problems is predicted, the early warning mechanism is triggered according to the set threshold and the early warning information is pushed to the workshop management personnel; Based on the reinforcement learning algorithm, by simulating the performance changes after the implementation of different management strategies, the optimal management strategy adjustment plan is recommended to managers according to the reinforcement learning strategy recommendation mechanism. Managers make corresponding adjustments in the workshop management system accordingly, and the adjusted performance change data is fed back to the system for further optimization of the parameters and models of each algorithm module.

Images