Coal chemical conversion process control method based on data driving

Abstract

The invention relates to the technical field of coal chemical industry process control, and discloses a coal chemical industry conversion process control method based on data driving. Temperature and pressure sensors are calibrated on line, independent collection channels are formed, full-field temperature and pressure state vectors are constructed, partition indexes and threshold values are established according to data distribution, and spatial differences are described in real time; generating a dimensionless error model by using a sample, introducing a comprehensive load constraint, constructing a dual Lagrange function, unifying a safety constraint and an optimization target, forming a temperature and pressure paired scheduling index, performing conversion to obtain an executable control increment, and differentially distributing the executable control increment to each measuring point according to a partition proportion, and closed-loop operation and dual factor dynamic switching are carried out in a fixed period, so that whole-process self-adaptive, safe and reliable cooperative control is realized.

Description

Technical Field

[0001] This invention relates to the field of coal chemical process control technology, specifically to a data-driven coal chemical conversion process control method. Background Technology

[0002] In coal chemical plants, shift processes or conversion technologies typically involve coal gas, which can be sourced from coke oven gas or coal gasification gas. These two types of gas differ in composition, pressure rating, dust and moisture content, and operating condition fluctuations. However, temperature and pressure control is crucial in the separation, purification, heat exchange, shift reactions, and connecting pipelines to ensure reaction conversion, equipment safety, and stable operation. Current engineering applications often employ single-loop control methods, such as separate temperature and pressure control loops, relying on fixed or segmented setpoints to maintain operating conditions within permissible ranges. When the plant load changes or the gas source switches, stability is often maintained by manually adjusting setpoints, switching valve positions, or altering heat exchange / throttling conditions. This type of control typically relies on a few key point measurements, making it difficult to comprehensively characterize the spatial distribution differences along the process flow. This can easily lead to localized overheating, localized overpressure, or abnormal pressure drops while the overall average remains normal.

[0003] Furthermore, existing technologies often employ uniform or fixed-weight allocation for multi-measuring-point control variable distribution. This means that after determining the overall adjustment direction, equal or pre-set proportions of adjustment are applied to the corresponding execution links of each measuring point, lacking a mechanism for differentiated allocation based on the degree of deviation of the measuring point, regional characteristics, and real-time operating status. Regarding the synergistic relationship between temperature and pressure, engineering practices often treat them as independent control variables. The lack of paired design and unified scheduling indicators for temperature and pressure regulation leads to contradictions such as temperature reaching the target but pressure approaching the upper limit, or pressure stabilizing but uneven temperature distribution during operating condition fluctuations. Simultaneously, existing technologies often achieve device safety load constraints through alarms, interlocks, or experience-based limits, often separating safety constraints from process accuracy targets. When the load approaches the upper limit, the control strategy tends to become conservative or rely on manual intervention, making it difficult to maintain operating accuracy while ensuring no over-limits are exceeded. At the data processing level, existing methods for the joint evaluation of variables with different dimensions, such as temperature and pressure, often rely on empirical proportional coefficients or fixed normalization, lacking a way to automatically form a scale by combining the distribution of field data. This makes them susceptible to outliers, sensor drift, or sampling noise. At the same time, the zoning of measuring points often relies on process experience or static area division, which is difficult to update dynamically with changes in gas source and load, resulting in a mismatch between zoning and control strategies.

[0004] Therefore, this case aims to propose a data-driven control method for coal chemical conversion processes. Based on the data-driven method, a scheduling and execution system that takes into account both process accuracy and load safety is formed by online calibration of temperature and pressure sensors, real-time state vector construction, partition index division, dimensionless error modeling, and the introduction of a dual Lagrangian structure with comprehensive load constraints. Summary of the Invention

[0005] This invention provides a data-driven control method for coal chemical conversion processes, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a data-driven coal chemical conversion process control method, comprising:

[0007] Temperature and pressure sensors were deployed and calibrated using a two-point method to form an independent voltage acquisition channel and calibration coefficient.

[0008] The temperature and pressure samples from each measuring point are combined to form a temperature and pressure state vector, and zoning indicators are established and the main zone and regulation zone are divided according to the zoning threshold.

[0009] Collect running samples, obtain the total dimensionless target value, and form a dimensionless error model;

[0010] By introducing comprehensive load constraints, a dimensionless Lagrangian function and dual factor are constructed.

[0011] Generate temperature and pressure scheduling indicators, set conversion factors, and obtain global temperature control increment and global pressure control increment.

[0012] The temperature and pressure control values ​​for each measuring point are allocated according to the ratio of the main zone to the regulation zone.

[0013] Perform temperature and pressure regulation and complete status updates within a fixed control cycle;

[0014] Store runtime data and perform dual factor switching updates.

[0015] Optionally, the deployment of temperature and pressure sensors and the use of a two-point calibration method to form independent voltage acquisition channels and calibration coefficients specifically include:

[0016] Multiple physical measurement points are arranged at equal intervals along the reaction pipeline and numbered sequentially.

[0017] At each physical measurement point, a temperature sensor and a pressure sensor are installed and connected to an independent voltage acquisition channel. The total number of temperature sensors and the total number of pressure sensors are the same as the total number of physical measurement points.

[0018] Temperature calibration: Select two sets of known temperatures and record the original voltage at each measuring point under the two sets of temperatures;

[0019] If the two voltages are the same, reselect the calibration point or check the connection.

[0020] If they are different, calculate the linear slope and linear intercept of the temperature channel at each measuring point using the two-point method, and establish a linear conversion relationship from voltage to temperature.

[0021] Pressure calibration: Select two sets of known pressures and record the original voltage at each measuring point under the two sets of pressures;

[0022] If the two voltages are the same, reselect the calibration point or check the connection.

[0023] If they are different, calculate the linear slope and linear intercept of the pressure channel at each measuring point using the two-point method to establish a linear conversion relationship from voltage to pressure.

[0024] During sampling, the voltage of the temperature channel is converted into the actual temperature and the voltage of the pressure channel is converted into the actual pressure according to their respective linear conversion relationships, thus forming the real-time temperature and real-time pressure of each measuring point.

