Automatic sampling and sample injection control system for intelligent calibration and calibration of gas mercury instrument

Abstract

The invention discloses an automatic sampling and sample introduction control system for intelligent calibration of a gas mercury analyzer, and relates to the technical field of earthquake underground fluid mercury analyzer equipment. The system comprises a data acquisition module, an attention head module, a central controller, an intelligent dead volume compensation and prediction module, a multi-sensor fusion module, a data storage and management module, an adaptive learning and calibration scheme optimization module and a user interface module. Multi-sensor fusion and an attention mechanism are utilized to realize multi-level feature extraction and information fusion, a central controller comprehensively decides and regulates and controls an execution mechanism in a closed loop mode to ensure that the gas mercury instrument rapidly and stably reaches a preset calibration target, and an intelligent dead volume compensation module predicts standard mercury gas residues in a gas circuit in real time by utilizing machine learning; and the adaptive learning module optimizes calibration parameters according to historical calibration data, so that the calibration precision and repeatability are effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of seismic underground fluid mercury detector equipment, in particular to an automatic sampling and injection control system for intelligent calibration of a mercury gas detector. Background Art

[0002] Some gas mercury stations are located in static water level observation wells, and their gas mercury measurements are relatively low. However, the minimum dose of standard mercury sample used in the existing calibration method is 5μL. Under conditions of 11°C to 25°C, its absolute mercury content ranges from 0.0305 to 0.0995ng, which is much higher than the actual measured values ​​of these low-background stations. This means that the calibration curve fitted with 5μL as the minimum calibration point is only applicable to stations with higher background concentrations and cannot cover weak signals below this standard. In addition, the existing calibration method has other problems: during manual sampling and injection, if the sampling volume is extremely small, it is difficult for the operator to accurately control it, which can easily introduce random and systematic errors; during manual operation, the handheld syringe may produce a heat conduction effect on the mercury standard due to the body temperature, resulting in a deviation in the mercury concentration injected into the instrument; differences in operating habits among different calibrators can also cause inconsistent calibration results.

[0003] Therefore, there is an urgent need for a calibration system that can realize automatic sampling and injection to solve the problems of insufficient lower range limit and human error in the existing methods when calibrating low-background mercury gas stations, thereby providing a scientific and reasonable calibration basis for low-signal mercury gas measurement points.

[0004] Chinese invention patent CN117471041A discloses a dilution ratio-based self-calibration method. This method utilizes a built-in elemental mercury generator to generate mercury-free zero gas and standard gas. The standard gas is then diluted by a dilution sampling assembly and fed into a mercury analyzer for measurement. Zero drift and dilution error are then corrected based on the measurement results. Although this method ensures measurement accuracy to a certain extent, it is highly dependent on the stability of the dilution system and is susceptible to environmental changes and system fluctuations. It also lacks the ability to make real-time dynamic adjustments.

[0005] In addition, Chinese utility model patent CN210269762U discloses an automatic detection device for a gas mercury analyzer. This device, through a control circuit and coordinated with various gas processing mechanisms, achieves carrier gas processing, saturated mercury gas delivery, and multi-point calibration and detection of multiple instruments. It has the advantages of accurate detection, high reliability, and easy operation. However, its design focuses on the automatic control and switching of gas circuits, and insufficient consideration is given to key links such as data processing, intelligent decision-making, and real-time closed-loop control. It is difficult to effectively respond to environmental changes and real-time dynamic adjustments to equipment status, and the calibration optimization function is relatively lacking. Summary of the Invention

[0006] The purpose of the present invention is to overcome the shortcomings of the existing technology and propose an automatic sampling and injection control system for the intelligent calibration of mercury gas analyzers. By introducing advanced multi-sensor fusion, attention mechanism feature extraction, intelligent dead volume compensation and adaptive learning optimization technology, it realizes the comprehensive real-time collection and intelligent processing of environmental parameters, instrument response and historical calibration data, thereby improving calibration accuracy and system stability, simplifying the operation process, and effectively improving the overall detection efficiency and reliability.

[0007] The object of the present invention is achieved through the following technical solutions: an automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer, comprising at least: a data acquisition module, an attention head module, a central controller, an intelligent dead volume compensation and prediction module, a multi-sensor fusion module, and a data storage and management module; The attention head module includes at least: sensor data attention head, instrument response attention head, dead volume prediction attention head, calibration workflow status attention head, historical data attention head and control action attention head; The sensor data attention head includes: a fusion data receiving unit for receiving the fused environment and system state data; an environment and system state feature extraction unit for extracting feature vectors representing the environment and system state; The instrument response attention head includes: a response signal receiving unit for receiving the response signal of the mercury gas analyzer; an instrument response state feature analysis unit for analyzing the signal strength, stability, noise level and change trend to generate a response state feature vector; The dead volume prediction attention head includes: a prediction result receiving unit, configured to receive a predicted dead volume or compensation parameter feature vector from the intelligent dead volume compensation and prediction module; The calibration workflow state attention head includes: a state tracking unit for tracking the steps of the current calibration process; a state encoding unit for encoding the current calibration step into a feature vector; The historical data attention head includes: a data retrieval unit for retrieving historical calibration data from the data storage and management module; a historical information feature extraction unit for extracting feature vectors of historical calibration curve parameters and performance indicators; The control action attention head includes: an instruction receiving unit for receiving instructions from the central controller; a control signal generating unit for generating control signals for controlling the actions of the servo motor and the air valve.

[0008] It also includes an adaptive learning and calibration scheme optimization module for adaptively optimizing future calibration schemes based on historical calibration data and performance indicators; a user interface module for displaying collected data and system status in real time; The multi-sensor fusion module is used to improve the accuracy of environmental and system status data; the data acquisition module is used to collect environmental parameters, system status data and the response signal of the mercury gas analyzer to be calibrated in real time; the attention head module is used to perform multi-level feature extraction and information fusion on the collected data; the central controller is used to receive the feature vectors from each attention head and make decisions based on the preset calibration strategy; the intelligent dead volume compensation and prediction module is used to collect historical calibration data and environmental and system operating parameters, train the dead volume prediction model through machine learning algorithms, and predict the residual amount of standard mercury gas in the gas path before calibration in real time; the data storage and management module is used to store and manage various types of system data.

