Bid evaluation expert extraction system based on RPA technology and black box design

Abstract

The invention provides a bid evaluation expert extraction system based on an RPA technology and a black box design. The system comprises a black box module, an expert extraction module, a dynamic parameter adjustment module and an automatic settlement function module. The black box module ensures data security and operation compliance through data isolation and user authority management. And the expert extraction module automatically executes expert extraction, parameter configuration and notification operation by utilizing an RPA technology. The dynamic parameter adjustment module performs intelligent prediction and analog simulation based on a machine learning algorithm, adjusts an extraction strategy in real time, and improves the extraction efficiency and accuracy. And the automatic settlement function module automatically adjusts a scheme state according to expert feedback and an extraction result, and optimizes a management process. Through cooperative work of the modules, the system realizes full automation and intellectualization of the expert extraction process, and ensures fairness and efficiency of operation.

Description

Technical Field

[0001] The present invention relates to a bid evaluation expert extraction system, and in particular to a bid evaluation expert extraction system based on RPA technology and black box design. Background Art

[0002] In the current bidding and procurement process, the selection and management of bid evaluation experts is an important part of ensuring the fairness and transparency of the bid evaluation process. Especially in the bidding activities of large-scale projects, the effective management of the selection process of bid evaluation experts is directly related to the quality of the project and the fairness of the bidding results. However, the existing process of selecting bid evaluation experts still faces a series of challenges that need to be solved in actual operation. These challenges include efficiency issues in the operation process, as well as compliance and data security issues.

[0003] The existing process of selecting bid evaluation experts usually includes six main links: preparation of the bid evaluation expert selection plan, review of the selection plan, expert selection, expert notification, expert feedback and confirmation of the expert list. Each link has specific operating procedures and management responsibilities. In the preparation of the bid evaluation expert selection plan, the bidding specialist of the agency prepares an expert demand plan based on the procurement package they are responsible for. The plan needs to clarify the key parameters such as the professional field, quantity, and selection conditions of the required experts, and adjust them according to the specific needs of the project. After completion, the plan needs to be submitted to the tenderer for review. Next, the tenderer reviews the submitted expert selection plan to ensure that the design of the plan meets the project needs and meets the requirements of relevant laws and regulations and management rules. After the review is passed, the plan enters the next step of expert selection and notification.

[0004] In the expert selection phase, the selection and notification personnel will start the expert selection according to the approved plan. The core of this process is to select qualified experts from the expert pool by random selection according to pre-set rules and conditions to ensure the fairness and unpredictability of the selection process. After the expert is successfully selected, a notification must be sent to the selected expert in a timely manner. The content of the notification usually includes the time, place and other relevant arrangements for the bid evaluation. The notification is required to be comprehensive and timely to ensure that all selected experts can respond within the specified time whether they will participate in the bid evaluation. The efficient execution of this link is crucial to the smooth progress of subsequent work.

[0005] After receiving the notification, the experts need to provide feedback within the specified time on whether they can participate in the bid evaluation. The collection of feedback is a key link in the expert selection process, because the expert's response directly affects the arrangement of the bid evaluation. If some experts are unable to attend, a supplementary plan must be formulated in a timely manner and the selection must be repeated. When all the experts' feedback is collected, the final list of experts attending will be generated. This list needs to be provided to the bidding and procurement evaluation committee and conference staff at the same time, so that the corresponding bid evaluation arrangements and preparations can be made.

[0006] Although the existing process of selecting bid evaluation experts has been standardized and proceduralized to a certain extent, there are still many deficiencies and challenges in actual operation. The first is the problem of high-intensity manual monitoring and management. According to the requirements of the expert management rules, the process of expert selection and notification usually takes three working days, and sometimes even extends to the weekend. This requires the extraction and notification personnel to conduct manual monitoring throughout the process to ensure that the extraction and notification process of each expert can be carried out in a timely and accurate manner. Especially when the expert feedback is not timely or fails to respond as required, the extraction and notification personnel need to continue to intervene manually to ensure the continuity and effectiveness of the work. This way of working is not only inefficient, but also occupies a lot of human resources. Especially in large-scale bidding activities, there are many experts and complex plans, which are prone to omissions and errors, directly affecting the quality and efficiency of bid evaluation.

[0007] The second is compliance risk caused by human factors. In the existing process of selecting bid evaluation experts, the operations of each link mostly rely on manual judgment and intervention. Especially when the expert feedback is not timely or additional plans are needed, the extraction and notification personnel may need to manually mark the attendance status of the experts. The subjectivity of this operation and the inevitability of human intervention make the management method prone to irregular operations and even the risk of violating relevant regulations. Especially in some high-value or sensitive projects, the compliance risk of this manual operation is more significant, which may have a serious impact on the fairness and transparency of the project.

[0008] Finally, there is the issue of data confidentiality and security. The existing system has insufficient data security in the expert extraction and notification process. Since the extraction personnel can directly access and operate the data related to the expert database, there is a risk of data leakage and abuse when the expert database information is in hand. In addition, during the expert notification process, the manifestation of relevant data at the front end also increases the possibility of data leakage. Especially when the expert database information involves a large amount of personal privacy and sensitive information, how to ensure the security and confidentiality of the extraction process and prevent data leakage has become an important issue that the current system needs to solve urgently.

[0009] In order to address the above-mentioned shortcomings and challenges, the existing bid evaluation expert extraction system needs technical improvements. Summary of the invention

[0010] In order to solve the above problems in the prior art, the present invention proposes a bid evaluation expert extraction system based on RPA technology and black box design, wherein the expert extraction system comprises:

[0011] Black-box module, used to perform black-box processing of the expert extraction process;

[0012] The expert extraction module automatically performs expert extraction operations based on RPA technology;

[0013] Dynamic parameter adjustment module, which performs intelligent prediction and simulation based on machine learning algorithms and historical data analysis, and adjusts the extraction strategy in real time;

[0014] The automatic settlement function module automatically adjusts the expert status and extraction plan status after the expert notification ends.

[0015] The black box module includes:

[0016] An independent data user setting module, wherein the independent data user setting module sets up one or more independent data users, each of which has specific access rights and operation rights, and the data user operates under an authentication and authorization mechanism;

[0017] A data isolation module, wherein the data isolation module isolates data through logical isolation and hardware isolation strategies. The logical isolation isolates data through a virtual data view, and the hardware isolation isolates data through a dedicated storage device or virtual storage. Data transmission is performed through an encrypted channel. The virtual data view extracts necessary information related to expert extraction from the original database and displays it as an independent data set.

[0018] A data operation normalization module defines a standardized data operation process and ensures the traceability and security of data operations through log recording and anomaly detection.