[0025] Optionally, the combination of temperature and pressure sampling at each measuring point forms a temperature-pressure state vector, establishes zoning indicators, and divides the main zone and regulation zone according to zoning thresholds, specifically including:

[0026] At any sampling moment, the real-time temperature and real-time pressure of all measuring points are combined into a temperature and pressure state vector according to the measuring point number;

[0027] Calculate the global average temperature and global average pressure;

[0028] Based on the temperature and pressure scales, for each measuring point, the deviation of the temperature at each measuring point relative to the global average temperature and the deviation of the pressure at each measuring point relative to the global average pressure are calculated. After normalization according to the corresponding scale, the results are summed to generate a zoning index for each measuring point.

[0029] Within a 600-second time window after system startup, all measurement point zoning indicators are collected, and the median is calculated as the zoning threshold.

[0030] At any given time, if the zoning index of a certain measuring point is not greater than the zoning threshold, the measuring point is assigned to the main zone set; if the zoning index of a certain measuring point is greater than the zoning threshold, the measuring point is assigned to the adjustment zone set.

[0031] Optionally, the process of collecting running samples, obtaining the total dimensionless target value, and forming a dimensionless error model specifically includes:

[0032] Within a 600-second time window after system startup, temperature and pressure sample sets are collected.

[0033] The 5th and 95th percentiles of the temperature sample and the 5th and 95th percentiles of the pressure sample were obtained respectively.

[0034] Obtain the minimum resolvable temperature difference of the temperature sensor and the minimum resolvable pressure difference of the pressure sensor;

[0035] Construct temperature and pressure scales: use the quantile difference when the corresponding quantile difference is not zero, and use the minimum resolvable temperature difference or the minimum resolvable pressure difference when the corresponding quantile difference is zero.

[0036] For each measuring point, based on the target temperature and target pressure set by the process, the data is first normalized according to the corresponding scale, and then the temperature deviation and pressure deviation are squared and added together to obtain the bivariate dimensionless deviation of each measuring point.

[0037] The dimensionless deviation of each measuring point in the main region set is averaged to obtain the main region error; the dimensionless deviation of each measuring point in the adjustment region set is averaged to obtain the adjustment region error; the main region error and the adjustment region error are added together to obtain the total dimensionless target value.

[0038] Optionally, the introduction of comprehensive load constraints and the construction of a dimensionless Lagrangian function and dual factor specifically include:

[0039] Within a 600-second time window after system startup, temperature and pressure sample sets are collected.

[0040] The 5th and 95th percentiles of the temperature sample and the 5th and 95th percentiles of the pressure sample were obtained respectively.

[0041] Obtain the minimum resolvable temperature difference of the temperature sensor and the minimum resolvable pressure difference of the pressure sensor;

[0042] Construct temperature and pressure scales: use the quantile difference when the corresponding quantile difference is not zero, and use the minimum resolvable temperature difference or the minimum resolvable pressure difference when the corresponding quantile difference is zero.

[0043] For each measuring point, based on the target temperature and target pressure set by the process, the data is first normalized according to the corresponding scale, and then the temperature deviation and pressure deviation are squared and added together to obtain the bivariate dimensionless deviation of each measuring point.

[0044] The dimensionless deviation of each measuring point in the main region set is averaged to obtain the main region error; the dimensionless deviation of each measuring point in the adjustment region set is averaged to obtain the adjustment region error; the main region error and the adjustment region error are added together to obtain the total dimensionless target value.

[0045] Optionally, the generation of temperature and pressure scheduling indices, setting conversion factors, and obtaining global temperature control increments and global pressure control increments specifically includes:

[0046] Constructing temperature scheduling indicators: The temperature scheduling indicators are obtained by converting the difference between the global average temperature and the target temperature using a temperature scale, and by superimposing the standardized load deviation obtained by standardizing the comprehensive load according to the temperature load coefficient.

[0047] The pressure scheduling index is constructed by converting the difference between the global average pressure and the target pressure using a pressure scale, and then adding the standardized load deviation obtained by standardizing the comprehensive load and weighting it by the pressure load coefficient.

[0048] Set conversion factors corresponding to temperature and pressure scales. The conversion factor is equal to the reciprocal of the corresponding scale.

[0049] Based on the scheduling indicators and conversion coefficients, global temperature control increments and global pressure control increments are generated in the opposite direction to the scheduling indicators, and the corresponding analytical relationships are given.

[0050] Optionally, the allocation of temperature and pressure control values ​​at each measuring point according to the ratio of the main zone to the adjustment zone specifically includes:

[0051] Based on the number of measuring points in the main area set and the adjustment area set, the proportion of the main area and the proportion of the adjustment area are calculated respectively.

[0052] For each measuring point in the main area, the global temperature control increment is allocated to the temperature control quantity of each measuring point according to the proportion of the main area, and the global pressure control increment is allocated to the pressure control quantity of each measuring point according to the proportion of the main area.

[0053] For each measuring point within the regulation zone, the global temperature control increment is allocated to the temperature control quantity of each measuring point according to the regulation zone ratio, and the global pressure control increment is allocated to the pressure control quantity of each measuring point according to the regulation zone ratio.

[0054] Optionally, the step of performing temperature and pressure regulation and completing state updates within a fixed control cycle specifically includes:

[0055] Set the control cycle length to the preset number of seconds;

[0056] At the end of each control cycle, the current temperature and pressure at each measuring point are updated according to the corresponding temperature control and pressure control values ​​to obtain the temperature and pressure at the next moment.

[0057] Each temperature actuator and each pressure actuator operates according to the temperature control value and pressure control value of the corresponding measuring point, respectively.

[0058] Optionally, the storage of runtime data and the execution of dual factor switching updates specifically include:

[0059] At the end of each control cycle, store the operating data for that cycle, including the temperature and pressure of each measuring point, the main zone set and the regulation zone set, the global temperature control increment and the global pressure control increment, the temperature control quantity and the pressure control quantity of each measuring point, the comprehensive load value and the standardized load deviation, and the value of the dual factor.

