Adaptive adjustment control system and method for combustion experiment of NOx generation mechanism

Abstract

This invention relates to the field of combustion experiment technology, specifically to an adaptive adjustment control system and method for NOx formation mechanism combustion experiments. The method includes: inputting user-entered current experimental data, historical experimental data, a preset knowledge base, and an experimental plan; inputting historical experimental data, the experimental plan, and the preset knowledge base into a pre-constructed experimental element identification model, outputting key fuzzy variables, a fuzzy range set, control quantities, and fuzzy rules, and establishing an initial fuzzy rule library; inputting current experimental data and the experimental plan into a fuzzy optimization model, outputting a weight set and an activation sequence; determining the activated rules and the set of adjustment variables based on the activation sequence; and performing model demodeling based on the initial fuzzy rule library, weight set, and adjustment variable set to obtain the adjustment value of each adjustment variable. This invention uses fuzzy algorithms to handle uncertainties and combines machine learning models to optimize the rule library and model demodeling method, thereby improving the adjustment accuracy of experimental parameters and the reliability of experimental results.

Description

Technical Field

[0001] This invention relates to the field of combustion experiment technology, and more specifically, to an adaptive adjustment control system and method for NOx formation mechanism combustion experiments. Background Technology

[0002] With increasingly stringent environmental protection requirements, controlling nitrogen oxides (NOx) during combustion has become particularly important. Current technologies for NOx generation control largely rely on theoretical calculations and simple experiments, lacking a systematic and precise experimental control system. This makes it difficult to effectively simulate actual combustion environments, hindering accurate research into NOx formation mechanisms and impacting the development and application of pollutant reduction technologies.

[0003] A patent with publication number CN117192022A discloses an integrated control system for a high-voltage discharge ignition mechanism experimental device, relating to the field of high-voltage ignition equipment technology. The system includes an ignition test module, a PCL controller, a data acquisition module, and a timing control module. The ignition test module is used for remote control of the test circuit and experimental chamber, and remote operation of switches, phase selection switches, circuit breaker closing, and the opening and closing of the test sample. The data acquisition module is used to collect test combustion data from the experimental chamber. The PCL controller is used to calculate the combustion index based on the test combustion data and to assist in determining the valve opening of the intake valve based on the combustion index to adjust the atmosphere inside the experimental chamber and improve test safety. The timing control module is used to generate two mutually exclusive opening and closing signals from the high and low levels of a clock signal to control the test time. This invention allows for remote and precise control of the required test circuit and sample switching, resulting in highly accurate test parameters. The system has good integration, high reliability, and strong practicality.

[0004] Although the above techniques have achieved high accuracy in obtaining experimental parameters, there are still issues with handling uncertainties. All experiments involve uncertainties, and the ultimate goal of the experiment is to directly bypass these uncertainties and arrive at the experimental conclusions. Otherwise, the uncertainties will interfere with the final experimental conclusions. Furthermore, the lack of a dynamic adjustment of experimental parameters based on real-time data by the experimental control system may affect the reliability of the experiment. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an adaptive adjustment control method for NOx formation mechanism combustion experiments, the method comprising:

[0006] Input the current experimental data, historical experimental data, preset knowledge base and experimental plan entered by the user;

[0007] Input historical experimental data, experimental schemes, and preset knowledge bases into the pre-constructed experimental element recognition model, output key fuzzy variables, fuzzy range sets, control quantities, and fuzzy rules, and establish an initial fuzzy rule base;

[0008] Input the current experimental data and experimental plan into the fuzzy optimization model to output the weight set and activation sequence;

[0009] The activation rules and the set of adjustment variables are determined based on the activation sequence;

[0010] The model is solved based on the initial fuzzy rule base, weight set, and adjustment variable set to obtain the adjustment value of each adjustment variable;

[0011] The furnace is adaptively adjusted using the adjustment values ​​of each adjustment variable.

[0012] Furthermore, the method for constructing the experimental element recognition model includes:

[0013] Each set of historical experimental data, experimental plan, and preset knowledge base is converted into the form of the first feature vector;

[0014] The set of all first feature vectors is used as input to the experimental element recognition model. The experimental element recognition model outputs the predicted key fuzzy variables, fuzzy range sets, and fuzzy rules for each set of historical experimental data, experimental schemes, and preset knowledge base. The training objective is to minimize the sum of the first prediction accuracies of all key fuzzy variables, fuzzy range sets, and fuzzy rules. The model parameters of the experimental element recognition model are optimized using the gradient descent algorithm. The experimental element recognition model is constructed by obtaining the model parameters of the experimental element recognition model that minimizes the sum of the first prediction accuracies of all key fuzzy variables, fuzzy range sets, and fuzzy rules.

[0015] Furthermore, the combination of the key fuzzy variables, the fuzzy range set, and the fuzzy rules constitutes the initial fuzzy rule base; the initial fuzzy rule base is a set of linguistic fuzzy definition rules.