[0019] The expert extraction module comprises:

[0020] The parameter configuration function submodule is used to extract and manage various parameters in the expert extraction process and store the adjusted parameters;

[0021] A random drawing order generation submodule is used to set the drawing conditions through the logic analysis module and generate an expert random drawing order using a pseudo-random number generation algorithm, wherein the expert random drawing order meets regulatory requirements;

[0022] The extraction plan preparation submodule is used to prepare the extraction plan according to the preset operation parameters, optimize the plan and publish it to the task scheduling queue for execution;

[0023] The supplementation plan optimization submodule is used to automatically generate and adjust the supplementation strategy when the expert is absent or refuses;

[0024] The scheduled task module is used to automatically perform expert extraction and notification operations according to the set time and conditions, and record the operation steps and results.

[0025] The expert extraction module further includes multiple RPA robots, which are configured in different computing environments and are responsible for executing independent subtasks in the expert extraction process. The multiple RPA robots have automation scripts that control the RPA robots to perform specific tasks and contain operation steps and conditional logic; the multiple RPA robots monitor the progress of the expert extraction process and automatically adjust the extraction and notification strategies based on feedback; and combine machine learning models to analyze and predict expert behavior, adjust parameter settings and extraction strategies; automatically assign tasks according to task priority and monitor execution status; generate operation logs and perform compliance checks.

[0026] The dynamic parameter adjustment module comprises:

[0027] The data collection and preprocessing submodule is used to collect multi-source data related to expert extraction, and clean, normalize, extract and select features of the data to generate high-quality data suitable for machine learning model training;

[0028] The machine learning model training and optimization submodule is used to train the machine learning model using preprocessed data, predict key parameters in the expert extraction process, and improve prediction accuracy through model optimization and updating;

[0029] The prediction and simulation submodule is used to perform intelligent prediction and simulation of expert extraction strategies based on the trained machine learning model, evaluate the possible results under different extraction strategies, and generate the optimal extraction plan;

[0030] The real-time strategy adjustment submodule is used to combine the prediction and simulation results to adjust the actual expert extraction and notification strategies in real time to cope with possible changes in the extraction process.

[0031] The machine learning model training and optimization submodule includes:

[0032] Model selection and initialization unit, used to select appropriate machine learning models based on data characteristics and task requirements;

[0033] A model training unit is used to train the selected model using the training set data and minimize the prediction error by iteratively optimizing the model parameters;

[0034] Model validation and cross-validation unit, used to evaluate the performance of the model through cross-validation technology and select the optimal model;

[0035] The model optimization and iterative update unit is used to regularly update and optimize the model, including retraining the model, adjusting hyperparameters, and using new data for incremental learning.

[0036] The prediction and simulation submodule includes:

[0037] The intelligent prediction unit is used to call the trained machine learning model to predict the key parameters in the expert extraction process. The output of the prediction model is a multi-output function:

[0038]

[0039] Among them, Y1 is the predicted value of the expert response rate, Y2 is the predicted value of the absence rate, and Y3 is the predicted value of the optimal number of extractions; functions f1 and f2 use logistic regression models with probability outputs between 0 and 1; f3 is a linear regression model used to predict the optimal number of extractions; parameter α ij , β i are the coefficients of the model, i = 1, 2, j = 1...n, n is the total number of all influencing factors, γ and ∈ are the intercept term and error term respectively; X1, X2,..., X n Various factors that affect the expert extraction results.

[0040] The prediction and simulation submodule includes:

[0041] The simulation unit is used to perform simulation based on the prediction results. The Monte Carlo method is used to perform multiple random samplings on the objective function Z to evaluate the effect of each extraction strategy. The objective function Z is expressed as:

[0042]

[0043] Among them, Y 1,i is the expert response rate prediction value under the i-th strategy, T i is the expert’s response time, Y 2,i is the predicted value of absence rate, Y 3,i is the predicted value of the optimal number of extractions, i=1…m, m is the total number of different extraction strategies evaluated during the simulation, and w1, w2, w3 are weight coefficients.

[0044] The real-time strategy adjustment submodule includes:

[0045] Feedback monitoring and adjustment triggering unit, which is used to monitor the actual feedback during the expert extraction and notification process, and automatically trigger the strategy adjustment process when the feedback rate deviates from the predicted value;

[0046] A dynamic adjustment strategy generation unit is used to generate a new extraction strategy according to real-time feedback, and the strategy is generated by optimizing the rule engine and dynamic optimization algorithm;

[0047] The strategy execution and continuous optimization unit is used to execute the adjusted extraction strategy through RPA technology and record all operation steps in the operation log for subsequent analysis and optimization.

[0048] The automatic settlement function module includes:

[0049] The non-responding expert status adjustment submodule is used to automatically update the status of non-responding experts to "not present" after each round of expert notification;

[0050] The notification rule adjustment submodule is used to automatically adjust the status of experts not notified in this round according to the preset notification rules;

[0051] The scheme status management submodule is used to dynamically adjust the scheme status according to expert feedback and extraction results. The status update is based on the number of shortfalls and notifications, and is executed according to the preset logic;

[0052] The notification trigger management submodule is used to automatically generate and send notifications under specific conditions, and update the solution status to ensure that the status information is synchronized with the notification content.

[0053] The beneficial effects of the present invention are:

[0054] Through automated RPA technology and machine learning models, the system can automatically perform expert extraction and notification operations, reduce manual operations, and improve the efficiency and accuracy of the extraction process.

[0055] The dynamic parameter adjustment module uses historical data and real-time feedback to perform intelligent prediction and simulation through machine learning algorithms. It can adjust the extraction strategy in real time, adapt to various changes, and improve the success rate and response speed of expert extraction.

[0056] The black box module adopts data isolation and authentication mechanisms to ensure data security and operational compliance; the data operation standardization module ensures the traceability and security of data operations.

[0057] The automatic settlement function module automatically adjusts the solution status and expert attendance according to expert feedback and extraction results to ensure a smooth process, while reducing human intervention and optimizing management efficiency.

[0058] Through simulation and multi-objective optimization, the system can evaluate the effects of different extraction strategies, provide decision support, and ensure the fairness and optimization of the extraction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application, but do not constitute an improper limitation of the present invention. In the drawings:

[0060] Figure 1 The overall architecture diagram of the bid evaluation expert extraction system of the present invention is shown;

[0061] Figure 2 A flow chart of the present invention is shown. DETAILED DESCRIPTION

[0062] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, wherein the illustrative embodiments and descriptions are only used to explain the present invention but are not intended to limit the present invention.

[0063] In order to better understand the present invention, specific implementation modes are described in detail below with reference to the accompanying drawings.