[0060] Based on whether the current period's comprehensive load exceeds the comprehensive load safety limit, the dual factor for the next period is switched and updated between zero and one.

[0061] The present invention has the following beneficial effects:

[0062] 1. Online calibration is performed using a two-point method. The output voltages of temperature and pressure sensors under two sets of known operating conditions are collected and converted into actual physical quantities through a linear relationship. Simultaneously, a voltage acquisition channel and corresponding calibration coefficient are independently assigned to each sensor channel. This embeds the traditional calibration process into the online sampling workflow, dynamically updating calibration parameters without additional downtime or manual intervention, thereby improving calibration efficiency and data reliability. This method effectively solves the problems of sensor drift and long-term operational error accumulation in coal chemical plants, ensuring that all subsequent decision-making processes are based on reliable data.

[0063] 2. Real-time temperature and pressure information from all measuring points is concatenated into a state vector according to the measuring point number. A zoning index function is then introduced to normalize and weight the temperature and pressure deviations at each measuring point, thereby quantifying the regional differences of each measuring point relative to the global average level. The median threshold is used to divide the main zone and the regulation zone, making the zoning judgment robust and resistant to anomalies. It can dynamically identify hotspots or bottleneck areas in the process, helping the control system to prioritize the allocation of resources or regulation weights to the areas most in need of adjustment. Compared with traditional methods based solely on global averages or empirical thresholds, this data-driven zoning strategy is more suitable for nonlinearly coupled, multivariate coal chemical applications, ensuring the real-time performance and reliability of zoning judgment.

[0064] 3. Based on the temperature and pressure sample data from the initialization phase, quantiles and minimum resolution are calculated to construct temperature and pressure scales. Subsequently, the deviation at each measuring point is normalized and summed by squares to form a bivariate dimensionless deviation. The average deviations of the main region and the control region are calculated separately, and the two are added together to obtain the overall target value. A second-order judgment strategy based on quantiles and minimum resolution is adopted, ensuring that the scale construction reflects both the data distribution characteristics and extreme deviation values. Simultaneously, the error aggregation takes into account the dual needs of the main region and the control region. This achieves a smooth mapping from multivariate deviations to a single overall target, providing a unified quantitative standard for subsequent optimization. Compared to traditional error models that rely solely on a single variable or empirical weights, this scheme is more stable and robust in multivariate coupled systems, effectively avoiding error amplification or conflicts.

[0065] 4. Obtain temperature and pressure load coefficients from the equipment nameplate. Linearly combine the global average temperature and pressure according to the coefficients to obtain the comprehensive load. After further standardization, introduce a dual factor to form a dimensionless Lagrangian function. Integrate process safety load constraints and target control quantities through a dual Lagrangian structure to dynamically adjust the influence of the dual factor without exceeding limits. This ensures strict adherence to safety constraints while providing a real-time relaxation or tightening mechanism through the dual factor, enabling the control system to intelligently switch between multiple objectives. Compared to existing methods that only impose hard limits or soft constraints at the control end, this scheme mathematically couples safety constraints with target optimization, enhancing both system safety and control performance.

[0066] 5. Within the Lagrange framework, temperature and pressure scheduling indices are calculated separately. Then, by setting conversion factors (i.e., the reciprocal of the scale), the dimensionless scheduling indices are converted into corresponding temperature and pressure control increments. A clear analytical mapping relationship is established between the dimensionless indices and physical control quantities, making the decision output both executable and physically meaningful. This simplifies the conversion steps from the decision-making layer to the execution layer, avoiding complex nonlinear mappings or manual empirical adjustments; simultaneously, the precise conversion factors ensure that the magnitude of the control increments is consistent with process requirements. Compared with current methods that rely on empirical formulas or offline simulation mapping, this scheme achieves truly data-driven, online calibrable scheduling index conversion, improving control accuracy and adaptability.

[0067] 6. Calculate the partition ratio based on the number of measuring points in the main zone and the control zone, and allocate the global temperature and pressure control increments to each measuring point proportionally. Using the size of the dynamic partition set as the weight, the control strategy for each measuring point possesses adaptive group response attributes. It can adjust the allocation ratio in real time according to changes in the number of measuring points within a partition, avoiding waste of control resources or response delays caused by uniform allocation; simultaneously, the partition set automatically shrinks in the event of measuring point failure or anomalies, ensuring the robustness of the allocation mechanism. Compared with traditional static or experience-based proportional allocation, this scheme achieves adaptive resource scheduling based on real-time online partitioning, improving control efficiency and system fault tolerance.

[0068] 7. A fixed control cycle is established. At the end of each cycle, the status of each measuring point is updated according to the allocated temperature and pressure control values, and physical adjustment is achieved through the actuator. Data-driven control decisions are embedded into the bounded-cycle execution framework, achieving an organic integration of decision-making and execution. While ensuring real-time performance, the system possesses predictability and verifiability, facilitating timely evaluation and adjustment of execution results. Simultaneously, the introduction of a fixed cycle avoids excessively frequent or sparse control command issuance. Compared to existing event-triggered or fully timed triggering methods, this solution enhances system stability and maintainability while ensuring response speed.

[0069] 8. At the end of each control cycle, comprehensive data including measurement point status, partition sets, control variables, overall load, and dual factors are stored. The dual factor value is dynamically switched based on whether the overall load exceeds limits. This coupling of data storage with a safety control adaptive mechanism enables a complete record of the process's operational history and intelligent switching of dual factor states. This provides a high-quality data foundation for subsequent fault diagnosis, performance analysis, and model updates, while ensuring that the dual factor can intervene in the control strategy in real time outside the safety boundary. Compared to traditional approaches that only store key variables or do not record full data online, this solution's panoramic data and dual factor linkage update method provides valuable historical support and safety assurance for process optimization and improvement. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Example, refer to Figure 1 A data-driven control method for coal chemical conversion processes includes:

[0073] Temperature and pressure sensors were deployed and calibrated using a two-point method to form an independent voltage acquisition channel and calibration coefficient.