[0016] Furthermore, the method for constructing the fuzzy optimization model includes:

[0017] Each set of historical experimental data and experimental scheme is converted into the form of a second feature vector;

[0018] The set of all second feature vectors is used as the input to the fuzzy optimization model. The fuzzy optimization model outputs the weight set and activation sequence predicted for each set of historical experimental data and experimental scheme. The training objective is to minimize the sum of the second prediction accuracies of all predicted weight sets and activation sequences. The model parameters of the fuzzy optimization model are optimized by the gradient descent algorithm. The model parameters of the fuzzy optimization model with the minimum sum of the second prediction accuracies of all predicted weight sets and activation sequences are used to construct the fuzzy optimization model.

[0019] Furthermore, the weight set is the set of weights in the weighted average method, that is, the set of membership degrees; the activation sequence is the sequence composed of the activated rules and the order of the activated rules; the demodulation calculation method adopts the weighted average method, and determines the adjustment value of the adjustment variable to be calculated according to the set of adjustment variables.

[0020] The adaptive adjustment control method for NOx formation mechanism combustion experiments also includes:

[0021] Input historical experimental data and experimental design into a pre-built window decision model, and output the size and number of time windows;

[0022] The adjustment values ​​are calibrated and delayed compensation is performed based on the size and number of time windows.

[0023] Perform stability testing; if the test passes, calculate the adjustment duration and obtain the adjustment duration.

[0024] The NOx generation mechanism experimental platform receives the adjustment duration and continuously adjusts the adjustment variable according to the adjustment value within the adjustment duration until the adjustment duration ends.

[0025] Furthermore, the method for constructing the window decision model includes:

[0026] Historical experimental data and experimental schemes are used as analysis data. Time windows corresponding to group c of analysis data are collected in advance, and the analysis data and corresponding time windows are converted into a set of third feature vectors.

[0027] Each group of third feature vectors is used as input to the window decision model, which outputs a set of prediction time windows corresponding to each set of analysis data, with the goal of minimizing the sum of prediction errors of all analysis data. The model parameters of the window decision model are optimized through natural heuristic optimization, and the model parameters of the window decision model with the minimum sum of prediction errors of all analysis data are obtained to construct the window decision model. The window decision model is a recurrent neural network model.

[0028] Furthermore, the data calibration method is the moving average method, which obtains the smoothed adjustment value by calculating the average of the original adjustment values ​​within a time window.

[0029] Furthermore, the stability test method includes: starting from the smoothed adjustment value, gradually checking the rate of change between adjacent data points; when the rate of change is consistently lower than a set threshold ϵ, the smoothed adjustment value is considered to have passed the test and entered a stable state, where ϵ is a preset value; the calculation method for the adjustment duration includes: in the sequence of multiple smoothed adjustment values, recording the time point v at which the first adjustment value reaches a stable state, then the adjustment duration is v + ϵ. , Used as a unit of time.

[0030] This invention also discloses an adaptive adjustment control system for NOx formation mechanism combustion experiments, the system comprising:

[0031] Collection module: used to input current experimental data, historical experimental data, preset knowledge base and experimental plan entered by the user;

[0032] Fuzzy New Creation Module: Used to input historical experimental data, experimental plans and preset knowledge bases into the pre-built experimental element recognition model, output key fuzzy variables, fuzzy range set, control quantities and fuzzy rules, and establish an initial fuzzy rule base;

[0033] Fuzzy optimization module: Used to input the current experimental data and experimental scheme into the fuzzy optimization model to output the weight set and activation sequence;

[0034] Activation confirmation module: used to determine the activated rules and adjustment variable set based on the activation sequence;

[0035] Model decomposition module: used to perform model decomposition based on the initial fuzzy rule base, weight set and adjustment variable set, and obtain the adjustment value of each adjustment variable;

[0036] Adjustment module: Used to adaptively adjust the furnace using the adjustment values ​​of each adjustment variable;

[0037] Window confirmation module: used to input historical experimental data and experimental plans into the pre-built window decision model, and output the size and number of time windows;

[0038] Data calibration module: used to perform data calibration and delay compensation on the adjustment values ​​according to the size and number of time windows;

[0039] Stability testing module: Used to perform stability tests. If the test passes, the adjustment duration is calculated and obtained.

[0040] Adjustment and optimization module: Receives the adjustment duration. The NOx generation mechanism experimental platform continuously adjusts the adjustment variables according to the adjustment value within the adjustment duration until the adjustment duration ends.

[0041] The technical effects and advantages of the adaptive adjustment control system and method for NOx formation mechanism combustion experiments provided by this invention are as follows:

[0042] Technical effects:

[0043] This invention addresses various uncertainties in NOx generation mechanism experiments by introducing fuzzy algorithms and machine learning techniques, thereby improving the precision of experimental control. By optimizing the fuzzy rule base and demodeling method using machine learning models, the experimental control system can dynamically adjust experimental parameters based on real-time data, ensuring the reliability and accuracy of the experiment. Furthermore, time calibration and data smoothing techniques are used to calibrate and compensate for delays in the experimental data, effectively reducing errors caused by data latency and ensuring the accuracy and continuity of the experimental data. The system's high degree of automation reduces human intervention and operational errors, improving experimental efficiency and accuracy.