[0009] The data acquisition module at least includes: a sensor data acquisition unit for collecting temperature, humidity, air pressure and its change rate, servo motor current and gas path switching valve status; a mercury gas meter response signal acquisition unit for acquiring the response signal of the mercury gas meter to be calibrated; The central controller includes an attention calculation unit, a decision logic unit and a closed-loop control unit.

[0010] The attention calculation unit uses a self-attention mechanism to calculate and fuse the weights of the feature vectors from each attention head. The attention calculation unit contains multi-head attention sub-units for parallel processing of different feature vectors. The decision logic unit is used to generate control action instructions based on the fused information and send the instructions to the control action attention head. The closed-loop control unit dynamically adjusts the injection volume or injection rate of the servo motor based on the real-time response signal of the mercury gas meter through PID control or adaptive control algorithm, so that the mercury gas meter responds quickly to the predetermined target and remains stable.

[0011] The intelligent dead volume compensation and prediction module further includes: The data collection unit is used to collect historical calibration data, environmental parameters and system operating parameters; the model training unit is used to train the dead volume prediction model using a machine learning algorithm; the real-time prediction unit is used to predict the residual amount of standard mercury gas in the gas path based on the current parameters before each calibration, and send the prediction results to the dead volume prediction attention head.

[0012] The adaptive learning and calibration scheme optimization module includes at least: The calibration data storage unit is used to store historical calibration data and performance indicators. The performance indicator evaluation unit is used to evaluate the linearity and repeatability of the calibration curve for each calibration. The calibration parameter optimization unit uses reinforcement learning or Bayesian optimization algorithms to adaptively optimize future calibration parameters based on historical performance and feed back optimization suggestions to the decision logic unit of the central controller.

[0013] The user interface module includes at least: The real-time data display unit is used to display real-time sensor data and mercury gas analyzer response signals; the calibration status display unit is used to display the real-time status of the calibration process; and the attention weight display unit is used to intuitively display the weight information of each attention head so that users can understand the system decision-making process.

[0014] The data storage and management module at least includes: The configuration storage unit is used to store calibration parameter configurations; the real-time data storage unit is used to store real-time collected sensor data and mercury analyzer response data; the result storage unit is used to store calibration results; the log recording unit is used to record system operation logs; and the historical data storage unit is used to store historical calibration data and adaptive learning model parameters.

[0015] The multi-sensor fusion module includes at least: The data receiving unit is used to receive data from temperature, humidity and air pressure sensors; the data fusion unit uses Kalman filtering or weighted averaging algorithm to fuse the data and send the fused data to the sensor data attention head.

[0016] The following steps are involved: Real-time data acquisition step: using the data acquisition module to collect environmental parameters, system status data and the response signal of the mercury gas analyzer to be calibrated in real time; Feature extraction and fusion steps: The sensor data attention head, instrument response attention head, dead volume prediction attention head, calibration workflow status attention head, and historical data attention head in the attention head module are used to extract features from the collected data, predicted dead volume, calibration status, and historical data, respectively. The resulting feature vectors are then transmitted to the attention calculation unit of the central controller for weighted fusion. Intelligent decision-making and control steps: The decision logic unit of the central controller generates control instructions based on the fused feature vectors, and converts them into control signals through the control action attention head of the attention head module to drive the servo motor and gas valve to perform automatic sampling and injection operations; Closed-loop feedback control steps: The closed-loop control unit of the central controller monitors the response signal of the mercury meter provided by the data acquisition module in real time, and uses PID or adaptive control algorithms to dynamically adjust the injection volume or injection rate to ensure that the mercury meter responds quickly and stably; Intelligent dead volume compensation step: The intelligent dead volume compensation and prediction module predicts the residual amount of standard mercury gas in the gas path based on historical data and current parameters, and transmits the prediction result to the dead volume prediction attention head of the attention head module; Adaptive Optimization Step: The adaptive learning and calibration scheme optimization module optimizes future calibration parameters based on historical calibration data and performance indicators, and feeds back the optimization suggestions to the decision logic unit of the central controller; Data storage and management steps: Use the data storage and management module to store and manage the collected raw data, fused feature vectors, control instructions, calibration results, system operation logs and adaptive learning model parameters.

[0017] The beneficial effects of the present invention are: 1. The system collects environmental parameters, system status and mercury gas analyzer response signals in real time, and uses multi-sensor fusion technology to effectively improve data accuracy. The multi-level feature extraction and information fusion mechanism enables various types of data to be processed carefully, providing a reliable basis for subsequent decision-making, thereby greatly reducing errors and improving calibration accuracy.

[0018] 2. An attention mechanism is used to perform weighted fusion of various data. Combined with the self-attention mechanism and multi-head parallel processing, a comprehensive analysis of environmental conditions, instrument responses, calibration procedures, and historical data is achieved. The control instructions generated from this can accurately drive the actuator and dynamically adjust the injection parameters through a real-time feedback mechanism, ensuring that the mercury analyzer quickly and stably reaches the predetermined calibration target in a short period of time.

[0019] 3. The system uses machine learning algorithms to train models based on historical calibration data, environmental and operating parameters, predicting the residual amount of standard mercury gas in the gas path in real time and performing intelligent compensation accordingly. At the same time, by evaluating the linearity and repeatability of the calibration curve, the system can adaptively optimize future calibration parameters to achieve continuous accuracy improvement and stability assurance in long-term operation.

[0020] 4. The user interface module displays information such as collected data, calibration status, and attention weight in real time, making the entire calibration process transparent to operators and facilitating real-time monitoring and manual intervention when necessary. A comprehensive data storage and management mechanism ensures that the original data, fusion features, control instructions, calibration results, and logs generated during system operation can be effectively stored and quickly retrieved, providing solid data support for subsequent analysis and fault diagnosis.