[0074] The temperature and pressure samples from each measuring point are combined to form a temperature and pressure state vector, and zoning indicators are established and the main zone and regulation zone are divided according to the zoning threshold.

[0075] Collect running samples, obtain the total dimensionless target value, and form a dimensionless error model;

[0076] By introducing comprehensive load constraints, a dimensionless Lagrangian function and dual factor are constructed.

[0077] Generate temperature and pressure scheduling indicators, set conversion factors, and obtain global temperature control increment and global pressure control increment.

[0078] The temperature and pressure control values ​​for each measuring point are allocated according to the ratio of the main zone to the regulation zone.

[0079] Perform temperature and pressure regulation and complete status updates within a fixed control cycle;

[0080] Store runtime data and perform dual factor switching updates.

[0081] First, by calibrating temperature and pressure sensors online and establishing independent voltage acquisition channels and calibration coefficients, the problem of sensor accuracy drift and signal error accumulation caused by long-term operation was solved, ensuring the reliability of measurement data from the source. Then, temperature and pressure data from all measuring points were combined into a unified state vector, and zoning indicators and thresholds were constructed based on data distribution to achieve real-time dynamic characterization of process space differences, avoiding over-adjustment caused by traditional control relying solely on average values ​​or empirical thresholds. Next, a dimensionless error model was generated by aggregating samples, aggregating multivariate deviations into structured target values, providing a unified quantitative standard for subsequent scheduling decisions. A comprehensive load constraint was introduced, and a dual Lagrangian function was constructed, integrating safety load constraints with optimization objectives, ensuring both process accuracy and safety without exceeding limits. Finally, scheduling indicators were generated based on the Lagrangian structure, and executable temperature and pressure control increments were obtained through conversion coefficients. The control quantities were then allocated to each measuring point according to the zoning ratio, forming an implementable and adaptive control output. The entire process is executed in a closed loop within a fixed period, dynamically switching the dual factor to ensure the system automatically converges to the optimal solution under different load conditions. A closed-loop system was constructed, encompassing the entire process from data acquisition to control execution. This system not only improves measurement and control accuracy but also ensures a real-time balance between intelligent decision-making and safety constraints. Compared with existing segmented optimization or static scheduling methods, it enhances the system's adaptability and reliability.

[0082] The temperature and pressure sensors are deployed and calibrated using a two-point method, forming an independent voltage acquisition channel and calibration coefficients, specifically including:

[0083] Multiple physical measurement points are arranged at equal intervals along the reaction pipeline and numbered sequentially.

[0084] At each physical measurement point, a temperature sensor and a pressure sensor are installed and connected to an independent voltage acquisition channel. The total number of temperature sensors and the total number of pressure sensors are the same as the total number of physical measurement points.

[0085] Temperature calibration: Select two sets of known temperatures and record the original voltage at each measuring point under the two sets of temperatures;

[0086] If the two voltages are the same, reselect the calibration point or check the connection.

[0087] If they are different, calculate the linear slope and linear intercept of the temperature channel at each measuring point using the two-point method, and establish a linear conversion relationship from voltage to temperature.

[0088] Pressure calibration: Select two sets of known pressures and record the original voltage at each measuring point under the two sets of pressures;

[0089] If the two voltages are the same, reselect the calibration point or check the connection.

[0090] If they are different, calculate the linear slope and linear intercept of the pressure channel at each measuring point using the two-point method to establish a linear conversion relationship from voltage to pressure.

[0091] During sampling, the voltage of the temperature channel is converted into the actual temperature and the voltage of the pressure channel is converted into the actual pressure according to their respective linear conversion relationships, thus forming the real-time temperature and real-time pressure of each measuring point.

[0092] Further specific implementation steps include:

[0093] In the controlled gas process section of the coal chemical conversion process, the gas source is coke oven gas or coal gasification gas; the controlled gas process section is a gas separation-conversion process section, including but not limited to separation, purification, heat exchange, conversion reaction and its upstream and downstream connecting pipe sections; reaction pipelines are equidistantly arranged along the reaction pipelines of the process section. There are 1 physical measurement point, numbered as follows: ;in, This represents the total number of physical measuring points laid out along the reaction pipeline; Number the physical measurement points;

[0094] At each physical measurement point, one temperature sensor and one pressure sensor are installed, each connected to an independent voltage acquisition channel; the total number of temperature sensors is denoted as . The total number of pressure sensors is recorded as ,satisfy ;

[0095] Temperature sensor calibration: Select two sets of known temperatures Record physical measurement points Voltage output of the temperature sensor at two sets of temperatures ;in, , These are the known calibration temperatures for the first and second groups of temperature channels, respectively. , Physical measuring points Temperature sensor at calibration temperature , The original voltage reading below;

[0096] like If so, reselect the calibration point or check the sensor connection;

[0097] If the two voltages are different, the linear calibration coefficient is calculated as follows: , ;in, For physical measurement points The voltage-temperature linear slope coefficient of the temperature sensor; For physical measurement points Linear calibration intercept of the temperature sensor;

[0098] Pressure sensor calibration: Select two sets of known pressures Record physical measurement points Voltage output of the pressure sensor under two pressure conditions ;in, The known calibration pressures are for the first and second groups of pressure channels, respectively. Physical measuring points Pressure sensor at calibration pressure , The original voltage reading below;

[0099] like If so, reselect the calibration point or check the sensor connection;

[0100] If the two voltages are different, the linear calibration coefficient is calculated as follows: , ;in, For physical measurement points The voltage and pressure linear slope coefficient of the pressure sensor; For physical measurement points Linear calibration intercept of the pressure sensor;

[0101] At any sampling time Physical measuring points The actual temperature and pressure at the location are calculated as follows: , ;in, For continuous time variables; For physical measurement points Temperature sensor at time The voltage output; For physical measurement points The pressure sensor is at all times The voltage output; For physical measurement points At the moment The actual temperature; For physical measurement points At the moment The actual pressure.