[0044] advantage:

[0045] This invention addresses uncertainties through fuzzy algorithms, combines machine learning models with optimized rule bases and demodeling methods to improve the accuracy of experimental parameter adjustments and the reliability of experimental results. Data calibration and delay compensation techniques effectively reduce error accumulation, ensuring the continuity and accuracy of experimental data. Dynamic adjustment and real-time optimization technologies enable the system to adapt to various changes during experiments, enhancing its robustness. High automation reduces reliance on manual operation, saving labor costs and improving experimental efficiency. Attached Figure Description

[0046] Figure 1 This is a flowchart of the adaptive adjustment control method for the NOx formation mechanism combustion experiment in Example 1;

[0047] Figure 2 This is a flowchart of the adaptive adjustment control method for the NOx formation mechanism combustion experiment in Example 2;

[0048] Figure 3 This is a schematic diagram of the adaptive adjustment control system connection for the NOx formation mechanism combustion experiment in Example 3. Detailed Implementation

[0049] 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.

[0050] Example 1

[0051] Please see Figure 1 As shown in this embodiment, the adaptive adjustment control method for the NOx formation mechanism combustion experiment includes:

[0052] Input the current experimental data, historical experimental data, preset knowledge base and experimental plan entered by the user;

[0053] Input historical experimental data, experimental schemes, and preset knowledge bases into the pre-constructed experimental element recognition model, output key fuzzy variables, fuzzy range sets, control quantities, and fuzzy rules, and establish an initial fuzzy rule base;

[0054] Input the current experimental data and experimental plan into the fuzzy optimization model to output the weight set and activation sequence;

[0055] The activation rules and the set of adjustment variables are determined based on the activation sequence;

[0056] The model is solved based on the initial fuzzy rule base, weight set, and adjustment variable set to obtain the adjustment value of each adjustment variable;

[0057] The furnace is adaptively adjusted using the adjustment values ​​of each adjustment variable.

[0058] The current experimental data refers to the real-time experimental data of the current experiment, while the historical experimental data refers to the experimental data recorded in the experimental log. When an experiment is completed, the experimental data is recorded in the experimental log. Both the current experimental data and the historical experimental data include NOx generation (unit: mg / m³), excess air coefficient (unitless, typically ranging from 0.8 to 1.5), temperature (unit: degrees Celsius), and reducing gas concentration (unit: % volume ratio, such as CO or H2). The NOx generation, reducing gas concentration, and excess air coefficient are obtained from the flue gas sampling device at the furnace outlet, and the temperature is obtained from the temperature sensor.

[0059] The experimental design involves manually set variable control. For example, before the experiment begins, technicians design the experimental plan, as follows:

[0060] Experiment 1: The effect of the concentration of reducing gas on the amount of NOx generated under the same conditions.

[0061] Experiment 2: The effect of temperature on NOx formation under the same conditions.

[0062] The preset knowledge base can be downloaded from the Internet or manually set by technical personnel, for example:

[0063] The results show that a high-temperature, strongly reducing atmosphere can reduce NOx formation.

[0064] NOx formation is moderate when the temperature is between 800-1000℃; however, NOx levels will increase significantly when the temperature exceeds 1000℃ and the space velocity is too high.

[0065] When a certain concentration of reducing additives is introduced, NOx generation can be effectively controlled.

[0066] In low-space-velocity environments, increasing the catalytic combustion cycle (CC cycle) can improve combustion efficiency while reducing NOx production. In this cycle, a catalyst is used to promote the combustion reaction, enabling the fuel to burn completely at a lower temperature.

[0067] In one experiment, a novel flame cooling technology was used, which demonstrated how to lower the temperature and achieve ideal results.

[0068] The method for constructing the experimental element recognition model includes:

[0069] Each set of historical experimental data, experimental plan, and preset knowledge base is converted into the form of a first feature vector.

[0070] The set of all first feature vectors is used as input to the experimental element recognition model. The experimental element recognition model outputs the predicted key fuzzy variables, fuzzy range sets, and fuzzy rules for each set of historical experimental data, experimental schemes, and a preset knowledge base. The training objective is to minimize the sum of the first prediction accuracies of all key fuzzy variables, fuzzy range sets, and fuzzy rules. The model parameters of the experimental element recognition model are optimized using a gradient descent algorithm. The experimental element recognition model is constructed by obtaining the model parameters of the experimental element recognition model that minimizes the sum of the first prediction accuracies of all key fuzzy variables, fuzzy range sets, and fuzzy rules. The first prediction accuracy is determined by the sum of the predicted key fuzzy variables, fuzzy range sets, and fuzzy rules corresponding to the e-th set of historical experimental data, experimental schemes, and the preset knowledge base, and the sum of the predicted key fuzzy variables, fuzzy range sets, and fuzzy rules for the e-th set of historical experimental data, experimental schemes, and the preset knowledge base. The results are obtained by calculating the squared difference of historical experimental data, experimental schemes, and the corresponding actual key fuzzy variables, fuzzy range sets, and fuzzy rules in the preset knowledge base; such as the first prediction accuracy. ,in, For the highest prediction accuracy, For the e-th group of historical experimental data, experimental scheme, and pre-set knowledge base, the key fuzzy variables, fuzzy range set, and fuzzy rules for prediction are: For the first The actual key fuzzy variables, fuzzy range sets, and fuzzy rules corresponding to the set of historical experimental data, experimental schemes, and preset knowledge base.