[0064] Example 1: Overall architecture of the bid evaluation expert extraction system

[0065] like Figure 1 As shown, this embodiment provides an overall architecture of a bid evaluation expert extraction system based on RPA (Robotic Process Automation) and black box design. The system aims to optimize the extraction and management process of bid evaluation experts through automation and intelligence to ensure the compliance of operations and the security of data. The system mainly includes the following modules:

[0066] 1. Blackbox module 101: This module is used to perform blackbox processing of the expert extraction process. By setting up independent data users, the blackbox module standardizes the extraction operation from the business operation level and isolates the extraction process data from the system level to ensure that the data operation does not directly interact with the original database to achieve data security and operational compliance.

[0067] 2. Expert extraction module 102 based on RPA technology: This module is responsible for applying RPA technology in the expert extraction process to improve extraction efficiency and operation accuracy. This module includes the following submodules:

[0068] Parameter configuration function submodule 102a: is used to extract parameters such as expert notification ratio and extraction times in the original process, and establish a unified parameter configuration function, allowing users to flexibly set these parameters according to specific needs, so as to reduce repeated operations and improve system efficiency.

[0069] Random selection order generation submodule 102b: through automatic system recognition and logic analysis, automatically generate expert random selection order that meets regulatory requirements to ensure the fairness and compliance of the selection process.

[0070] Extraction plan preparation submodule 102c: In the expert extraction plan preparation link, parameters such as notification time, response time and reporting time are pre-set and limited to the prescribed working hours (8:30-17:30) to ensure the timeliness and humanity of the notification.

[0071] Supplementary plan optimization submodule 102d: optimizes the extraction process of supplementary plans through automated processes, transforms it into a one-by-one control of supplementary plan operations, reduces human intervention, and improves the response speed and accuracy of the system.

[0072] Scheduled task module 102e: automatically executes expert extraction and notification operations according to set time and conditions to ensure the timeliness and automation of tasks.

[0073] 3. Dynamic parameter adjustment module 103: This module improves the intelligence level of the system. Based on machine learning algorithms and historical data analysis, this module performs intelligent prediction and simulation before expert extraction. The specific functions are as follows:

[0074] Data collection and preprocessing submodule 103a: used to collect and preprocess various data related to expert extraction, including historical expert extraction records, real-time project data, expert personal data and environmental data. The preprocessed data will be used for subsequent model training and prediction.

[0075] Machine learning model training and optimization submodule 103b: Use the preprocessed data to train the machine learning model, and through model optimization and iterative updating, achieve accurate prediction of key parameters in the expert extraction process, such as the expert's response rate, absence rate, and optimal number of extractions.

[0076] Prediction and simulation submodule 103c: Based on the trained model, the system performs intelligent prediction and simulation before each expert extraction, predicts the possible results under different extraction strategies, and dynamically adjusts the extraction strategy.

[0077] Real-time strategy adjustment submodule 103d: Combines the prediction and simulation results to adjust the actual expert extraction and notification strategy in real time to cope with various changes that may occur during the extraction process, such as the actual feedback rate of the expert deviating from the predicted value.

[0078] 4. Automatic settlement function module 104: After each round of expert notification, the system settlement function is automatically executed to adjust the solution status and expert attendance. Specifically, it includes:

[0079] Adjustment of the status of experts who have not responded: Change the status of experts who have not responded to the meeting to “Not Attended”.

[0080] Notification rule adjustment: Adjust the status of experts not notified in this round according to the notification rules.

[0081] Plan status management: Dynamically adjust the plan status based on expert feedback and drawing results. For example, when the number of difference is 0, it will be changed to "pending confirmation"; when the number of difference is greater than 0 and there are no unnotified people, it will be changed to "pending selection", etc.

[0082] Notification trigger management: Under certain conditions, such as when the number of people with a difference is greater than 0 but the number of people who have not been notified is 0, and the number of people who can be drawn is 0, the status of the change plan will be "pending confirmation" and a notification will be triggered.

[0083] Through the collaborative work of the above modules, the system realizes the full automation and intelligent operation of the expert extraction process, reduces human intervention, improves the system's work efficiency and security, and can adapt to complex business needs and dynamic changes.

[0084] Example 2 Black Box Module

[0085] The black box module 101 is designed to achieve secure data processing and operational compliance control through independent data user settings and data isolation policies. The design of the black box module ensures that data operations do not directly interact with the original database, thereby effectively preventing data leakage and unauthorized access.

[0086] 1. Independent data user settings:

[0087] The black box module 101 controls the data operations in the expert extraction process by setting up independent data users. The specific implementation steps are as follows:

[0088] During the system initialization phase, the black box module creates one or more independent data users, which have strictly restricted access rights and can only access data related to the expert extraction process. These users are isolated in a dedicated subsystem that is independent of the main database but can exchange data through a controlled interface.

[0089] Assign specific access rights and operation rights to each data user. The permission settings include data reading, data writing and data updating rights, but do not include the rights to delete and modify the master database data. The permission configuration is managed through a set of policy files, which define the operation boundaries of each data user to ensure that data users can only operate within their authority.

[0090] Before a data user performs any data operation, the system first verifies the user's identity through the authentication and authorization mechanism. The authentication mechanism includes multiple methods such as username, password and two-factor authentication (2FA), and the authorization mechanism is based on a predefined policy rule file. Only when both authentication and authorization are passed can the data user perform the corresponding operation.

[0091] 2. Data isolation strategy:

[0092] To ensure data security and prevent unauthorized data access, the black box module 101 implements a data isolation strategy.

[0093] The core of the data isolation strategy lies in logical isolation. Logical isolation is achieved by establishing virtual data views, which are a subset of data extracted from the original database and only include the necessary information related to expert extraction. These virtual data views are displayed as independent data sets, but in fact do not contain all the data in the original database. In this way, the black box module can provide the necessary data access without exposing sensitive information.

[0094] At the hardware level, data isolation is achieved through dedicated physical storage devices or virtual private storage (VPS). The black box module stores all data related to the expert extraction process in isolated physical or virtual storage space, which is independent of the physical storage of the main database, and data transmission is only carried out through controlled interfaces. This physical isolation further enhances data security and prevents unauthorized data from being obtained through hardware-level attacks.

[0095] During the data exchange process between the black-box module and the main database, all data transmission must go through an encrypted channel (such as SSL / TLS) to prevent data from being eavesdropped or tampered with during transmission. The black-box module uses a dedicated data transfer gateway to monitor and control the data transmission process to ensure that all data transmission complies with the predefined security protocol.

[0096] 3. Standardization of data operations:

[0097] The black box module 101 standardizes the relevant data operations extracted by experts from the business operation level.

[0098] The black box module defines a standardized data operation process, including the steps and sequence of data reading, writing and updating. All operations must strictly follow these standardized processes. Any violation of the operation process will trigger an automatic alarm mechanism and be recorded in the security log.