[0102] By arranging measuring points equidistantly along key reaction pipelines and independently connecting temperature and pressure sensor channels at each measuring point, a one-to-one acquisition structure is formed, effectively avoiding the problems of mutual interference or signal crosstalk between sensor channels. In the calibration process, two sets of known calibration points are used for real-time calibration. If the two voltage readings are consistent, a recalibration or connection check step is triggered, thereby solving the dead zone and blind spot that may occur during the calibration process. When the voltage response is normal, the linear slope and intercept are calculated to map the voltage signal into physical quantities. The entire process does not require offline shutdown or the use of a dedicated calibration station, realizing dynamic calibration in online continuous operation. In particular, at the sampling time, the original voltage is converted into actual temperature and pressure in real time, ensuring that the data relied upon by all subsequent logic and algorithm modules are reliable real-time values.

[0103] The combined temperature and pressure sampling at each measuring point forms a temperature and pressure state vector, establishes zoning indicators, and divides the main zone and regulation zone according to zoning thresholds. Specifically, this includes:

[0104] At any sampling moment, the real-time temperature and real-time pressure of all measuring points are combined into a temperature and pressure state vector according to the measuring point number;

[0105] Calculate the global average temperature and global average pressure;

[0106] Based on the temperature and pressure scales, for each measuring point, the deviation of the temperature at each measuring point relative to the global average temperature and the deviation of the pressure at each measuring point relative to the global average pressure are calculated. After normalization according to the corresponding scale, the results are summed to generate a zoning index for each measuring point.

[0107] Within a 600-second time window after system startup, all measurement point zoning indicators are collected, and the median is calculated as the zoning threshold.

[0108] At any given time, if the zoning index of a certain measuring point is not greater than the zoning threshold, the measuring point is assigned to the main zone set; if the zoning index of a certain measuring point is greater than the zoning threshold, the measuring point is assigned to the adjustment zone set.

[0109] Further specific implementation steps include:

[0110] At sampling time Combine the temperature and pressure data of all physical measuring points in order of their numbers to construct a temperature-pressure state vector: ;in, For at any time The state vector composed of the temperature and pressure at all physical measuring points;

[0111] The global average temperature and global average pressure are calculated separately as follows: , .

[0112] in, For at any time The arithmetic mean of the temperatures at all physical measurement points; For at any time The arithmetic mean of the pressure at all physical measuring points.

[0113] Constructing physical measurement points The partition indicator function is as follows: ;in, For physical measurement points At the moment The values ​​of the partition indicators; Temperature scale; As a pressure scale;

[0114] Time interval after system startup Within seconds, construct a set of partitioned indicators for all measurement points within that time period: And take the median as the partition threshold: ;in, The continuous time points during system startup; This is the partition threshold; To find the median for a finite set of samples;

[0115] If a time exists Make Then the physical measurement point Included in the main area physical measurement point set Otherwise, include them in the set of physical measurement points in the adjustment zone. .

[0116] First, the real-time temperature and pressure data of all measuring points are assembled into a unified state vector in numerical order, providing a foundation for multi-dimensional data fusion. Then, the global average value is calculated, and a partition index function is further constructed. The deviation of each measuring point is normalized and weighted and summed to transform it into a single quantitative index. The median of all indicators within a fixed time window after system startup is used as the partition threshold. The anti-anomaly characteristics of the median are cleverly utilized to solve the interference of extreme deviations or noise on the threshold calculation. Then, at any time, based on the relationship between the partition index and the threshold, the measuring points are dynamically divided into the main zone (small deviation) and the adjustment zone (large deviation), realizing real-time mapping of process spatial heterogeneity.

[0117] The process of collecting and running samples, obtaining the total dimensionless target value, and forming a dimensionless error model specifically includes:

[0118] Within a 600-second time window after system startup, temperature and pressure sample sets are collected.

[0119] The 5th and 95th percentiles of the temperature sample and the 5th and 95th percentiles of the pressure sample were obtained respectively.

[0120] Obtain the minimum resolvable temperature difference of the temperature sensor and the minimum resolvable pressure difference of the pressure sensor;

[0121] Construct temperature and pressure scales: use the quantile difference when the corresponding quantile difference is not zero, and use the minimum resolvable temperature difference or the minimum resolvable pressure difference when the corresponding quantile difference is zero.

[0122] For each measuring point, based on the target temperature and target pressure set by the process, the data is first normalized according to the corresponding scale, and then the temperature deviation and pressure deviation are squared and added together to obtain the bivariate dimensionless deviation of each measuring point.

[0123] The dimensionless deviation of each measuring point in the main region set is averaged to obtain the main region error; the dimensionless deviation of each measuring point in the adjustment region set is averaged to obtain the adjustment region error; the main region error and the adjustment region error are added together to obtain the total dimensionless target value.

[0124] Further specific implementation steps include:

[0125] During the initial operation phase time interval Within seconds, the temperature sample set was statistically analyzed. With pressure sample set ;

[0126] Temperature sample quantiles are denoted as , quantiles are denoted as ;

[0127] Pressure sample quantiles are denoted as , quantiles are denoted as ;

[0128] The minimum resolvable temperature difference of the temperature sensor is denoted as . ;

[0129] The minimum resolvable pressure difference of the pressure sensor is obtained, denoted as . ;

[0130] The temperature and pressure scales are constructed as follows: , .

[0131] Constructing physical measurement points The bivariate deviation function is as follows: ;in, The target temperature set for the process; The target pressure set for the process; For physical measurement points At any moment The dimensionless comprehensive deviation value;

[0132] The error functions for the main region and the adjustment region are constructed separately as follows: , .

[0133] in, For a moment Dimensionless deviation of all measuring points in the main area The average value; For a moment The average value of the dimensionless deviation of all measuring points within the adjustment zone; For set The number of elements in the middle; For set The number of elements in the middle.

[0134] The overall objective function of the system is set as the sum of the errors in the main region and the adjustment region, specifically as follows: ;in, For a moment The total dimensionless target value.