[0021] 5. The present invention adopts a modular design with clear interfaces and efficient collaboration between functional modules. It can not only realize the complex functions of current automatic sampling and intelligent calibration, but also provide a flexible technical platform for future functional upgrades, maintenance and system expansion. The high integration and flexibility of the overall system effectively reduce the difficulty and cost of maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Figure 1 of the present invention; Figure 2 Figure 1 is a diagram of the present invention. DETAILED DESCRIPTION

[0023] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0024] It is to be noted that the directions of "left", "right", "up", "down", "front", "back", "inside" and "outside" in the following schemes are all relative directions and are not listed here one by one.

[0025] The complete claims are summarized below. Their structure includes the main claim and each dependent claim, which describes in detail each module and unit of the system and their functions and interrelationships. Each module and unit is clearly numbered. The specific contents are as follows: Example 1 This embodiment provides an automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer, which is mainly composed of the following six modules: Data acquisition module 1, attention head module 2, central controller 3, intelligent dead volume compensation and prediction module 4, multi-sensor fusion module 5, data storage and management module 8.

[0026] Through high-speed data transmission and signal interaction, each module realizes real-time collection, fusion, feature extraction, prediction, decision-making and storage of environmental parameters, mercury gas analyzer response and historical calibration data, forming a closed-loop calibration control system.

[0027] Specific structure and working principle of each module and subunit The data acquisition module 1 is responsible for real-time acquisition of various raw data generated during the operation of the system, including environmental parameters, system status data, and response signals of the mercury gas analyzer to be calibrated.

[0028] Receive the original signals from the temperature, humidity, air pressure sensors and mercury gas meter, and pass the collected raw data to the multi-sensor fusion module 5 and the subsequent attention head module 2 for processing.

[0029] The multi-sensor fusion module 5 includes: The data receiving unit 501 is used to receive real-time data from sensors such as temperature, humidity and air pressure.

[0030] The data fusion unit 502 fuses the received sensor data using Kalman filtering or weighted average algorithm, thereby improving the accuracy of the environment and system status data.

[0031] The raw sensor data from the data acquisition module 1 is received, and the fused environment and system status data are sent to the sensor data attention head 201 in the attention head module 2.

[0032] Attention head module 2 is used to perform multi-level feature extraction and information fusion on data from different sources. It includes the following six attention heads: Sensor Data Attention Head 201 The fusion data receiving unit 2011 receives data from the multi-sensor fusion module 5 .

[0033] The environment and system status feature extraction unit 2012 performs feature extraction on the received data to generate a feature vector representing the environment and system status.

[0034] Instrument response attention head 202 The response signal receiving unit 2021 receives the mercury gas analyzer response signal from the data acquisition module 1 .

[0035] The instrument response state characteristic analysis unit 2022 analyzes the signal strength, stability, noise level and change trend to generate an instrument response state characteristic vector.

[0036] Dead Volume Prediction Attention Head 203 The prediction result receiving unit 2031 is used to receive the predicted dead volume or compensation parameter feature vector from the intelligent dead volume compensation and prediction module 4 .

[0037] Calibrate Workflow State Attention Head 204 The state tracking unit 2041 tracks each step of the current calibration process, and the state encoding unit 2042 encodes the current calibration step into a feature vector to reflect the process state.

[0038] Historical Data Attention Head 205 The data retrieval unit 2051 retrieves historical calibration data from the data storage and management module 8, and the historical information feature extraction unit 2052 extracts features from historical calibration curve parameters and performance indicators to form a historical data feature vector.

[0039] Control action attention head 206 The instruction receiving unit 2061 receives the control instruction from the central controller 3 , and the control signal generating unit 2062 generates a control signal for controlling the servo motor and the air valve according to the received instruction, thereby driving the automatic sampling and injection operation.

[0040] The central controller 3 is responsible for aggregating the feature vectors of each attention head from the attention head module 2, making a comprehensive decision based on the preset calibration strategy, and converting the decision results into control instructions.

[0041] Receive feature vectors from each attention head.

[0042] The generated control instructions are passed to the execution unit through the control action attention head 206 to complete the real-time control of the servo motor and the air valve.

[0043] The intelligent dead volume compensation and prediction module 4 is mainly used to solve the problem of standard mercury gas residue in the gas path during the calibration process, and improve the calibration accuracy through prediction compensation. Its internal structure is as follows: The data collection unit 401 collects historical calibration data, environmental parameters and system operating parameters to provide data support for prediction. The model training unit 402 uses a machine learning algorithm to train the collected data to build and optimize the dead volume prediction model. Before each calibration, the real-time prediction unit 403 predicts the residual amount of standard mercury gas in the gas path in real time based on the current environment and system parameters, and sends the prediction results to the dead volume prediction attention head 203.

[0044] Receive historical and real-time data from the data acquisition module 1 and the data storage and management module 8.

[0045] The predicted dead volume data (or compensation parameter feature vector) is passed to the prediction result receiving unit 2031 in the attention head module 2.

[0046] The data storage and management module 8 is used for hierarchical storage and management of system data, supporting real-time monitoring, historical data analysis, and model training. It receives raw data, fused feature vectors, calibration results, and log information from each module, and provides data retrieval and storage services for the historical data attention head 205 and other modules to call. Specifically, it includes: The configuration storage unit 801 stores the calibration parameter configuration of the system. The real-time data storage unit 802 stores the real-time collected sensor data and mercury analyzer response data to facilitate instant system control. The result storage unit 803 stores the calibration results generated by each calibration operation. The log recording unit 804 records various log information generated during the operation of the system to support fault analysis and maintenance. The historical data storage unit 805 stores historical calibration data and adaptive learning model parameters to provide data support for the intelligent dead volume compensation and prediction module 4.

[0047] Data collection and fusion: The data acquisition module 1 collects environmental parameters, system status data and mercury gas analyzer response signals in real time; the multi-sensor fusion module 5 obtains the original sensor data through the data receiving unit 501, and uses the data fusion unit 502 to fuse the data using Kalman filtering or weighted averaging algorithm to output accurate environmental and system status data; the fused data is sent to the sensor data attention head 201 (fusion data receiving unit 2011) in the attention head module 2.