[0099] To ensure the traceability of operations, the black box module logs all data operations. The log records include operation time, operation user, operation type (read, write, update), data change details, etc. The operation log is stored in an immutable log database to ensure that any data operation can be audited and traced back.

[0100] The black box module includes an anomaly detection engine, which uses machine learning algorithms to monitor data operation patterns in real time and detect any abnormal behavior (such as abnormal data access frequency, illegal data access paths, etc.). Once an anomaly is detected, the system will respond immediately, including blocking operations, notifying security administrators, and triggering emergency measures.

[0101] Through the above detailed implementation scheme, the black box module 101 can effectively realize the security management and operational compliance control of the data in the bidding expert extraction process.

[0102] Example 3 Expert extraction module 102 based on RPA technology

[0103] This embodiment provides an expert extraction module 102 based on RPA (Robotic Process Automation) technology, which improves the efficiency and operation accuracy of the expert extraction process through automation technology. The expert extraction module based on RPA technology includes five submodules, each of which has specific functions and implementation steps.

[0104] 1. Parameter configuration function submodule 102a:

[0105] The parameter configuration function submodule is used to extract and manage various parameters in the expert extraction process.

[0106] The system first extracts parameters such as expert notification ratio, extraction times, expert type, and regional distribution from the original process. The parameter extraction process can be achieved through predefined parameter extraction rules, which define how to extract the required parameters from historical data and project requirements.

[0107] The extracted parameters will be displayed and managed in a unified parameter configuration interface. Users can adjust the parameters through the graphical user interface (GUI) or command line interface (CLI) according to specific needs. This submodule provides a parameter checking mechanism to ensure that all set parameters are logically reasonable and comply with business rules.

[0108] Parameter storage and application: The adjusted parameters will be stored in a dedicated parameter database and called up at different stages of expert extraction. The parameter database uses a version control mechanism to manage the historical versions and change records of the parameters to ensure the traceability of the parameters and the recoverability of the versions.

[0109] 2. Randomly extract sequence generation submodule 102b:

[0110] The random drawing sequence generation submodule generates an expert random drawing sequence that meets regulatory requirements through automatic system identification and logical analysis.

[0111] The logic analysis module analyzes the input project requirements, expert database information and regulatory requirements, and sets extraction conditions, including the expert's professional field, participation limit, and geographical distribution requirements.

[0112] A pseudo-random number generation algorithm (such as the Mersenne twister algorithm or the linear congruential generator) is used to generate a random expert extraction order. To ensure the fairness and unpredictability of the extraction, the system can combine multiple randomization algorithms and enhance randomness through entropy sources.

[0113] The generated random drawing sequence will be checked by the legality verification module to ensure that the drawing sequence complies with regulatory requirements and business rules. If an illegal sequence is detected, the system will regenerate it until all conditions are met.

[0114] 3. Extraction plan preparation submodule 102c:

[0115] The extraction plan preparation submodule pre-sets various operating parameters in the expert extraction plan preparation stage to ensure the timeliness and humanity of the notification.

[0116] After receiving the extraction task, the system automatically initializes the extraction plan and calls the parameters set in the parameter configuration function submodule 102a.

[0117] According to project requirements and regulations, the system presets key parameters such as notification time, response time and reporting time. Time parameters must be limited to the prescribed working hours (8:30-17:30). To ensure the rationality of parameters, the system uses a time conflict detection algorithm to avoid time overlap or unreasonable arrangements.

[0118] The system optimizes the initially set extraction plan, taking into account various constraints and business needs (such as the actual availability of experts, priority of urgent projects, etc.). The optimized plan will be published to the task scheduling queue and wait for execution.

[0119] 4. Supplementary solution optimization submodule 102d:

[0120] The supplementary plan optimization submodule optimizes the process of extracting supplementary plans through automated processes, reduces human intervention, and improves response speed and accuracy.

[0121] The system monitors the feedback of experts and automatically triggers the supplementary plan when it detects that the expert is absent or refuses to participate. The supplementary conditions are detected in real time by the feedback monitoring module.

[0122] The supplementation plan optimization submodule generates a supplementation strategy based on the current absence of experts and the urgency of the project. The strategy generation is based on a multi-objective optimization algorithm to balance the number of experts, time requirements and project priorities.

[0123] The system executes the supplementation extraction process according to the generated supplementation strategy. During the supplementation execution process, the system continuously monitors the feedback from experts and adjusts the supplementation strategy in real time to ensure that the supplementation operation achieves the expected goals.

[0124] 5. Scheduled task module 102e:

[0125] The scheduled task module is responsible for automatically executing expert extraction and notification operations according to the set time and conditions.

[0126] The user sets the time and conditions of the expert extraction and notification operations through the parameter configuration function submodule 102a, and the system stores the set task information in the task scheduling database.

[0127] The system has built-in timing triggers, which automatically start the corresponding expert extraction and notification operations according to the time settings and conditions in the task scheduling database.

[0128] During the task execution process, the system automatically executes all preset operations through RPA technology, including expert extraction, notification sending, etc. All operation steps and results will be recorded in the task log file to ensure that the execution of each task can be traced and audited.

[0129] Through the above implementation scheme, the expert extraction module 102 based on RPA technology can efficiently and automatically perform expert extraction and management tasks, significantly reduce manual operations, improve the accuracy and efficiency of operations, and ensure the fairness and compliance of the entire expert extraction process.

[0130] 6. Specific applications of RPA technology

[0131] The expert extraction module 102 uses RPA (Robotic Process Automation) technology to achieve automation and intelligent operation of the expert extraction process. This module automatically performs various tasks by simulating human operation behavior on the computer, reducing human intervention, thereby improving work efficiency and operation accuracy. In order to effectively perform these tasks, the module is configured with multiple RPA robots that can simulate human operations in different computing environments and complete their respective tasks.

[0132] During the implementation process, the expert extraction process is first decomposed into several independent subtasks, such as parameter extraction, random extraction, notification sending, etc. One or more RPA robots are configured for each subtask. These robots perform specific tasks by using written automation scripts, such as Python, UiPath, or Automation Anywhere. Each script contains clear operation steps and conditional logic to ensure the accuracy and consistency of the operation. The configuration files of all robots are centrally managed in a robot configuration library, which supports version control and can track and manage the configuration history and update records of each robot.

[0133] The core of RPA technology is automated operation execution. The expert extraction module 102 uses RPA robots to perform key operations. For example, the RPA robot reads parameters such as the expert notification ratio and the number of extractions in the original process, and automatically fills them into the parameter configuration interface of the system. By simulating mouse clicks and keyboard input, the robot can accurately and quickly perform parameter extraction and setting, avoiding errors and inefficiencies that may be caused by manual operations. In the random extraction sequence generation submodule, the RPA robot automatically generates the expert extraction sequence according to the set randomization algorithm. The robot generates a random sequence by calling the system's built-in random number generator function and stores the results in the extraction database. In addition, during the extraction plan preparation and scheduled task execution process, the RPA robot automatically generates expert notification content and sends notifications through an email system, SMS platform or other communication tools. The robot can automatically detect the contact information of each expert, fill in the notification template, send notifications, and monitor the sending status and feedback.