[0135] By collecting samples during the initial operation phase, statistically analyzing the quantile differences and minimum resolvable differences of temperature and pressure, and constructing scale parameters, the normalization process reflects the actual distribution characteristics of the data while avoiding the inability to classify due to zero bias. A bivariate dimensionless bias is formed by summing the squared biases, quantifying the overall bias of each measuring point relative to the process target. Subsequently, the average biases of the main zone and the control zone are calculated and summed to obtain the total dimensionless target value, providing a unified optimization index for subsequent scheduling.

[0136] The introduction of comprehensive load constraints and the construction of a dimensionless Lagrangian function and dual factor specifically include:

[0137] Within a 600-second time window after system startup, temperature and pressure sample sets are collected.

[0138] The 5th and 95th percentiles of the temperature sample and the 5th and 95th percentiles of the pressure sample were obtained respectively.

[0139] Obtain the minimum resolvable temperature difference of the temperature sensor and the minimum resolvable pressure difference of the pressure sensor;

[0140] Construct temperature and pressure scales: use the quantile difference when the corresponding quantile difference is not zero, and use the minimum resolvable temperature difference or the minimum resolvable pressure difference when the corresponding quantile difference is zero.

[0141] For each measuring point, based on the target temperature and target pressure set by the process, the data is first normalized according to the corresponding scale, and then the temperature deviation and pressure deviation are squared and added together to obtain the bivariate dimensionless deviation of each measuring point.

[0142] The dimensionless deviation of each measuring point in the main region set is averaged to obtain the main region error; the dimensionless deviation of each measuring point in the adjustment region set is averaged to obtain the adjustment region error; the main region error and the adjustment region error are added together to obtain the total dimensionless target value.

[0143] Further specific implementation steps include:

[0144] The temperature load coefficient is obtained from the equipment nameplate and denoted as . Obtain the pressure load factor, denoted as... ;

[0145] The thermal pressure combined load function is constructed as follows: ;in, For a moment The combined load value of hot pressing;

[0146] Obtain the maximum permissible comprehensive load safety limit for the equipment, denoted as ;

[0147] The load function is standardized to a dimensionless form, and a standardized load function is constructed as follows: ;in, For a moment Standardized load deviation;

[0148] By introducing a dual factor, a dimensionless Lagrangian function is constructed, specifically: ;in, For a moment The dual factor; For a moment The Lagrange function;

[0149] At system startup Dual factor The initial value is determined based on the relationship between the comprehensive load and the safety limit, specifically: ;in, To be at startup time The initial value of the dual factor.

[0150] First, the temperature and pressure load coefficients are obtained from the equipment nameplate. The global average temperature and pressure are then linearly combined to obtain the comprehensive load value. By comparing it with the equipment's rated upper limit and standardizing it, the dimensionless deviation is obtained, and the safe load is introduced into the target framework. Then, the safety constraint is added to the dimensionless target value using the dual factor coefficient to form a Lagrangian function. When the system starts, the dual factor value is initialized according to whether the load exceeds the limit, providing a basis for safety relaxation or tightening in subsequent algorithms.

[0151] The generation of temperature and pressure scheduling indices, setting conversion factors, and obtaining global temperature control increments and global pressure control increments specifically includes:

[0152] Constructing temperature scheduling indicators: The temperature scheduling indicators are obtained by converting the difference between the global average temperature and the target temperature using a temperature scale, and by superimposing the standardized load deviation obtained by standardizing the comprehensive load according to the temperature load coefficient.

[0153] The pressure scheduling index is constructed by converting the difference between the global average pressure and the target pressure using a pressure scale, and then adding the standardized load deviation obtained by standardizing the comprehensive load and weighting it by the pressure load coefficient.

[0154] Set conversion factors corresponding to temperature and pressure scales. The conversion factor is equal to the reciprocal of the corresponding scale.

[0155] Based on the scheduling indicators and conversion coefficients, global temperature control increments and global pressure control increments are generated in the opposite direction to the scheduling indicators, and the corresponding analytical relationships are given.

[0156] Further specific implementation steps include:

[0157] The temperature regulation index function and the pressure regulation index function are constructed respectively as follows: , ;in, For a moment Temperature control indicators; For a moment Pressure scheduling indicators;

[0158] Using temperature scale With pressure scale ,set up , ;in, It represents the reciprocal of the temperature scale and is used to convert dimensionless temperature scheduling indicators into temperature control quantities. It is the reciprocal of the pressure scale, used to convert dimensionless pressure scheduling indicators into pressure control quantities;

[0159] Let the temperature control variable be denoted as The pressure control variable is denoted as Analytical expressions for the temperature and pressure control quantities are constructed respectively, as follows: , ;in, For at any time Temperature control increments calculated from scheduling indicators; For at any time Pressure control increment.

[0160] By calculating the scheduling indices for temperature and pressure separately within the Lagrange framework, and converting the dimensionless indices into actual control increments based on pre-set conversion coefficients (i.e., scale reciprocals), a one-to-one mapping between decision quantities and physical quantities is ensured. At the same time, the standardized deviation of the comprehensive load is weighted and superimposed into the scheduling indices according to the load coefficient, so that safety constraints directly affect the magnitude of the control quantity, thus solving the problem that it is difficult to balance safety and accuracy in control increments under multi-objective environments.

[0161] The allocation of temperature and pressure control values ​​at each measuring point according to the ratio of the main zone to the adjustment zone specifically includes:

[0162] Based on the number of measuring points in the main area set and the adjustment area set, the proportion of the main area and the proportion of the adjustment area are calculated respectively.

[0163] For each measuring point in the main area, the global temperature control increment is allocated to the temperature control quantity of each measuring point according to the proportion of the main area, and the global pressure control increment is allocated to the pressure control quantity of each measuring point according to the proportion of the main area.

[0164] For each measuring point within the regulation zone, the global temperature control increment is allocated to the temperature control quantity of each measuring point according to the regulation zone ratio, and the global pressure control increment is allocated to the pressure control quantity of each measuring point according to the regulation zone ratio.