[0048] Feature extraction and information fusion: Each attention head in attention head module 2 extracts features from the data it receives: The sensor data attention head 201 generates a feature vector by the environment and system state feature extraction unit 2012; the instrument response attention head 202 generates a response state feature vector through the instrument response state feature analysis unit 2022; the calibration workflow state attention head 204 generates a calibration process state vector through the state tracking unit 2041 and the state encoding unit 2042; the historical data attention head 205 extracts historical data features through the data retrieval unit 2051 and the historical information feature extraction unit 2052; the dead volume prediction attention head 203 receives the prediction results of the intelligent dead volume compensation and prediction module 4 through the prediction result receiving unit 2031.

[0049] Decision-making and control: The central controller 3 summarizes the feature vectors from each attention head, analyzes and processes them according to the preset calibration strategy, and generates calibration control instructions; the control instructions are converted into specific control signals through the instruction receiving unit 2061 and the control signal generating unit 2062 in the control action attention head 206, driving the servo motor and the air valve to perform automatic sampling and injection operations.

[0050] Intelligent prediction and feedback: The intelligent dead volume compensation and prediction module 4 collects historical and real-time data through the data collection unit 401 and uses the model training unit 402 to build a prediction model. Before each calibration, the real-time prediction unit 403 predicts the residual amount of standard mercury gas in the gas path based on the current parameters and transmits the prediction result to the dead volume prediction attention head 203 (prediction result receiving unit 2031). The prediction result is used to adjust the calibration strategy and further fed back to the central controller 3.

[0051] Data storage and management: The data storage and management module 8 stores various types of data in a hierarchical manner: The configuration storage unit 801 saves the system configuration; the real-time data storage unit 802 saves the collected real-time data; the result storage unit 803 records the calibration results; the log recording unit 804 records the operation log; the historical data storage unit 805 stores historical data and model parameters to support subsequent data retrieval and model optimization. The historical data is also called by the historical data attention head 205 to support the reference basis of system decision-making. The multi-sensor fusion module 5 adopts Kalman filtering or weighted average algorithm to realize data fusion, significantly improving the accuracy of environmental and system status data, and providing a reliable basis for subsequent calibration decisions. The attention head module 2 uses multi-channel feature extraction and information fusion to realize in-depth analysis of environmental status, instrument response, calibration process and historical data; the intelligent dead volume compensation and prediction module 4 realizes real-time prediction of the residual amount of standard mercury gas in the gas path based on machine learning, thereby effectively compensating for dead volume errors and improving calibration accuracy.

[0052] The data acquisition module 1 and the data storage and management module 8 ensure that the system can collect and save key data in real time; the central controller 3 makes decisions quickly based on the real-time fused feature vectors, and accurately drives the execution unit by controlling the action attention head 206 to achieve fast response and stable calibration.

[0053] Each module and sub-unit is clearly identified by a number and has a clear structure, which facilitates subsequent system maintenance, fault location and function expansion. At the same time, it lays a solid data foundation and structural guarantee for the subsequent optimization of adaptive learning and calibration solutions.

[0054] In summary, Example 1 integrates functions such as data acquisition, fusion, feature extraction, intelligent prediction, decision control, and data management through efficient collaboration between modules. It not only realizes high-precision automatic sampling and calibration of the mercury gas analyzer, but also improves the system response speed and calibration stability, providing reliable technical support and broad prospects for promotion and application for practical applications.

[0055] Example 2 This embodiment is composed of the following modules based on the first embodiment: The data acquisition module 1 includes a sensor data acquisition unit 101 and a mercury gas analyzer response signal acquisition unit 102; The multi-sensor fusion module 5 includes a data receiving unit 501 and a data fusion unit 502; The attention head module 2 includes a sensor data attention head 201 (including a fusion data receiving unit 2011 and an environment and system state feature extraction unit 2012), an instrument response attention head 202 (including a response signal receiving unit 2021 and an instrument response state feature analysis unit 2022), a dead volume prediction attention head 203 (including a prediction result receiving unit 2031), a calibration workflow state attention head 204 (including a state tracking unit 2041 and a state encoding unit 2042), a history data attention head 205 (including a data retrieval unit 2051 and a history information feature extraction unit 2052), and a control action attention head 206 (including an instruction receiving unit 2061 and a control signal generating unit 2062). The central controller 3 includes an attention calculation unit 301 (with a multi-head attention sub-unit 3011), a decision logic unit 302 and a closed-loop control unit 303; The intelligent dead volume compensation and prediction module 4 includes a data collection unit 401, a model training unit 402, and a real-time prediction unit 403; The adaptive learning and calibration scheme optimization module 6 includes a calibration data storage unit 601, a performance indicator evaluation unit 602, and a calibration parameter optimization unit 603; The user interface module 7 includes a real-time data display unit 701, a calibration status display unit 702, and an attention weight display unit 703; The data storage and management module 8 includes a configuration storage unit 801, a real-time data storage unit 802, a result storage unit 803, a log recording unit 804, and a historical data storage unit 805; Data and control signals are transmitted between modules through a high-speed data bus or dedicated interface, forming a closed-loop system of data acquisition, fusion, feature extraction, intelligent prediction, decision control, parameter optimization and human-computer interaction.

[0056] The sensor data acquisition unit 101 collects environmental and system status data such as temperature, humidity, air pressure and its rate of change, servo motor current, and gas path switching valve status. The mercury gas meter response signal acquisition unit 102 obtains the response signal of the mercury gas meter to be calibrated to ensure the collection of real-time dynamic information. It receives the original signals from each sensor and the mercury gas meter and transmits the collected data to the multi-sensor fusion module 5 (raw sensor data) and the subsequent attention head module 2 (mercury gas meter response signal).

[0057] The data acquisition module 1 sends the collected signals to the multi-sensor fusion module 5 through the high-speed interface for data fusion, and at the same time transmits the mercury gas analyzer response signal to the instrument response attention head 202 (response signal receiving unit 2021).

[0058] The data receiving unit 501 receives real-time data from the sensor data acquiring unit 101 .

[0059] The data fusion unit 502 uses Kalman filtering or weighted average algorithm to fuse the sensor data to improve data accuracy.

[0060] Receive raw sensor data from data acquisition module 1.