[0071] The combination of the key fuzzy variables, the fuzzy range set, and the fuzzy rules constitutes the initial fuzzy rule base; the initial fuzzy rule base is a set of linguistic fuzzy definition rules, for example:

[0072] Adjust the concentration of reducing gas (such as oxygen): If the temperature is "high" and the oxygen concentration is "low" and the excess air coefficient is "low", then the concentration of reducing gas needs to be increased by 10%.

[0073] If the temperature is "medium", the concentration of reducing gas is "high", and the excess air coefficient is "medium", then the concentration of reducing gas needs to be reduced and the temperature lowered by 50°C.

[0074] Among them, "concentration of reducing gas", "temperature" and "oxygen concentration" are key fuzzy variables, "low", "medium" and "high" are fuzzy range sets, "if '' then increase ''" and "if '' then decrease ''" are fuzzy rules, and "decrease temperature by 50℃" and "increase concentration of reducing gas by 10%" are control quantities.

[0075] The terms "high," "medium," and "low" in the above text were preset by technicians based on actual experimental conditions and are not specifically defined here.

[0076] Example: The specific process of establishing and initializing the fuzzy rule base is as follows:

[0077] 1. Input data

[0078] (1) Historical experimental data

[0079] Historical experimental data is derived from experimental logs, recording partial results from the past 10 combustion experiments. The following is sample data (simplified version, assuming 5 parameters are recorded for each experiment).

[0080] Experiment number NOx generation (unit: mg / m³) Temperature (°C) Reducing gas concentration (%) Excess air coefficient 1 600 950 4 1.1 2 450 850 6 1.0 3 700 1050 2 1.4 4 500 900 5 1.2

[0081] (2) Experimental scheme

[0082] The experimental design specifies the control targets for variables set by the technicians, for example:

[0083] Experimental objective: To investigate the effects of temperature and reducing gas concentration on NOx formation.

[0084] Controlled variables: Adjust temperature (T) and reducing gas concentration (R) to keep the excess air coefficient (A) relatively stable. "Stable" means that the preset value is controlled within the allowable error range.

[0085] (3) Pre-set knowledge base

[0086] The pre-defined knowledge base contains combustion experiment patterns extracted from literature or the experience of technical personnel, for example:

[0087] "High temperatures (>1000℃) and low concentrations of reducing gases (<3%) can lead to a significant increase in NOx formation."

[0088] "Increasing the concentration of reducing gases (>5%) at medium temperatures (800-1000℃) can effectively reduce NOx formation."

[0089] "When the excess air coefficient is between 1.0 and 1.3, the combustion efficiency is relatively high and the NOx generation is moderate."

[0090] 2. Experimental Element Recognition Model Processing Procedure

[0091] The experimental element recognition model is a machine learning-based model (such as a neural network or decision tree) that extracts key fuzzy variables, fuzzy range sets, and fuzzy rules from the input data. Assuming the model has been trained on historical data, the following are the processing steps:

[0092] (1) Data preprocessing: Converting historical experimental data, experimental plans, and pre-set knowledge bases into feature vectors. For example:

[0093] The data from Experiment 1 were converted into a vector: [600, 950, 4, 1.1, 55].

[0094] Preset knowledge base rules are converted into structured text input.

[0095] (2) Model training objective: Minimize the error of the key fuzzy variables, fuzzy range set and fuzzy rules in the prediction (compared with the actual experimental law).

[0096] (3) Model output: Based on the input data, identify key variables and generate fuzzy rules.

[0097] 3. Output Results

[0098] After model processing, the following output is given:

[0099] (1) Key fuzzy variables

[0100] Temperature (T); reducing gas concentration (R); excess air coefficient (A).

[0101] (2) Fuzzy range set

[0102] Based on historical data and a pre-set knowledge base, the model defines a fuzzy range:

[0103] Temperature (T): Low: ≤800℃, Medium: 800-1000℃, High: >1000℃.

[0104] Reducing gas concentration (R): Low: ≤3%, Medium: 3-7%, High: >7%.

[0105] Excess air coefficient (A): Low: ≤1.0, Medium: 1.0-1.3, High: >1.3.

[0106] (3) Controlled amount

[0107] The control amount is a suggested adjustment based on the experimental objective, for example:

[0108] Temperature adjustment: ±30℃, ±50℃.

[0109] Adjustment of reducing gas concentration: ±1%, ±2%.

[0110] (4) Fuzzy rules

[0111] The model generates the following fuzzy rules based on the input data and the knowledge base:

[0112] If T is "high", R is "low" and A is "medium", then the adjustment amount of R is increased by +2%.

[0113] If T is "medium", R is "medium", and A is "medium", then the adjustment amount for reducing T is -30℃.

[0114] If T is "high", R is "medium" and A is "high", then the adjustment amount for reducing T is -50℃.

[0115] If T is "medium", R is "high" and A is "low", then the adjustment amount of R is increased by +1%.