[0165] Further specific implementation steps include:

[0166] Based on the set of physical measurement points in the main area With the set of physical measurement points in the regulation zone The allocation ratio factors for the main area and the adjustment area are constructed separately, as follows: , ;in, For a moment The proportion of the number of measuring points in the main area to the total number of measuring points; For a moment The proportion of measuring points in the adjustment zone to the total number of measuring points;

[0167] For physical measurement points in the main area The allocated temperature control and pressure control values ​​are as follows: , ;in, , At time respectively For physical measurement points in the main area Temperature control increment and pressure control increment;

[0168] For physical measuring points within the adjustment zone The allocated temperature control and pressure control values ​​are as follows: , ;in, , At time respectively For physical measuring points within the regulation zone Temperature control increment and pressure control increment.

[0169] First, the partition ratio is calculated based on the number of measuring points in the main area and the regulation area. Then, the global temperature and pressure control increments are allocated to each measuring point according to the corresponding ratio, realizing a group adaptive response. When some measuring points are abnormal or out of the group, the size of the partition set changes automatically, and the allocation ratio is adjusted accordingly, thus ensuring that the fault point does not affect the overall allocation.

[0170] The process of performing temperature and pressure regulation and completing state updates within a fixed control cycle specifically includes:

[0171] Set the control cycle length to the preset number of seconds;

[0172] At the end of each control cycle, the current temperature and pressure at each measuring point are updated according to the corresponding temperature control and pressure control values ​​to obtain the temperature and pressure at the next moment.

[0173] Each temperature actuator and each pressure actuator operates according to the temperature control value and pressure control value of the corresponding measuring point, respectively.

[0174] Further specific implementation steps include:

[0175] Set the control cycle length to Second;

[0176] Physical measuring points At any moment The temperature and pressure at that time were updated as follows: , ;in, , Physical measuring points In the next moment Temperature and pressure;

[0177] Each temperature actuator and pressure actuator is based on its corresponding physical measuring point. of and The signal is used to coordinate the temperature and pressure fields of the gas (coke oven gas or coal gasification gas) process in the gas separation-conversion process section.

[0178] By setting a uniform cycle length, the real-time nature of control commands is ensured, and the system has a predictable execution rhythm. At the end of each cycle, the temperature and pressure of each measuring point are updated according to the control increment, and physical adjustments are implemented through the actuator, realizing a closed-loop feedback of data-driven decision-making.

[0179] The storage of runtime data and the execution of dual factor switching updates specifically include:

[0180] At the end of each control cycle, store the operating data for that cycle, including the temperature and pressure of each measuring point, the main zone set and the regulation zone set, the global temperature control increment and the global pressure control increment, the temperature control quantity and the pressure control quantity of each measuring point, the comprehensive load value and the standardized load deviation, and the value of the dual factor.

[0181] Based on whether the current period's comprehensive load exceeds the comprehensive load safety limit, the dual factor for the next period is switched and updated between zero and one.

[0182] Further specific implementation steps include:

[0183] At the end of each control cycle, the operational data for the current cycle is stored, including but not limited to: the temperature of all physical measuring points. With pressure The calculated sets of physical measuring points in the main area and the physical measuring points in the adjustment area. , Temperature control quantity and pressure control quantity , and control quantities at each measuring point , Current load function value With standardized load function value Current dual factor ;

[0184] Dual factor The update will be performed as follows: ;in, For the next moment The value of the dual factor depends only on the current time. Switching based on whether the load exceeds the limit.

[0185] At the end of each cycle, key data including measurement point status, partition results, control increment, comprehensive load deviation, and dual factor are stored, providing a complete data foundation for subsequent diagnosis, optimization, and model retraining; at the same time, the dual factor is switched only based on whether the current load exceeds the limit, realizing the dynamic intervention of safety constraints.

[0186] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0187] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A data-driven control method for coal chemical conversion processes, characterized in that, include: Temperature and pressure sensors were deployed and calibrated using a two-point method to form an independent voltage acquisition channel and calibration coefficient. Combine the temperature and pressure samples from each measuring point to form a temperature and pressure state vector, establish zoning indicators, and divide the main zone and regulation zone according to the zoning threshold; Collect running samples, obtain the total dimensionless target value, and form a dimensionless error model; By introducing comprehensive load constraints, a dimensionless Lagrangian function and dual factor are constructed. Generate temperature and pressure scheduling indicators, set conversion factors, and obtain global temperature control increment and global pressure control increment. The temperature and pressure control values ​​for each measuring point are allocated according to the ratio of the main zone to the regulation zone. Perform temperature and pressure regulation and complete status updates within a fixed control cycle; Store runtime data and perform dual factor switching updates.

2. The data-driven coal chemical conversion process control method according to claim 1, characterized in that, The temperature and pressure sensors are deployed and calibrated using a two-point method, forming an independent voltage acquisition channel and calibration coefficients, specifically including: Multiple physical measurement points are arranged at equal intervals along the reaction pipeline and numbered sequentially. At each physical measurement point, a temperature sensor and a pressure sensor are installed and connected to an independent voltage acquisition channel. The total number of temperature sensors and the total number of pressure sensors are the same as the total number of physical measurement points. Temperature calibration: Select two sets of known temperatures and record the original voltage at each measuring point under the two sets of temperatures; If the two voltages are the same, reselect the calibration point or check the connection. If they are different, calculate the linear slope and linear intercept of the temperature channel at each measuring point using the two-point method to establish a linear conversion relationship from voltage to temperature. Pressure calibration: Select two sets of known pressures and record the original voltage at each measuring point under the two sets of pressures; If the two voltages are the same, reselect the calibration point or check the connection. If they are different, calculate the linear slope and linear intercept of the pressure channel at each measuring point using the two-point method to establish a linear conversion relationship from voltage to pressure. During sampling, the voltage of the temperature channel is converted into the actual temperature and the voltage of the pressure channel is converted into the actual pressure according to their respective linear conversion relationships, thus forming the real-time temperature and real-time pressure of each measuring point.