[0134] When performing automated tasks, the RPA robot can also monitor the progress of the entire expert extraction process in real time and make adjustments based on preset rules. By continuously monitoring the feedback from experts, such as the time and content of responses, the robot updates this information to the feedback monitoring module in real time, triggering the start conditions of the supplementary solution optimization submodule. Based on the real-time monitoring results, the robot automatically adjusts the extraction and notification strategies. For example, if some experts fail to respond within the specified time, the robot will automatically execute the supplementary solution, re-extract and notify other experts. The robot can also adjust the frequency and content of notifications based on feedback to ensure that all operations comply with preset business rules and regulatory requirements.

[0135] In order to further improve the level of automation, the expert extraction module 102 combines intelligent algorithms and implements adaptive learning and optimization through RPA technology. The RPA robot can call pre-trained machine learning models to analyze and predict the behavior patterns of experts. For example, the system can predict which experts are more likely to reply within a specific time period and adjust the extraction order and notification strategy. Through integration with the machine learning model, the RPA robot continuously collects and learns new expert feedback data and automatically adjusts parameter settings and extraction strategies. This adaptive learning capability enables the system to continuously optimize itself over time and improve the accuracy and efficiency of expert extraction.

[0136] In terms of task management, the expert extraction module 102 uses the task scheduling function in the RPA technology to achieve the management and execution of automated tasks. All automated tasks enter the task queue according to the preset priority. The task scheduling module automatically assigns the tasks to idle RPA robots for execution based on the urgency and priority order of the tasks. During the execution of the tasks, the RPA robot monitors the execution status of each step in real time. If an error occurs, such as a network failure or data loss, the robot can automatically record the error information, try to re-execute the task, or perform a preset error handling operation based on the error type.

[0137] In addition, during the expert extraction process, RPA technology also provides security and compliance management functions. When performing each operation, the RPA robot automatically generates a detailed operation log, including the operation time, operation content, and operation results. These logs are stored in the security log database for subsequent auditing and analysis. Before performing an operation, the robot automatically performs a compliance check to ensure that each operation complies with the preset business rules and regulatory requirements. If an illegal operation is found, the robot will automatically stop the task and notify the administrator to handle it.

[0138] Through the above implementation scheme, the expert extraction module 102 fully utilizes the advantages of RPA technology to realize the full automation and intelligent operation of the expert extraction process, which not only improves work efficiency and operation accuracy, but also ensures the fairness and compliance of the expert extraction process.

[0139] Embodiment 3 Dynamic parameter adjustment module 103

[0140] The dynamic parameter adjustment module 103 uses machine learning algorithms and historical data analysis to perform intelligent prediction and simulation before expert extraction, aiming to improve the accuracy and efficiency of expert extraction. The dynamic parameter adjustment module 103 includes four submodules: a data collection and preprocessing submodule 103a, a machine learning model training and optimization submodule 103b, a prediction and simulation submodule 103c, and a real-time strategy adjustment submodule 103d.

[0141] 1. Data collection and preprocessing submodule 103a:

[0142] The data collection and preprocessing submodule is responsible for collecting and processing various data related to expert extraction to ensure that the data input into the machine learning model is high-quality and structured. The implementation steps are as follows:

[0143] Data Collection: The system continuously collects multi-source data related to the expert extraction process, including but not limited to:

[0144] Historical expert extraction records: such as past extraction success rate, expert response time, absence rate, etc.

[0145] Real-time project data: such as the scale, type, geographical distribution and other characteristics of the current project.

[0146] Expert personal data: including the expert’s area of ​​expertise, engagement history, current availability, geographic location, etc.

[0147] Environmental data: includes external factors that may affect expert participation, such as changes in policies and regulations, natural disasters, etc.

[0148] Data cleaning and normalization: The collected data may contain noise, missing values, or inconsistencies. The system first cleans the data to remove outliers and invalid data. Then, the system normalizes different types of data and converts the data into a unified scale and format to facilitate subsequent model training and analysis. Normalization methods include Min-Max Scaling, Z-score normalization, etc.

[0149] Feature extraction and selection: In order to improve the predictive power of the model, the system extracts key features from the raw data through feature extraction algorithms (such as principal component analysis PCA and linear discriminant analysis LDA). These features include the distribution of expert response time, the impact of project type on extraction results, etc. The system can also select the most representative features based on correlation analysis and feature importance evaluation (such as information gain and Gini coefficient) to reduce the complexity of the model.

[0150] Data storage and management: The processed data will be stored in a dedicated data warehouse. The data warehouse uses efficient indexing and query mechanisms to support multi-dimensional data analysis and fast retrieval, ensuring the availability of data during model training and prediction.

[0151] 2. Machine learning model training and optimization submodule 103b:

[0152] The machine learning model training and optimization submodule uses preprocessed data to train the model and continuously optimize it to achieve accurate prediction of the key parameters extracted by experts. The implementation steps are as follows:

[0153] Model selection and initialization: Select an appropriate machine learning model based on the characteristics of the data and task requirements. Common models include decision trees, random forests, support vector machines (SVMs), neural networks, etc. For complex prediction tasks, consider using ensemble learning methods such as random forests or gradient boosted decision trees (GBDT) to improve the generalization ability of the model.

[0154] Model training: The system divides the preprocessed data into a training set and a validation set, and uses the training set data to train the selected model. During the training process, the system iteratively optimizes the model parameters (such as learning rate, regularization coefficient, etc.) to minimize the prediction error. The training algorithm can use optimization methods such as gradient descent and stochastic gradient descent.

[0155] Model validation and cross-validation: To avoid model overfitting, the system uses cross-validation technology to evaluate the performance of the model. Specifically, the system divides the data into multiple subsets and performs training and validation separately to obtain a more robust performance evaluation. By comparing the cross-validation scores (such as accuracy, AUC values) of different models, the optimal model is selected.

[0156] Model optimization and iterative updates: As data increases and business needs change, the system needs to regularly update and optimize the model. The update process includes retraining the model, adjusting hyperparameters, and using new data for incremental learning. The system can also automatically tune the model through automated machine learning tools to further improve prediction accuracy.