[0116] 4. Establish an initial fuzzy rule base

[0117] The above outputs are integrated into an initial fuzzy rule base, which is a set of linguistic fuzzy definition rules. The following is a specific example:

[0118] Rule 1: If the temperature is "high" (>1000℃), the reducing gas concentration is "low" (≤3%), and the excess air coefficient is "medium" (1.0-1.3), then increase the reducing gas concentration by 2%.

[0119] Rule 2: If the temperature is “medium” (800-1000℃), the concentration of reducing gas is “medium” (3-7%), and the excess air coefficient is “medium” (1.0-1.3), then reduce the temperature by 30°C.

[0120] Rule 3: If the temperature is "high" (>1000°C), the concentration of reducing gas is "medium" (3-7%), and the excess air coefficient is "high" (>1.3), then reduce the temperature by 50°C.

[0121] Rule 4: If the temperature is "medium" (800-1000°C), the concentration of reducing gas is "high" (>7%), and the excess air coefficient is "low" (≤1.0), then increase the concentration of reducing gas by 1%.

[0122] 5. Verification

[0123] Take Experiment 4 (T = 900°C, R = 5%, A = 1.2, NOx = 500 mg / m³) in the historical experimental data as an example:

[0124] Matching rule: Rule 2 (T = "medium", R = "medium", A = "medium").

[0125] Regulation result: Reduce the temperature by 30°C, and adjust T from 900°C to 870°C.

[0126] Expected effect: According to the historical trend, the NOx generation amount is expected to be reduced to about 460 mg / m³.

[0127] The construction method of the fuzzy optimization model includes:

[0128] Convert each set of historical experimental data and experimental schemes into the form of a second feature vector.

[0129] Take the set of all second feature vectors as the input of the fuzzy optimization model. The fuzzy optimization model takes the weight set and activation sequence predicted for each set of historical experimental data and experimental schemes as the output, and takes minimizing the sum of the second prediction accuracies of all predicted weight sets and activation sequences as the training objective. Optimize the model parameters of the fuzzy optimization model through the gradient descent algorithm, and obtain the model parameters of the fuzzy optimization model with the minimum sum of the second prediction accuracies of all predicted weight sets and activation sequences to build the fuzzy optimization model; where the second prediction accuracy is calculated from the square difference between the predicted weight set and activation sequence corresponding to the gth set of historical experimental data and experimental schemes, and the actual weight set and activation sequence corresponding to the gth set of historical experimental data and experimental schemes; like the first prediction accuracy , where is the second prediction accuracy, is the predicted weight set and activation sequence corresponding to the gth set of historical experimental data and experimental schemes, is the actual weight set and activation sequence corresponding to the gth set of historical experimental data and experimental schemes.

[0130] The weight set is the set of weights in the weighted average method, i.e., the set of membership degrees; the activation sequence is the sequence of activated rules and the order of activated rules; in the subsequent modeling, the adjustment variables are the key fuzzy variables that need to be controlled, and the set of adjustment variables is the set of all adjustment variables.

[0131] It is particularly important to note that the demodeling process involves an order and priority. Each demodeling operation yields an adjustment value for the adjustment variables that need to be calculated, based on the set of adjustment variables. Since this invention calculates the adjustment values ​​on a computer platform and not all rules are used, an activation code in a programming language is required to activate the rule program, thereby enabling the demodeling operation.

[0132] The calculation method for the model decomposition adopts the weighted average method. Based on the set of adjustment variables, the adjustment values ​​of the adjustment variables to be calculated are determined, such as the adjustment values. The calculation formula is:

[0133] .

[0134] In the formula, For the first The membership degree of each activated rule. For the first One activated rule control quantity; The activated rule number; The total number of rules that have been activated.

[0135] It is particularly important to note that the calculation of the unmodulation remains applicable even when the same adjustment variable appears in multiple activated rules, and does not require repeated calculation. This is only used as an expression for calculating adjustment values ​​and is not unique, because there may be multiple adjustment variables in the activated rule at the same time. Therefore, the formula for calculating the adjustment values ​​of all adjustment variables is still universal.

[0136] To facilitate understanding of the solution in this embodiment, the following example is provided:

[0137] 1. Input data

[0138] (1) Current experimental data

[0139] Assuming the experiment is currently underway, the real-time data collected is as follows:

[0140] NOx generation: 520 mg / m³; Temperature (T): 920℃; Reducing gas concentration (R): 4.5%; Excess air coefficient (A): 1.15.

[0141] (2) Experimental scheme

[0142] The experimental design is as follows:

[0143] Objective: To reduce NOx generation to below 450 mg / m³.

[0144] Controlled variables: Adjust temperature (T) and reducing gas concentration (R) to keep the excess air coefficient (A) stable.

[0145] 2. Initialize the fuzzy rule base (refer to the example above)

[0146] Rule 1: If T is "high" (>1000℃), R is "low" (≤3%), and A is "medium", then the adjustment amount of R is increased by +2%.

[0147] Rule 2: If T is "medium" (800-1000℃), R is "medium" (3-7%), and A is "medium" (1.0-1.3), then the adjustment amount of T should be reduced by -30℃.