3. The data-driven coal chemical conversion process control method according to claim 2, characterized in that, The combined temperature and pressure sampling at each measuring point forms a temperature and pressure state vector, establishes zoning indicators, and divides the main zone and regulation zone according to zoning thresholds. Specifically, this includes: At any sampling moment, the real-time temperature and real-time pressure of all measuring points are combined into a temperature and pressure state vector according to the measuring point number; Calculate the global average temperature and global average pressure; Based on the temperature and pressure scales, for each measuring point, the deviation of the temperature at each measuring point relative to the global average temperature and the deviation of the pressure at each measuring point relative to the global average pressure are calculated. After normalization according to the corresponding scale, the results are summed to generate a zoning index for each measuring point. Within a 600-second time window after system startup, all measurement point zoning indicators are collected, and the median is calculated as the zoning threshold. At any given time, if the zoning index of a certain measuring point is not greater than the zoning threshold, the measuring point is assigned to the main zone set; if the zoning index of a certain measuring point is greater than the zoning threshold, the measuring point is assigned to the adjustment zone set.

4. The data-driven coal chemical conversion process control method according to claim 3, characterized in that, The process of collecting and running samples, obtaining the total dimensionless target value, and forming a dimensionless error model specifically includes: Within a 600-second time window after system startup, temperature and pressure sample sets are collected. The 5th and 95th percentiles of the temperature sample and the 5th and 95th percentiles of the pressure sample were obtained respectively. Obtain the minimum resolvable temperature difference of the temperature sensor and the minimum resolvable pressure difference of the pressure sensor; Construct temperature and pressure scales: use the quantile difference when the corresponding quantile difference is not zero, and use the minimum resolvable temperature difference or the minimum resolvable pressure difference when the corresponding quantile difference is zero. For each measuring point, based on the target temperature and target pressure set by the process, the data is first normalized according to the corresponding scale, and then the temperature deviation and pressure deviation are squared and added together to obtain the bivariate dimensionless deviation of each measuring point. The dimensionless deviation of each measuring point in the main region set is averaged to obtain the main region error; the dimensionless deviation of each measuring point in the adjustment region set is averaged to obtain the adjustment region error; the main region error and the adjustment region error are added together to obtain the total dimensionless target value.

5. The data-driven coal chemical conversion process control method according to claim 4, characterized in that, The introduction of comprehensive load constraints and the construction of a dimensionless Lagrangian function and dual factor specifically include: Within a 600-second time window after system startup, temperature and pressure sample sets are collected. The 5th and 95th percentiles of the temperature sample and the 5th and 95th percentiles of the pressure sample were obtained respectively. Obtain the minimum resolvable temperature difference of the temperature sensor and the minimum resolvable pressure difference of the pressure sensor; Construct temperature and pressure scales: use the quantile difference when the corresponding quantile difference is not zero, and use the minimum resolvable temperature difference or the minimum resolvable pressure difference when the corresponding quantile difference is zero. For each measuring point, based on the target temperature and target pressure set by the process, the data is first normalized according to the corresponding scale, and then the temperature deviation and pressure deviation are squared and added together to obtain the bivariate dimensionless deviation of each measuring point. The dimensionless deviation of each measuring point in the main region set is averaged to obtain the main region error; the dimensionless deviation of each measuring point in the adjustment region set is averaged to obtain the adjustment region error; the main region error and the adjustment region error are added together to obtain the total dimensionless target value.

6. The data-driven coal chemical conversion process control method according to claim 5, characterized in that, The generation of temperature and pressure scheduling indices, setting conversion factors, and obtaining global temperature control increments and global pressure control increments specifically includes: Constructing temperature scheduling indicators: The temperature scheduling indicators are obtained by converting the difference between the global average temperature and the target temperature using a temperature scale, and by superimposing the standardized load deviation obtained by standardizing the comprehensive load according to the temperature load coefficient. The pressure scheduling index is constructed by converting the difference between the global average pressure and the target pressure using a pressure scale, and then adding the standardized load deviation obtained by standardizing the comprehensive load and weighting it by the pressure load coefficient. Set conversion factors corresponding to temperature and pressure scales. The conversion factor is equal to the reciprocal of the corresponding scale. Based on the scheduling indicators and conversion coefficients, global temperature control increments and global pressure control increments are generated in the opposite direction to the scheduling indicators, and the corresponding analytical relationships are given.

7. The data-driven coal chemical conversion process control method according to claim 6, characterized in that, The allocation of temperature and pressure control values ​​at each measuring point according to the ratio of the main zone to the adjustment zone specifically includes: Based on the number of measuring points in the main area set and the adjustment area set, the proportion of the main area and the proportion of the adjustment area are calculated respectively. For each measuring point in the main area, the global temperature control increment is allocated to the temperature control quantity of each measuring point according to the proportion of the main area, and the global pressure control increment is allocated to the pressure control quantity of each measuring point according to the proportion of the main area. For each measuring point within the regulation zone, the global temperature control increment is allocated to the temperature control quantity of each measuring point according to the regulation zone ratio, and the global pressure control increment is allocated to the pressure control quantity of each measuring point according to the regulation zone ratio.

8. The data-driven coal chemical conversion process control method according to claim 7, characterized in that, The process of performing temperature and pressure regulation and completing state updates within a fixed control cycle specifically includes: Set the control cycle length to the preset number of seconds; At the end of each control cycle, the current temperature and pressure at each measuring point are updated according to the corresponding temperature control and pressure control values ​​to obtain the temperature and pressure at the next moment. Each temperature actuator and each pressure actuator operates according to the temperature control value and pressure control value of the corresponding measuring point, respectively.

9. A data-driven coal chemical conversion process control method according to claim 8, characterized in that, The storage of runtime data and the execution of dual factor switching updates specifically include: At the end of each control cycle, store the operating data for that cycle, including the temperature and pressure of each measuring point, the main zone set and the regulation zone set, the global temperature control increment and the global pressure control increment, the temperature control quantity and the pressure control quantity of each measuring point, the comprehensive load value and the standardized load deviation, and the value of the dual factor. Based on whether the current period's comprehensive load exceeds the comprehensive load safety limit, the dual factor for the next period is switched and updated between zero and one.

Images