[0157] 3. Prediction and simulation submodule 103c:

[0158] The prediction and simulation submodule uses the trained machine learning model to perform intelligent prediction and simulation before each expert extraction to predict the possible results under different extraction strategies and dynamically adjust the extraction strategy. This module predicts multiple key parameters and uses these prediction results for simulation to optimize the expert extraction plan. The implementation steps are as follows:

[0159] a. Intelligent prediction: The system calls the trained machine learning model to predict multiple key parameters in the expert extraction process, including expert response rate (Y1), absence rate (Y2), optimal number of extractions (Y3), etc. These parameters are the core indicators that determine the effectiveness of the expert extraction strategy. The system performs multivariate predictions based on the input project characteristics and expert personal data.

[0160] Let X1, X2, …, X n To account for various factors that affect the expert extraction results (such as project size, expert experience, time, etc.), the output of the prediction model can be expressed as a multi-output function:

[0161]

[0162] Among them, Y1 is the predicted value of the expert response rate, Y2 is the predicted value of the absence rate, and Y3 is the predicted value of the optimal number of extractions. Functions f1 and f2 use logistic regression models to ensure that the probability output is between 0 and 1; f3 is a linear regression model used to predict the optimal number of extractions. Parameter α ij , β i are the coefficients of the model, i=1,2, j=1…n, n is the total number of all influencing factors, obtained through training of the machine learning algorithm, γ and ∈ are the intercept term and error term respectively.

[0163] In the above formula, n represents the number of independent variables that affect the expert extraction results, that is, the total number of influencing factors involved. For example:

[0164] If three influencing factors such as project scale, expert experience, and notification time are considered, then n=3.

[0165] If more factors are added, such as geographical distribution, task urgency, etc., n will increase accordingly.

[0166] Therefore, n is the number of all relevant influencing factors, which are X1,X2,…,X n In the form of i Represents a specific influencing factor. These factors are weighted by the model (such as α ij , β i) reflects their different degrees of influence on the prediction results.

[0167] 2. Simulation: Based on the prediction results, the system performs simulation to evaluate the possible results under different extraction strategies. In order to optimize the expert extraction strategy, the system needs to consider multiple goals, including improving the response rate, reducing the absence rate, and reducing the number of extractions. To this end, a multi-objective optimization function Z is defined to measure the effect of the extraction strategy. This function comprehensively considers factors such as the expert's response rate, absence rate, and number of extractions:

[0168]

[0169] Among them, Y 1,i is the expert response rate prediction value under the i-th strategy, T i is the expert’s response time, Y 2,i is the predicted value of absence rate, Y 3,i is the predicted value of the optimal number of extractions, i=1…m, m is the total number of different extraction strategies evaluated during the simulation, and w1, w2, w3 are weight coefficients used to adjust the importance of different factors.

[0170] The simulation module uses the Monte Carlo method to perform multiple random sampling of the objective function Z to evaluate the effect of each extraction strategy. Through multiple simulations, the system can obtain indicators such as the average effect value, variance, and strategy success rate of different strategies. These indicators help the system identify the best strategy and make a selection.

[0171] In the above formula, m represents the total number of strategies considered during the simulation, that is, the number of different extraction strategies evaluated during the simulation.

[0172] Each strategy corresponds to a set of specific parameter configurations (such as expert notification ratio, number of extractions, notification time, etc.), and the system evaluates the effectiveness of these strategies one by one through simulation. For example:

[0173] If the system simulates 5 different extraction strategies (such as changing the notification time period, adjusting the expert grouping, etc.), then m=5.

[0174] If more policies are added (such as different supplementary rules or dynamic notification priorities, etc.), the value of m will increase.

[0175] Summation symbol in formulas It represents the weighted sum of the evaluation values ​​of all m strategies in order to compare and optimize the overall effect of the strategy. The effect of each strategy is determined by its corresponding response rate Y 1,i , absenteeism rate Y 2,i , extraction times Y 3,i , and response time T i Determined by other indicators.

[0176] 3. Scheme optimization: Based on the simulation results, the system performs multi-objective optimization to balance the efficiency, accuracy and fairness of expert extraction and generate the optimal extraction scheme. The system uses algorithms such as genetic algorithms or particle swarm optimization to further optimize the objective function Z and find the optimal solution for the parameter combination.

[0177] Through the above implementation plan, the system can effectively predict and evaluate the effects of different expert extraction strategies, ensuring that each extraction can be optimized and adjusted according to the latest prediction results and simulation analysis to maximize the overall extraction efficiency and accuracy.

[0178] 4. Real-time strategy adjustment submodule 103d:

[0179] The real-time strategy adjustment submodule combines the prediction and simulation results to adjust the actual expert extraction and notification strategies in real time to cope with various changes that may occur during the extraction process. The implementation steps are as follows:

[0180] Feedback monitoring and adjustment triggering: During the expert extraction and notification process, the system continuously monitors the actual feedback situation (such as the expert's response time and feedback content). When the feedback rate deviates from the predicted value, the system automatically triggers the adjustment process. The feedback monitoring mechanism uses real-time data stream processing technology to ensure the timely capture and processing of feedback information.

[0181] Dynamic adjustment strategy generation: The system calls the prediction and simulation submodule 103c to generate a new extraction strategy based on real-time feedback. For example, if the feedback rate of a certain round of extraction is low, the system can automatically increase the notification frequency, adjust the notification time, or start the supplementary extraction program. The system uses the rule engine and dynamic optimization algorithm to generate a new strategy and execute it immediately.

[0182] Strategy execution and continuous optimization: The adjusted strategy is automatically executed in the system through RPA technology, and all operation steps are recorded in the operation log for subsequent analysis and optimization. The system will also continue to collect feedback data after adjustment for further analysis and learning to optimize future extraction strategies.

[0183] Through the above implementation scheme, the dynamic parameter adjustment module 103 can apply intelligent prediction and simulation technology in real time during the expert extraction process, automatically adjust the extraction strategy, ensure that the system can always perform expert extraction tasks efficiently and accurately under various changing situations, and improve the overall efficiency and responsiveness of expert extraction.

[0184] Embodiment 4 Automatic settlement function module 104

[0185] The automatic settlement function module 104 is designed to automatically perform the system settlement function after each round of expert notification, and dynamically adjust the solution status and expert attendance according to the expert feedback and extraction results. This module ensures the smooth progress and efficient management of the expert extraction process through a series of automated logic and rules.

[0186] 1. Adjustment of the status of unanswered experts:

[0187] The non-responding expert status adjustment submodule is used to automatically change the status of non-responding experts to "not present" after each round of notification. The specific implementation steps are as follows:

[0188] Data collection and identification: After each round of expert notification, the system automatically collects all expert feedback information. For experts who do not respond within the specified time, the system marks them as "unanswered".

[0189] Status change rule application: Based on the preset status change rules, the system automatically updates all experts with "unresponsive" status to "not present".