[0148] Rule 3: If T is "high" (>1000℃), R is "medium" (3-7%), and A is "high" (>1.3), then the adjustment amount for reducing T is -50℃.

[0149] Rule 4: If T is “Medium” (800-1000℃), R is “High” (>7%), and A is “Low” (≤1.0), then the adjustment amount of R is increased by +1%.

[0150] 3. Fuzzy optimization model output: weight set and activation sequence

[0151] Input the current experimental data (T=920℃, R=4.5%, A=1.15) and experimental scheme into the fuzzy optimization model, and output the weight set and activation sequence.

[0152] Fuzzy range matching:

[0153] T=920℃→“Medium” (800-1000℃);

[0154] R=4.5% → "Medium" (3-7%)

[0155] A=1.15 → "Medium" (1.0-1.3);

[0156] Weight set (membership degrees are shown in a simple assignment example; in practice, membership functions, such as triangular membership functions, can be used for calculation):

[0157] Rule 1: Membership degree 0.1 (T is not "high", R is not "low", A is "medium");

[0158] Rule 2: Membership degree 0.9 (T is "medium", R is "medium", A is "medium");

[0159] Rule 3: Membership degree 0.2 (T is not "high", R is "medium", A is not "high");

[0160] Rule 4: Membership degree 0.3 (T is "medium", R is not "high", A is not "low");

[0161] Activation sequence:

[0162] Activated rule: Rule 2 (highest membership, highest priority).

[0163] Sequence: Only rule 2 is activated (other rules have low membership and have not reached the activation threshold, assuming the threshold is 0.5).

[0164] 4. Determine the activation rules and adjustment variable set based on the activation sequence.

[0165] Activated rule: Rule 2 (If T is "medium", R is "medium", and A is "medium", then reduce the regulation amount of T by -30°C).

[0166] Adjustment variable set: {T} (Rule 2 only involves the adjustment of temperature T).

[0167] 5. Demodel based on the initialized fuzzy rule base, weight set, and adjustment variable set.

[0168] The model decomposition uses a weighted average method to calculate the adjustment value for each adjustment variable:

[0169] Adjustment variable: Temperature (T);

[0170] Regulation quantity and membership degree:

[0171] Rule 2: Controlled temperature -30℃, membership degree 0.9

[0172] No other rules are activated; only rule 2 is used.

[0173] Adjustment value calculation:

[0174] ℃.

[0175] Result: The temperature adjustment value was reduced by 30℃.

[0176] 6. Adaptive adjustment of the furnace using adjustment values.

[0177] Adjustment: The furnace temperature is reduced from 920℃ to 890℃ through a cooling device.

[0178] Adjusted data (assuming real-time data collection after adjustment):

[0179] Temperature (T): 890℃

[0180] NOx generation: 470 mg / m³ (reduced by 50 mg / m³, close to the target).

[0181] Adaptive validation: NOx generation was reduced from 520 mg / m³ to 470 mg / m³, showing significant effect; if the target (450 mg / m³) is not reached, the next round of optimization can proceed.

[0182] Multivariate Adjustment Extended Example

[0183] If both T and R need to be adjusted simultaneously, assuming rule 2 is activated and another rule (such as rule 1 with slight modifications), the process is as follows:

[0184] Modify rule 1: If T is "medium", R is "medium", and A is "medium", then increase R+1.

[0185] Activation sequence: Rule 2 (membership 0.9), Rule 1 (membership 0.8).

[0186] Order: Rule 2 → Rule 1.

[0187] Adjust the variable set: {T, R}.

[0188] Demolding:

[0189] ℃.

[0190] .

[0191] Adjustments: T was reduced from 920℃ to 890℃, and R was increased from 4.5% to 5.5%.

[0192] Example 2

[0193] like Figure 2As shown, this embodiment is a further improvement on the design of Embodiment 1. The difference is that the adaptive adjustment control method for the NOx formation mechanism combustion experiment provided in Embodiment 1, in actual operation, determines the adjustment value of the adjustment variable to be calculated based on the set of adjustment variables. The adjustment value cannot be reached instantly, resulting in time consumption. Furthermore, due to the closed nature of the combustion experiment, it is difficult for the outside world to directly observe whether the adjustment of the adjustment variable has been completed. It can only be measured by external instruments. Therefore, there will be a time difference between the technicians and the furnace. The internal adjustment has not yet been completed, but the instruments on the NOx formation mechanism experimental platform have already shown that the adjustment value has been reached, and the next experiment can be carried out. For example, adjusting the concentration of reducing gas by releasing reducing gas into the furnace. When the flue gas sampling device at the furnace outlet detects that the concentration of reducing gas has reached the standard, the furnace may not actually have fully met the experimental requirements. This is because the spread of gas takes time, and a certain amount of time is needed for the concentration of reducing gas in each space inside to fully reach the standard. When the technicians substitute the experimental data of the entire furnace into the experiment, errors will occur. Experiments are generally continuous experiments, and errors will accumulate after the next experiment. When the error accumulates to a certain amount, it will eventually lead to data distortion. Based on this, the adaptive adjustment control method for the NOx formation mechanism combustion experiment provided in this example also includes:

[0194] Input historical experimental data and experimental design into a pre-built window decision model, and output the size and number of time windows;

[0195] The adjustment values ​​are calibrated and delayed compensation is performed based on the size and number of time windows.