[0061] The fused environment and system status data are sent to the sensor data attention head 201 (fused data receiving unit 2011) in the attention head module 2.

[0062] Ensure that the fused data is passed to the attention head module 2 after processing to provide high-precision basic data for subsequent feature extraction.

[0063] The attention head module 2 performs multi-level feature extraction and information fusion for different data sources, specifically including: Sensor data attention head 201: The fusion data receiving unit 2011 receives the fusion data from the multi-sensor fusion module 5, and the environment and system state feature extraction unit 2012 extracts the feature vectors of the environment and system state.

[0064] Instrument response attention head 202: The response signal receiving unit 2021 receives the mercury gas analyzer response signal from the data acquisition module 1, and the instrument response state feature analysis unit 2022 analyzes the signal characteristics to generate a response state feature vector.

[0065] Dead volume prediction attention head 203: The prediction result receiving unit 2031 receives the prediction feature vector output by the intelligent dead volume compensation and prediction module 4.

[0066] Calibration Workflow State Attention Head 204: The state tracking unit 2041 tracks the calibration process steps, and the state encoding unit 2042 encodes the steps into feature vectors.

[0067] Historical data attention head 205: The data retrieval unit 2051 retrieves historical calibration data from the data storage and management module 8, and the historical information feature extraction unit 2052 extracts historical data feature vectors.

[0068] Control action attention head 206: The instruction receiving unit 2061 receives the control instruction from the central controller 3, and the control signal generating unit 2062 generates the control signal to drive the servo motor and the air valve action.

[0069] Receive data from the multi-sensor fusion module 5, the data acquisition module 1, the intelligent dead volume compensation and prediction module 4 and the data storage and management module 8.

[0070] The feature vector extracted by each attention head is transmitted to the central controller 3 (attention calculation unit 301).

[0071] Each sub-unit extracts corresponding features independently, and after information fusion within the module, the multi-dimensional feature vector is uniformly transmitted to the central controller 3 for decision-making.

[0072] The central controller 3 receives optimization suggestions from the attention head module 2 (the feature vectors of each sub-unit) and the adaptive learning and calibration scheme optimization module 6. At the same time, the closed-loop control unit 303 receives real-time feedback from the data acquisition module 1 (through the mercury gas meter response signal acquisition unit 102), generates and outputs control instructions to the control action attention head 206 (instruction receiving unit 2061), and dynamically adjusts the execution units such as the servo motor and the gas valve.

[0073] The central controller 3 is the system decision center, which is divided into three parts: The attention calculation unit 301 uses a self-attention mechanism to perform weighted calculation and fusion on the feature vectors from each attention head, and has a built-in multi-head attention sub-unit 3011 to achieve parallel processing. The decision logic unit 302 generates control action instructions based on the fused information, and at the same time receives optimization suggestions from the adaptive learning and calibration scheme optimization module 6. The closed-loop control unit 303 monitors the real-time response signal of the mercury gas analyzer and dynamically adjusts the injection volume or injection rate of the servo motor through PID or adaptive control algorithms to ensure that the calibration target is achieved quickly and stably.

[0074] The information output by each attention head is integrated through the attention calculation unit 301, and instructions are generated by the decision logic unit 302. The closed-loop control unit 303 is regulated in combination with real-time feedback to form a precise control closed loop. At the same time, the suggestions of the adaptive learning and calibration scheme optimization module 6 are accepted to further optimize the decision.

[0075] The intelligent dead volume compensation and prediction module 4 receives historical and real-time data from the data acquisition module 1 and the data storage and management module 8, and uses the internal algorithm model to continuously update the prediction results, providing key compensation information to the central controller 3 to help make calibration decisions more accurate. The predicted dead volume or compensation parameter feature vector is sent to the dead volume prediction attention head 203 of the attention head module 2.

[0076] The data collection unit 401 collects historical calibration data, environmental parameters and system operating parameters. The model training unit 402 uses a machine learning algorithm to train a dead volume prediction model. The real-time prediction unit 403 predicts the residual amount of standard mercury gas in the gas path in real time before calibration based on the currently collected parameters, and transmits the prediction result to the dead volume prediction attention head 203 (prediction result receiving unit 2031).

[0077] The adaptive learning and calibration scheme optimization module 6 receives the historical calibration data and real-time performance indicators from the data storage and management module 8 (historical data storage unit 805), and transmits the generated optimization suggestions to the central controller 3 (decision logic unit 302). It continuously tracks the calibration effect and adjusts future calibration parameters through internal evaluation and optimization algorithms, so that the system can continuously optimize itself in long-term operation and improve calibration accuracy and repeatability.

[0078] The calibration data storage unit 601 stores system historical calibration data and performance indicators, providing a data basis for parameter optimization.

[0079] The performance index evaluation unit 602 evaluates the linearity and repeatability of the calibration curve generated by each calibration to form a performance index.

[0080] The calibration parameter optimization unit 603 utilizes reinforcement learning or Bayesian optimization algorithm to adaptively optimize future calibration parameters based on historical performance indicators, and feeds back optimization suggestions to the decision logic unit 302 of the central controller 3 .

[0081] The user interface module 7 receives relevant data from the data storage and management module 8, the central controller 3 and the attention head module 2, and presents the system status, sensor data and calibration process information in real time in the form of graphics or text. The user interface module 7 makes the dynamic data within the system transparent, and through real-time display and status feedback, it is convenient for operators to monitor the calibration process and perform manual intervention or adjustments when necessary.

[0082] The real-time data display unit 701 displays the real-time collected sensor data and mercury analyzer response signal for the user to monitor in real time. The calibration status display unit 702 displays the status and progress of each step of the current calibration process.

[0083] The attention weight display unit 703 intuitively displays the weight information of each attention head in the information fusion process to help users understand the basis of system decision-making.

[0084] The data storage and management module 8 receives various types of data from the data acquisition module 1, the central controller 3, the intelligent dead volume compensation and prediction module 4 and the adaptive learning and calibration scheme optimization module 6. The data is called by the historical data attention head 205, the adaptive learning and calibration scheme optimization module 6 and the user interface module 7. Through the structured storage and retrieval mechanism, it provides data support for each module to ensure seamless connection between the system historical data and real-time data, and realize intelligent decision-making and parameter adaptive optimization.