[0190] Logging and auditing: After each status change operation, the system will record the list of experts, time, operator and other information of the change in the operation log to ensure data traceability and transparency of operations.

[0191] 2. Notification rule adjustment:

[0192] The notification rule adjustment submodule is used to adjust the status of experts not notified in this round according to the notification rules. The specific implementation steps are as follows:

[0193] Rule identification and application: The system first identifies the current notification rules for experts not participating in this round. These rules may include whether experts not participating in this round of notifications should remain in the "pending" status or whether a status update is required.

[0194] Status adjustment: Based on the identified rules, the system automatically adjusts the status of experts who are not notified in this round. For example, if the rules require that the status of experts who are not involved in this round of notifications should remain unchanged.

[0195] Exception handling: During the status adjustment process, if the system detects data anomalies (such as duplicate status or invalid status), the system will automatically trigger the exception handler, send a warning message to the administrator, and record the exception details in the log.

[0196] 3. Solution status management:

[0197] The solution status management submodule dynamically adjusts the solution status based on expert feedback and extraction results. The specific implementation steps are as follows:

[0198] Difference calculation and judgment: After each round of drawing, the system automatically calculates the current difference number, that is, the difference between the actual number of experts present and the target number. The calculation formula is as follows:

[0199] Number of people missing = target number - actual number of people present

[0200] Status update logic: Based on the number of people missing and notifications, the system applies the following status update logic:

[0201] a. When the number of remaining experts is equal to 0, all required experts have confirmed their participation:

[0202] -The status of the project is changed to "pending confirmation".

[0203] b. When the number of people with a shortfall is greater than 0, and the number of people who have not been notified is 0 and the number of people who can be drawn is 0, there is a shortage of experts, but there are no more experts to be notified and no experts to be drawn:

[0204] -The status of the plan is changed to "pending confirmation", and the relevant personnel are notified to add additional experts.

[0205] c. When the number of people with a shortfall is greater than 0, the number of people who have not been notified is 0, the number of people who can be drawn is greater than 0, and the number of draws is less than the maximum number of draws, there is a gap in experts, there are still experts who can be drawn, and the maximum number of draws has not been reached:

[0206] -The status of the proposal is changed to "pending selection", ready for a new round of expert selection.

[0207] d. When the number of candidates with a shortfall is greater than 0, the number of candidates who have not been notified is 0, and the number of candidates who can be drawn is greater than 0, but the number of draws is equal to the maximum number of draws, there is a shortage of experts, and there are still experts who can be drawn, but the maximum number of draws has been reached:

[0208] -The status of the plan is changed to "pending confirmation", and the relevant personnel are notified to add additional experts.

[0209] e. When the number of the difference is greater than 0 and the number of the unnotified is greater than 0, it means that there are still experts who have not been notified and there is a shortage of experts:

[0210] -The status of the plan is changed to "pending issuance" and the remaining experts will continue to be notified.

[0211] Among them, the system selects the corresponding status update path and performs corresponding database operations based on the current status and the calculated difference in the number of people.

[0212] Automatic notification and feedback: When the system status changes to "pending confirmation" and a notification needs to be triggered, the system calls the SMS platform API to send SMS notifications to relevant personnel, including the number of experts who need to be added, status description, etc.

[0213] 4. Notification trigger management:

[0214] The notification trigger management submodule automatically triggers notifications under certain conditions and updates the status. The specific implementation steps are as follows:

[0215] Condition detection and judgment: The system automatically detects the current draw status based on preset conditions (such as the number of people with a difference is greater than 0 but the number of people who have not been notified is 0, and the number of people who can be drawn is 0). If the conditions are met, the system marks that a notification needs to be triggered.

[0216] Notification generation and sending: The system automatically generates notification content based on the conditional judgment results, including the current status of the extraction plan, the next action to be taken, etc. The notification is sent to relevant personnel via a preset SMS or email platform.

[0217] Status synchronization and update: After the notification is sent, the system automatically updates the solution status to "pending confirmation" to ensure that the status information is synchronized with the notification content. Status synchronization is achieved through transactional database operations to ensure data consistency and atomicity of operations.

[0218] Log and audit function: Each notification trigger and status update operation is recorded in the log database, including operation time, trigger conditions, notification content, etc. The system provides real-time audit function, and administrators can trace the details of each operation based on the log to ensure the compliance and transparency of the operation.

[0219] Through the above implementation scheme, the automatic settlement function module 104 can efficiently and accurately adjust the expert status and extraction plan during the expert extraction process, ensure the smooth process, and respond to different extraction results and expert feedback in a timely manner. The module minimizes human intervention through automated rules and intelligent judgment, and optimizes the overall management and execution efficiency of expert extraction.

[0220] Example 5 Detailed Flowchart

[0221] Figure 2 Shown is a detailed flow chart of the expert extraction module and dynamic parameter adjustment module based on RPA technology.

[0222] The specific implementation process is as follows:

[0223] 1. Process start 201: After the system is started, the black box module 101 first isolates the direct interaction between the data user and the main database to ensure the compliance of the extraction process and data security.

[0224] 2. Parameter initialization 202: The expert extraction module 102 based on RPA technology extracts the parameters such as the expert notification ratio and the number of extractions preset by the parameter configuration function submodule 102a, and initializes the extraction operation.

[0225] 3. Intelligent prediction and simulation 203: Before performing the extraction operation, the dynamic parameter adjustment module 103 is started, and the data collection and preprocessing submodule 103a collects data from the current project and expert library, and inputs it into the machine learning model training and optimization submodule 103b for intelligent prediction.

[0226] 4. Extraction strategy adjustment 204: Based on the results of the prediction and simulation submodule 103c, the real-time strategy adjustment submodule 103d dynamically optimizes and adjusts the expert extraction strategy.

[0227] 5. Execute expert extraction 205: The expert extraction task after strategy adjustment is executed by the random extraction sequence generation submodule 102b to complete the random extraction of experts.

[0228] 6. Supplementary plan and scheduled task execution 206: If it is detected that the expert feedback does not meet the requirements, the supplementary plan optimization submodule 102d is started, combined with the scheduled task module 102e, to automatically execute the supplementary extraction process to ensure that a sufficient number of experts can respond and attend.

[0229] 7. Settlement and status update 207: After the expert extraction and notification process is completed, the automatic settlement function module 104 is started, and the solution status and expert attendance are adjusted according to the actual feedback to complete the entire extraction process.

[0230] 8. Process End 208: After all tasks are completed, the system returns to standby mode and waits for the next task instruction.

[0231] Through the above implementation, the bid evaluation expert extraction system of the present invention can achieve efficient, compliant and intelligent expert extraction and management, adapt to various complex project requirements and dynamic changes, and improve the quality and efficiency of bidding work.