[0196] Perform stability testing; if the test passes, calculate the adjustment duration and obtain the adjustment duration.

[0197] The NOx generation mechanism experimental platform receives the adjustment duration and continuously adjusts the adjustment variable according to the adjustment value within the adjustment duration until the adjustment duration ends.

[0198] The method for constructing the window decision model includes:

[0199] Historical experimental data and experimental schemes are used as analysis data. Time windows corresponding to group c of analysis data are collected in advance, and the analysis data and corresponding time windows are converted into a set of third feature vectors.

[0200] Each group of third feature vectors is used as input to the window decision model, which outputs a set of prediction time windows corresponding to each set of analysis data. The prediction time windows are the pre-collected time windows corresponding to the analysis data. The training objective is to minimize the sum of prediction errors of all analysis data. The model parameters of the window decision model are optimized through natural heuristic optimization to obtain the model parameters of the window decision model with the minimum sum of prediction errors of all analysis data. The window decision model is a recurrent neural network model.

[0201] The data calibration method is the moving average method, which obtains the smoothed adjustment value by calculating the average of the original adjustment values ​​within a time window; for example, the smoothed adjustment value... In the formula, Indicates the first One original adjustment value, The time window size for the moving average. The number of the original adjustment value. This represents the number of time windows.

[0202] The stability test method includes: starting from the smoothed adjustment value, gradually checking the rate of change between adjacent data points. When the rate of change is continuously lower than the set threshold ϵ, the smoothed adjustment value is considered to have passed the test and entered a stable state. ϵ is a preset value, which is set by those skilled in the art according to the actual situation, such as "if the furnace volume increases to 2m³, ϵ can be adjusted to 0.3℃ / s".

[0203] The method for calculating the adjustment duration is as follows: in a sequence of smoothed adjustment values, record the time point v at which the first adjustment value reaches a stable state; then the adjustment duration is v + ... ,, The unit of measurement is time. This ensures that the data during the adjustment process has stabilized and is suitable for the next experimental step.

[0204] Example 3

[0205] like Figure 3 As shown, the adaptive adjustment control system for the NOx formation mechanism combustion experiment includes:

[0206] Collection module: used to input current experimental data, historical experimental data, preset knowledge base and experimental plan entered by the user;

[0207] Fuzzy New Creation Module: Used to input historical experimental data, experimental plans and preset knowledge base into the pre-built experimental element recognition model, output key fuzzy variables, fuzzy range set, control quantity and fuzzy rules, and establish an initial fuzzy rule base;

[0208] Fuzzy optimization module: Used to input the current experimental data and experimental scheme into the fuzzy optimization model to output the weight set and activation sequence;

[0209] Activation confirmation module: used to determine the activated rules and adjustment variable set based on the activation sequence;

[0210] Model decomposition module: used to perform model decomposition based on the initial fuzzy rule base, weight set and adjustment variable set, and obtain the adjustment value of each adjustment variable;

[0211] Adjustment module: Used to adaptively adjust the furnace using the adjustment values ​​of each adjustment variable;

[0212] Window confirmation module: used to input historical experimental data and experimental plans into the pre-built window decision model, and output the size and number of time windows;

[0213] Data calibration module: used to perform data calibration and delay compensation on the adjustment values ​​according to the size and number of time windows;

[0214] Stability testing module: Used to perform stability tests. If the test passes, the adjustment duration is calculated and obtained.

[0215] Adjustment and optimization module: Receives the adjustment duration. The NOx generation mechanism experimental platform continuously adjusts the adjustment variables according to the adjustment value within the adjustment duration until the adjustment duration ends.

[0216] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0217] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive adjustment control method for NOx formation mechanism combustion experiments, wherein the NOx formation mechanism combustion experiments are conducted on an experimental platform with a furnace, characterized in that the method... include: Input the current experimental data, historical experimental data, preset knowledge base and experimental plan entered by the user; Input historical experimental data, experimental schemes, and preset knowledge bases into the pre-constructed experimental element recognition model, output key fuzzy variables, fuzzy range sets, control quantities, and fuzzy rules, and establish an initial fuzzy rule base; Input the current experimental data and experimental plan into the fuzzy optimization model to output the weight set and activation sequence; The activation rules and the set of adjustment variables are determined based on the activation sequence; The model is solved based on the initial fuzzy rule base, weight set, and adjustment variable set to obtain the adjustment value of each adjustment variable; The furnace is adaptively adjusted using the adjustment values ​​of each adjustment variable.

2. The adaptive adjustment control method for NOx formation mechanism combustion experiments according to claim 1, characterized in that, The method for constructing the experimental element recognition model includes: Each set of historical experimental data, experimental plan, and preset knowledge base is converted into the form of the first feature vector; The set of all first feature vectors is used as the input to the experimental element recognition model. The experimental element recognition model outputs the predicted key fuzzy variables, fuzzy range sets, and fuzzy rules for each set of historical experimental data, experimental schemes, and preset knowledge base. The training objective is to minimize the sum of the first prediction accuracies of all key fuzzy variables, fuzzy range sets, and fuzzy rules. The model parameters of the experimental element recognition model are optimized using the gradient descent algorithm. The experimental element recognition model is constructed by obtaining the model parameters of the experimental element recognition model with the minimum sum of the first prediction accuracies of all key fuzzy variables, fuzzy range sets, and fuzzy rules.