[0085] The configuration storage unit 801 stores the system calibration parameter configuration file, the real-time data storage unit 802 stores the real-time collected sensor data and mercury meter response data, the result storage unit 803 stores the calibration results of each calibration operation, the log recording unit 804 records the system operation log and error information to facilitate maintenance and fault analysis, and the historical data storage unit 805 stores historical calibration data and adaptive learning model parameters to provide support for subsequent optimization and data retrieval.

[0086] Data collection and preprocessing: The data acquisition module 1 collects the environment, system status and instrument response data in real time through the sensor data acquisition unit 101 and the mercury gas analyzer response signal acquisition unit 102. The collected sensor data is sent to the multi-sensor fusion module 5, fused by the data fusion unit 502 and output to the sensor data attention head 201 (fused data receiving unit 2011).

[0087] Multi-level feature extraction and information fusion: Each sub-unit of the attention head module 2 extracts features from sensor data, instrument response, historical data, calibration process and dead volume prediction data, and generates its own feature vector. These feature vectors are sent to the attention calculation unit 301 (including the multi-head attention sub-unit 3011) of the central controller 3, and the self-attention mechanism is used to achieve weighted fusion.

[0088] Decision making and closed-loop control: The fused feature vectors are processed by the decision logic unit 302 of the central controller 3. At the same time, they are combined with the optimization suggestions fed back by the adaptive learning and calibration scheme optimization module 6 to generate specific control instructions. The control instructions drive the servo motor and the gas valve to perform the injection action through the control action attention head 206 (instruction receiving unit 2061, control signal generation unit 2062). The closed-loop control unit 303 uses the real-time feedback obtained from the mercury gas analyzer response signal acquisition unit 102 to dynamically adjust the execution parameters through PID or adaptive algorithms to ensure that the calibration target is achieved quickly and stably.

[0089] Intelligent prediction and adaptive optimization: The intelligent dead volume compensation and prediction module 4 uses the data collection unit 401, the model training unit 402 and the real-time prediction unit 403 to predict the standard mercury gas residual in the gas path in real time, and transmits the prediction result to the dead volume prediction attention head 203 (prediction result receiving unit 2031). The adaptive learning and calibration scheme optimization module 6 obtains historical calibration data from the data storage and management module 8 (historical data storage unit 805), evaluates the calibration effect through the performance index evaluation unit 602, and uses the calibration parameter optimization unit 603 to adaptively adjust future parameters, and the optimization suggestions are fed back to the central controller 3 (decision logic unit 302).

[0090] Data storage and human-computer interaction: The data storage and management module 8 stores and manages all collected, fused, predicted and controlled data, and provides data support for the historical data attention head 205, the adaptive learning and calibration scheme optimization module 6 and the user interface module 7. The user interface module 7 displays the collected data in real time (through the real-time data display unit 701), the calibration status (calibration status display unit 702) and the weight information of each attention head (attention weight display unit 703), enabling the operator to intuitively monitor the system status.

[0091] The adaptive learning and calibration scheme optimization module 6 evaluates calibration performance and historical data in real time, and uses reinforcement learning or Bayesian optimization algorithms to continuously adjust calibration parameters, so that the system can continuously improve calibration accuracy and repeatability in long-term operation.

[0092] The self-attention mechanism (attention calculation unit 301 and multi-head attention sub-unit 3011) within the central controller 3 works in conjunction with the closed-loop control unit 303, enabling the system to achieve precise decision-making and dynamic regulation based on multi-dimensional features, ensuring that the mercury gas analyzer quickly and stably reaches the preset calibration target.

[0093] The user interface module 7 enhances the transparency of the system operation process through real-time data display, calibration status display and intuitive display of attention weight, which helps users to monitor in real time and intervene when necessary, thereby improving system reliability and user trust.

[0094] The multi-sensor fusion module 5 and the attention head module 2 implement multi-level feature extraction, which, combined with the intelligent dead volume compensation and real-time prediction of the prediction module 4, provides the system with accurate received data, greatly improving the accuracy and stability of the overall calibration.

[0095] The modular design (each module and sub-unit is clearly numbered) makes the system structure clear, easy to maintain and expand, and provides a good platform foundation for subsequent functional upgrades and the integration of new technologies.

[0096] In summary, Example 2, based on Example 1, introduces the adaptive learning and calibration scheme optimization module 6 and the user interface module 7, and refines the data acquisition module 1 and the central controller 3, thereby forming a high-performance mercury gas analyzer calibration system that is tightly integrated with data acquisition, fusion, intelligent decision-making, real-time feedback, and self-optimization. This system not only improves the calibration accuracy and response speed, but also enhances the system transparency and maintainability, providing more advanced and reliable technical support for practical applications.

[0097] Example 3 Based on the module configurations of Examples 1 and 2, this embodiment further proposes an automatic sampling and injection control method for realizing intelligent calibration of a mercury gas analyzer based on a step-by-step process. This embodiment closely integrates the modules with the method steps to achieve a complete process from real-time data acquisition, feature extraction and fusion, intelligent decision-making and closed-loop control, dead volume compensation, adaptive learning optimization, to data storage and management.

[0098] Real-time data collection steps The sensor data acquisition unit 101 collects data such as temperature, humidity, air pressure, servo motor current, and gas path switching valve status in real time; the mercury gas meter response signal acquisition unit 102 collects the response signal of the mercury gas meter to be calibrated; and the collected raw data are sent to the multi-sensor fusion module 5 and the attention head module 2 respectively.

[0099] Attention mechanism feature extraction and fusion steps The sensor data attention head 201 (via the fusion data receiving unit 2011 and the environment and system state feature extraction unit 2012) extracts feature vectors from the fused data. The instrument response attention head 202 (via the response signal receiving unit 2021 and the instrument response state feature analysis unit 2022) extracts features from the mercury analyzer response signal. The dead volume prediction attention head 203 obtains the prediction results of the intelligent dead volume compensation and prediction module 4 via the prediction result receiving unit 2031. The calibration workflow state attention head 204 and the historical data attention head 205 respectively extract features of the current calibration process and historical calibration data. All feature vectors are sent to the attention calculation unit 301 of the central controller 3 (with a multi-head attention sub-unit 3011) for weighted calculation and fusion. The fused multidimensional feature vector is then passed to the decision logic unit 302.