[0232] The above description is only a preferred embodiment of the present invention, so all equivalent changes or modifications made according to the structure, characteristics and principles described in the scope of the patent application of the present invention are included in the scope of the patent application of the present invention.

Claims

1. A bid evaluation expert extraction system based on RPA technology and black box design, characterized by: The expert extraction system comprises: A black box module (101), used for performing black box processing of the expert extraction process; An expert extraction module (102), based on RPA technology, automatically performs expert extraction operations; A dynamic parameter adjustment module (103) performs intelligent prediction and simulation based on machine learning algorithms and historical data analysis, and adjusts the extraction strategy in real time; The automatic settlement function module (104) automatically adjusts the expert status and the extraction scheme status after the expert notification is completed.

2. A bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 1, characterized in that: The black box module (101) comprises: An independent data user setting module, wherein the independent data user setting module sets up one or more independent data users, each of which has specific access rights and operation rights, and the data user operates under an authentication and authorization mechanism; A data isolation module, wherein the data isolation module isolates data through logical isolation and hardware isolation strategies. The logical isolation isolates data through a virtual data view, and the hardware isolation isolates data through a dedicated storage device or virtual storage. Data transmission is performed through an encrypted channel. The virtual data view extracts necessary information related to expert extraction from the original database and displays it as an independent data set. A data operation normalization module defines a standardized data operation process and ensures the traceability and security of data operations through log recording and anomaly detection.

3. The bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 1 is characterized by: The expert extraction module (102) comprises: The parameter configuration function submodule (102a) is used to extract and manage various parameters in the expert extraction process and store the adjusted parameters; A random extraction sequence generation submodule (102b), used to set extraction conditions through a logic analysis module and generate an expert random extraction sequence using a pseudo-random number generation algorithm, wherein the expert random extraction sequence complies with regulatory requirements; An extraction plan preparation submodule (102c) is used to prepare an extraction plan according to preset operation parameters, optimize the plan and publish it to the task scheduling queue for execution; A supplementation plan optimization submodule (102d), used to automatically generate and adjust supplementation strategies when the expert is absent or refuses; The scheduled task module (102e) is used to automatically perform expert extraction and notification operations according to set time and conditions, and record operation steps and results.

4. A bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 3, characterized in that: The expert extraction module (102) further includes a plurality of RPA robots, which are configured in different computing environments and are responsible for executing independent subtasks in the expert extraction process. The plurality of RPA robots have automation scripts, which control the RPA robots to execute specific tasks and contain operation steps and conditional logic; the plurality of RPA robots monitor the progress of the expert extraction process and automatically adjust the extraction and notification strategies based on feedback; and combine machine learning models to analyze and predict expert behavior, adjust parameter settings and extraction strategies; automatically assign tasks according to task priorities and monitor execution status; generate operation logs and perform compliance checks.

5. The bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 1 is characterized by: The dynamic parameter adjustment module (103) comprises: A data collection and preprocessing submodule (103a), which is used to collect multi-source data related to expert extraction, and clean, normalize, extract and select features of the data to generate high-quality data suitable for machine learning model training; A machine learning model training and optimization submodule (103b), used to train the machine learning model using the preprocessed data, predict key parameters in the expert extraction process, and improve prediction accuracy through model optimization and updating; The prediction and simulation submodule (103c) is used to perform intelligent prediction and simulation on the expert extraction strategy based on the trained machine learning model, evaluate the possible results under different extraction strategies, and generate the optimal extraction plan; The real-time strategy adjustment submodule (103d) is used to adjust the actual expert extraction and notification strategy in real time in combination with the prediction and simulation results to cope with possible changes in the extraction process.

6. A bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 5, characterized in that: The machine learning model training and optimization submodule (103b) includes: Model selection and initialization unit, used to select appropriate machine learning models based on data characteristics and task requirements; A model training unit is used to train the selected model using the training set data and minimize the prediction error by iteratively optimizing the model parameters; Model validation and cross-validation unit, used to evaluate the performance of the model through cross-validation technology and select the optimal model; The model optimization and iterative update unit is used to regularly update and optimize the model, including retraining the model, adjusting hyperparameters, and using new data for incremental learning.

7. The bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 5 is characterized by: The prediction and simulation submodule (103c) comprises: The intelligent prediction unit is used to call the trained machine learning model to predict the key parameters in the expert extraction process. The output of the prediction model is a multi-output function: Among them, Y1 is the predicted value of the expert response rate, Y2 is the predicted value of the absence rate, and Y3 is the predicted value of the optimal number of extractions; functions f1 and f2 use logistic regression models, and the probability output is between 0 and 1; f3 is a linear regression model used to predict the optimal number of extractions; parameter α ij , β i are the coefficients of the model, i = 1, 2, j = 1...n, n is the total number of all influencing factors, γ and ∈ are the intercept term and error term respectively; X1, X2, ..., X n Various factors that affect the expert extraction results.

8. The bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 5, characterized in that: The prediction and simulation submodule (103c) comprises: The simulation unit is used to perform simulation based on the prediction results. The Monte Carlo method is used to perform multiple random samplings on the objective function Z to evaluate the effect of each extraction strategy. The objective function Z is expressed as: Among them, Y 1，i is the expert response rate prediction value under the i-th strategy, T i is the expert’s response time, Y 2，i is the predicted value of absence rate, Y 3,i is the predicted value of the optimal number of extractions, i=1...m, m is the total number of different extraction strategies evaluated during the simulation, and w1, w2, w3 are weight coefficients.

9. The bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 1, characterized in that: The real-time strategy adjustment submodule (103d) comprises: Feedback monitoring and adjustment triggering unit, which is used to monitor the actual feedback during the expert extraction and notification process, and automatically trigger the strategy adjustment process when the feedback rate deviates from the predicted value; A dynamic adjustment strategy generation unit is used to generate a new extraction strategy according to real-time feedback, and the strategy is generated by optimizing the rule engine and dynamic optimization algorithm; The strategy execution and continuous optimization unit is used to execute the adjusted extraction strategy through RPA technology and record all operation steps in the operation log for subsequent analysis and optimization.

10. The bid evaluation expert extraction system based on RPA technology and black box design as claimed in claim 1, characterized in that: The automatic settlement function module (104) includes: The non-responding expert status adjustment submodule is used to automatically update the status of non-responding experts to "not present" after each round of expert notification; The notification rule adjustment submodule is used to automatically adjust the status of experts not notified in this round according to the preset notification rules; The scheme status management submodule is used to dynamically adjust the scheme status according to expert feedback and extraction results. The status update is based on the number of shortfalls and notifications, and is executed according to the preset logic; The notification trigger management submodule is used to automatically generate and send notifications under specific conditions, and update the solution status to ensure that the status information is synchronized with the notification content.

Images