3. The adaptive adjustment control method for the NOx formation mechanism combustion experiment according to claim 1, characterized in that, The method for constructing the fuzzy optimization model includes: Each set of historical experimental data and experimental scheme is converted into the form of a second feature vector; The set of all second feature vectors is used as the input to the fuzzy optimization model. The fuzzy optimization model outputs the weight set and activation sequence predicted for each set of historical experimental data and experimental scheme. The training objective is to minimize the sum of the second prediction accuracies of all predicted weight sets and activation sequences. The model parameters of the fuzzy optimization model are optimized by the gradient descent algorithm. The model parameters of the fuzzy optimization model with the minimum sum of the second prediction accuracies of all predicted weight sets and activation sequences are used to construct the fuzzy optimization model.

4. The adaptive adjustment control method for NOx formation mechanism combustion experiments according to claim 3, characterized in that, The weight set is the set of weights in the weighted average method, i.e., the set of membership degrees; the activation sequence is the sequence composed of the activated rules and the order of the activated rules; the demodulation calculation method adopts the weighted average method, and determines the adjustment value of the adjustment variable to be calculated according to the set of adjustment variables.

5. The adaptive adjustment control method for NOx formation mechanism combustion experiments according to claim 1, characterized in that, The method further includes: Input historical experimental data and experimental design into a pre-built window decision model, and output the size and number of time windows; The adjustment values ​​are calibrated and delayed compensation is performed based on the size and number of time windows. Perform stability testing; if the test passes, calculate the adjustment duration and obtain the adjustment duration. The NOx generation mechanism experimental platform receives the adjustment duration and continuously adjusts the adjustment variable according to the adjustment value within the adjustment duration until the adjustment duration ends.

6. The adaptive adjustment control method for NOx formation mechanism combustion experiments according to claim 5, characterized in that, The method for constructing the window decision model includes: Historical experimental data and experimental schemes are used as analysis data. Time windows corresponding to group c of analysis data are collected in advance, and the analysis data and corresponding time windows are converted into a set of third feature vectors. Each group of third feature vectors is used as input to the window decision model, which outputs a set of prediction time windows corresponding to each set of analysis data, with the goal of minimizing the sum of prediction errors of all analysis data. The model parameters of the window decision model are optimized through natural heuristic optimization, and the model parameters of the window decision model with the minimum sum of prediction errors of all analysis data are obtained to construct the window decision model. The window decision model is a recurrent neural network model.

7. The adaptive adjustment control method for NOx formation mechanism combustion experiments according to claim 6, characterized in that, The data calibration method is the moving average method, which obtains the smoothed adjustment value by calculating the average of the original adjustment values ​​within a time window.

8. The adaptive adjustment control method for NOx formation mechanism combustion experiments according to claim 7, characterized in that, The stability test method includes: starting from the smoothed adjustment value, gradually checking the rate of change between adjacent data points; when the rate of change is consistently lower than a set threshold ϵ, the smoothed adjustment value is considered to have passed the test and entered a stable state, where ϵ is a preset value; the calculation method for the adjustment duration includes: in the sequence of multiple smoothed adjustment values, recording the time point v at which the first adjustment value reaches a stable state, then the adjustment duration is v+ , Used as a unit of time.

9. An adaptive adjustment control system for NOx formation mechanism combustion experiments, characterized in that, The adaptive adjustment control method for the NOx formation mechanism combustion experiment according to any one of claims 1-8, the system comprising: Collection module: used to input current experimental data, historical experimental data, preset knowledge base and experimental plan entered by the user; Fuzzy New Creation Module: Used to input historical experimental data, experimental plans and preset knowledge bases into the pre-built experimental element recognition model, output key fuzzy variables, fuzzy range set, control quantities and fuzzy rules, and establish an initial fuzzy rule base; Fuzzy optimization module: Used to input the current experimental data and experimental scheme into the fuzzy optimization model to output the weight set and activation sequence; Activation confirmation module: used to determine the activated rules and adjustment variable set based on the activation sequence; Model decomposition module: used to perform model decomposition based on the initial fuzzy rule base, weight set and adjustment variable set, and obtain the adjustment value of each adjustment variable; Adjustment module: Used to adaptively adjust the furnace using the adjustment values ​​of each adjustment variable; Window confirmation module: used to input historical experimental data and experimental plans into the pre-built window decision model, and output the size and number of time windows; Data calibration module: used to perform data calibration and delay compensation on the adjustment values ​​according to the size and number of time windows; Stability testing module: Used to perform stability tests. If the test passes, the adjustment duration is calculated and obtained. Adjustment and optimization module: Receives the adjustment duration. The NOx generation mechanism experimental platform continuously adjusts the adjustment variables according to the adjustment value within the adjustment duration until the adjustment duration ends.

Images