[0100] Intelligent decision-making and control steps The decision logic unit 302 generates specific control action instructions based on the fused feature vectors and the optimization suggestions fed back by the adaptive learning and calibration scheme optimization module 6; the control instructions are converted into control signals by the control action attention head 206 (instruction receiving unit 2061, control signal generating unit 2062), driving the servo motor and the air valve to perform automatic sampling and injection operations, and the control action signals are output to the execution unit to realize automatic calibration operations.

[0101] Closed-loop feedback control steps The closed-loop control unit 303 monitors the response signal of the mercury meter in the data acquisition module 1 (provided by the mercury meter response signal acquisition unit 102) in real time. Based on the real-time feedback, the closed-loop control unit 303 dynamically adjusts the injection volume or injection rate of the servo motor through PID control or adaptive control algorithm to ensure that the mercury meter responds quickly to the preset target and remains stable. The dynamically adjusted control signal is fed back to the actuator to form a closed-loop regulation.

[0102] Smart dead volume compensation steps The data collection unit 401 collects historical calibration data and current environmental parameters; the model training unit 402 uses a machine learning algorithm to train and update the dead volume prediction model; the real-time prediction unit 403 predicts the residual amount of standard mercury gas in the gas path based on the current parameters before each calibration, and sends the prediction result to the dead volume prediction attention head 203. The prediction result participates in subsequent feature fusion and decision-making to compensate for the influence of residual mercury gas in the gas path.

[0103] Adaptive learning and calibration scheme optimization steps The calibration data storage unit 601 stores and updates historical calibration data and performance indicators; the performance indicator evaluation unit 602 evaluates the effect of each calibration to obtain the linearity and repeatability of the calibration curve; the calibration parameter optimization unit 603 uses reinforcement learning or Bayesian optimization algorithm to adaptively optimize future calibration parameters based on the evaluation results, and feeds back the optimization suggestions to the decision logic unit 302 of the central controller 3. The optimization suggestions are used to correct subsequent calibration strategies and improve the overall calibration accuracy of the system.

[0104] Data storage and management steps The raw data collected by the data acquisition module 1, the fusion features extracted by the attention head module 2, the control instructions generated by the central controller 3, the calibration results, the system operation logs, and the adaptive learning model parameters are stored according to the configuration; at the same time, it supports historical data retrieval (for historical data attention head 205 to call) and subsequent adaptive learning and calibration scheme optimization module 6 to optimize use. Structured data storage ensures the security, traceability and efficient call of system data, providing data protection for the overall calibration process.

[0105] By clearly decomposing the steps of real-time data acquisition, feature extraction, intelligent decision-making, closed-loop feedback, dead volume compensation, adaptive optimization and data storage, Example 3 constructs a full-process closed-loop system from data acquisition to decision execution to ensure the efficiency and accuracy of the calibration process.

[0106] The attention head module 2 uses multi-dimensional feature extraction and the self-attention mechanism (attention calculation unit 301 and multi-head attention sub-unit 3011) within the central controller 3 to achieve information fusion, enabling the system to conduct a comprehensive analysis of environmental conditions, instrument responses, calibration process status, historical data and prediction results, thereby generating more accurate control instructions.

[0107] The closed-loop control unit 303 combines the real-time collected mercury gas analyzer response signal and dynamically adjusts the injection parameters through PID or adaptive algorithms to ensure a rapid and stable system response. At the same time, the adaptive learning and calibration scheme optimization module 6 continuously learns historical data to achieve dynamic optimization of calibration parameters, improving the accuracy and consistency of long-term operation.

[0108] The user interface module 7 displays key data and calibration status in real time, allowing operators to intuitively monitor system operation and promptly grasp the calibration progress and attention head weight information, thereby enhancing the transparency and reliability of the system.

[0109] In summary, Example 3 closely integrates the hardware module with the step-based calibration process to construct a complete system including real-time data acquisition, feature extraction and fusion, intelligent decision-making, closed-loop control, intelligent dead volume compensation, adaptive optimization and data storage management, thereby realizing high-precision automatic calibration and injection control of the mercury gas analyzer. This method not only improves the system's response speed and calibration accuracy, but also effectively ensures the long-term stability and reliability of the system through transparent human-computer interaction and dynamic adaptive optimization, providing efficient technical support for practical applications.

[0110] The above description is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be regarded as excluding other embodiments. Instead, it can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or technology or knowledge in related fields. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention and should be within the scope of protection of the claims attached to the present invention.

Claims

1. An automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer, characterized in that: At least including: a data acquisition module (1), an attention head module (2), a central controller (3), an intelligent dead volume compensation and prediction module (4), a multi-sensor fusion module (5), and a data storage and management module (8); The attention head module (2) at least includes: a sensor data attention head (201), an instrument response attention head (202), a dead volume prediction attention head (203), a calibration workflow status attention head (204), a historical data attention head (205), and a control action attention head (206); The sensor data attention head (201) comprises: a fusion data receiving unit (2011) for receiving fused environment and system state data; an environment and system state feature extraction unit (2012) for extracting feature vectors representing the environment and system state; The instrument response attention head (202) comprises: a response signal receiving unit (2021) for receiving a response signal from the mercury gas analyzer; an instrument response state feature analysis unit (2022) for analyzing signal strength, stability, noise level and change trend to generate a response state feature vector; The dead volume prediction attention head (203) comprises: a prediction result receiving unit (2031) for receiving a predicted dead volume or compensation parameter feature vector from the intelligent dead volume compensation and prediction module (4); The calibration workflow state attention head (204) comprises: a state tracking unit (2041) for tracking the steps of the current calibration process; a state encoding unit (2042) for encoding the current calibration step into a feature vector; The historical data attention head (205) comprises: a data retrieval unit (2051) for retrieving historical calibration data from the data storage and management module (8); a historical information feature extraction unit (2052) for extracting feature vectors of historical calibration curve parameters and performance indicators; The control action attention head (206) comprises: an instruction receiving unit (2061) for receiving instructions from a central controller (3); and a control signal generating unit (2062) for generating control signals for controlling the actions of the servo motor and the air valve.

2. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 1, characterized in that: It also includes an adaptive learning and calibration scheme optimization module (6) for adaptively optimizing future calibration schemes based on historical calibration data and performance indicators; a user interface module (7) for displaying collected data and system status in real time; The multi-sensor fusion module (5) is used to improve the accuracy of environmental and system status data; the data acquisition module (1) is used to collect environmental parameters, system status data and response signals of the mercury gas meter to be calibrated in real time; the attention head module (2) is used to perform multi-level feature extraction and information fusion on the collected data; the central controller (3) is used to receive feature vectors from each attention head and make decisions based on a preset calibration strategy; the intelligent dead volume compensation and prediction module (4) is used to collect historical calibration data and environmental and system operating parameters, train a dead volume prediction model through a machine learning algorithm, and predict in real time the residual amount of standard mercury gas in the gas path before calibration; the data storage and management module (8) is used to store and manage various types of system data.

3. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 2, characterized in that: The data acquisition module (1) comprises at least: a sensor data acquisition unit (101) for acquiring temperature, humidity, air pressure and its change rate, servo motor current and gas path switching valve status; and a mercury gas meter response signal acquisition unit (102) for acquiring a response signal of the mercury gas meter to be calibrated; The central controller (3) includes an attention calculation unit (301), a decision logic unit (302) and a closed-loop control unit (303).

4. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 3, characterized in that: The attention calculation unit (301) uses a self-attention mechanism to perform weight calculation and fusion on the feature vectors from each attention head; the attention calculation unit (301) includes a multi-head attention sub-unit (3011) for processing different feature vectors in parallel; the decision logic unit (302) is used to generate a control action instruction based on the fused information and send the instruction to the control action attention head (206); the closed-loop control unit (303) dynamically adjusts the injection volume or injection rate of the servo motor through PID control or adaptive control algorithm based on the real-time response signal of the mercury gas meter, so that the mercury gas meter responds quickly to reach a predetermined target and remains stable.

5. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 1, characterized in that: The intelligent dead volume compensation and prediction module (4) further comprises: A data collection unit (401) is used to collect historical calibration data, environmental parameters and system operating parameters; a model training unit (402) is used to train a dead volume prediction model using a machine learning algorithm; and a real-time prediction unit (403) is used to predict the residual amount of standard mercury gas in the gas path based on current parameters before each calibration, and send the prediction result to the dead volume prediction attention head (203).

6. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 2, characterized in that: The adaptive learning and calibration scheme optimization module (6) at least includes: A calibration data storage unit (601) is used to store historical calibration data and performance indicators; a performance indicator evaluation unit (602) is used to evaluate the linearity and repeatability of the calibration curve for each calibration; and a calibration parameter optimization unit (603) uses reinforcement learning or Bayesian optimization algorithms to adaptively optimize future calibration parameters based on historical performance and feed back optimization suggestions to the decision logic unit (302) of the central controller (3).

7. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 2, characterized in that: The user interface module (7) comprises at least: The real-time data display unit (701) is used to display real-time sensor data and mercury gas meter response signals; the calibration status display unit (702) is used to display the real-time status of the calibration process; and the attention weight display unit (703) is used to intuitively display the weight information of each attention head so that the user can understand the system decision process.

8. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 1, characterized in that: The data storage and management module (8) at least includes: A configuration storage unit (801) is used to store calibration parameter configurations; a real-time data storage unit (802) is used to store sensor data and mercury gas analyzer response data collected in real time; a result storage unit (803) is used to store calibration results; a log recording unit (804) is used to record system operation logs; and a historical data storage unit (805) is used to store historical calibration data and adaptive learning model parameters.

9. The automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to claim 1, characterized in that: The multi-sensor fusion module (5) at least includes: The data receiving unit (501) is used to receive data from temperature, humidity and air pressure sensors; the data fusion unit (502) fuses the data using a Kalman filter or a weighted average algorithm, and sends the fused data to the sensor data attention head (201).

10. An automatic sampling and injection control system for intelligent calibration of a mercury gas analyzer according to any one of claims 1 to 9, characterized in that: The following steps are involved: Real-time data acquisition step: using the data acquisition module (1) to collect environmental parameters, system status data and the response signal of the mercury gas meter to be calibrated in real time; Feature extraction and fusion steps: the sensor data attention head (201), instrument response attention head (202), dead volume prediction attention head (203), calibration workflow status attention head (204) and historical data attention head (205) in the attention head module (2) respectively extract features from the collected data, predicted dead volume, calibration status and historical data, and transmit the obtained feature vectors to the attention calculation unit (301) of the central controller (3) for weighted fusion; Intelligent decision-making and control steps: the decision logic unit (302) of the central controller (3) generates a control instruction based on the fused feature vector, and converts it into a control signal through the control action attention head (206) of the attention head module (2), driving the servo motor and the air valve to perform the automatic sampling and injection operation; Closed-loop feedback control step: the closed-loop control unit (303) of the central controller (3) monitors the response signal of the mercury meter provided by the data acquisition module (1) in real time, and dynamically adjusts the injection volume or injection rate using a PID or adaptive control algorithm to ensure that the mercury meter responds quickly and stably; Intelligent dead volume compensation step: the intelligent dead volume compensation and prediction module (4) predicts the residual amount of standard mercury gas in the gas path based on historical data and current parameters, and transmits the prediction result to the dead volume prediction attention head (203) of the attention head module (2); Adaptive optimization step: The adaptive learning and calibration scheme optimization module (6) optimizes future calibration parameters based on historical calibration data and performance indicators, and feeds back the optimization suggestions to the decision logic unit (302) of the central controller (3); Data storage and management step: Use the data storage and management module (8) to store and manage the collected raw data, fused feature vectors, control instructions, calibration results, system operation logs and adaptive learning model parameters.

Images