Entrained-flow bed gasifier control method and electronic equipment

Abstract

The invention discloses an entrained-flow bed gasifier control method and electronic equipment. The entrained flow gasifier control method comprises the following steps: constructing a burner life prediction model, and training the burner life prediction model; inputting the current data of the entrained-flow gasifier into the burner life prediction model to obtain the predicted residual life of the burner of the entrained-flow gasifier; constructing an optimization objective equation of the entrained-flow bed gasification furnace; and according to the predicted residual life, adjusting parameters of the optimization target equation, and based on the adjusted optimization target equation, controlling the entrained-flow gasifier. Real-time prediction of the service life of the burner is achieved by building a high-precision multi-mode burner service life prediction model, the limitation of linear adjustment of a single parameter is broken through, the parameters of the target equation are adjusted and optimized according to the predicted residual service life, automatic coupling between burner service life prediction and a control strategy is achieved, and the service life of the burner is predicted. And the control strategy accurately adapts to the equipment degradation process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gasification furnace, and particularly relates to a control method of an entrained-flow gasification furnace, an electronic device, a storage medium and a computer program product. BACKGROUND

[0002] As a core process of clean coal utilization, the entrained-flow gasification technology is widely used in coal chemical industry and IGCC power generation. The burner, as a core component of the gasification furnace, directly affects the stability and economy of the system in service life. In current industrial production, the control strategy of the gasification furnace depends on fixed process parameter thresholds (such as oxygen-to-coal ratio, coal slurry concentration, etc.) and manual experience adjustment, and lacks dynamic response capability for the degradation characteristics of the burner life. The traditional control architecture has the following significant drawbacks: first, single parameter threshold control cannot adapt to complex working conditions such as coal switching and load fluctuation, resulting in long-term deviation of the burner from the optimal working interval and accelerating the accumulation of local thermal stress; second, existing life management is mostly based on periodic shutdown detection or fixed period replacement, lacking predictive maintenance support, causing waste of maintenance resources and unplanned downtime losses; third, parameter adjustment and life extension goals are disjointed, and process optimization often focuses on improving gasification efficiency, ignoring the coupling between burner material degradation and process parameters. The current intelligent control level of the gasification furnace cannot meet the long-period operation requirements under severe working conditions such as high-sulfur coal and multi-coal blending.

[0003] In recent years, machine learning technology has made progress in equipment life prediction, but its closed-loop integration with the gasification furnace control system still faces bottlenecks: first, existing prediction models are mostly based on single data types (such as temperature time series data), without deeply integrating multi-modal features such as process operating parameters (such as coal slurry concentration, operating pressure), material performance data (such as coating thickness detection), and environmental parameters (such as cooling water pH value), resulting in limited prediction accuracy; second, there is a lack of real-time feedback channel between the prediction model and the control execution system, and the control strategy adjustment lags behind the burner state changes; third, the traditional control logic cannot achieve dynamic balance between extending the burner life and maintaining the gasification efficiency, and it is urgent to establish a self-consistent control system based on multi-objective optimization.

[0004] For example, Chinese patent "Gasification furnace and oxygen flow control method of one-stage coal burner of gasification furnace" (Application No. 202411577505.4). In this scheme, first, there is a lack of burner life prediction model, mainly focusing on oxygen flow control to improve conversion rate and operation stability, without involving the burner life prediction model, without considering how to optimize the long-term performance of the burner through control logic, and lacking monitoring and prediction of life-related factors such as burner wear and aging. Second, the limitation of control logic, based on the preset load demand and pressure-related limit, lacks adaptive adjustment capability, and may not achieve optimal control when facing complex working conditions or frequent load changes, affecting the service life of the burner. Third, it does not consider the collaborative optimization of multiple burners, focusing on the oxygen flow control of a single one-stage coal burner, without fully considering the collaborative optimization between multiple burners. In actual operation, the mutual influence of multiple burners may lead to poor overall control effect, affecting the long-term operation performance of the gasification furnace. Fourth, it lacks intelligent and predictive maintenance, without introducing intelligent predictive maintenance strategies, and cannot provide early warning of potential burner failures or end-of-life, which is insufficient in equipment management and maintenance, and difficult to achieve fine operation management.

[0005] For example, Chinese patent "A safe logic control method for a pulverized coal gasification furnace" (Application No. 200910018249.4). In this scheme: first, there is a lack of burner life prediction model, mainly focusing on the safe operation control of the gasification furnace, without involving the burner life prediction model, without considering how to optimize the long-term performance of the burner through control logic, and lacking monitoring and prediction of life-related factors such as burner wear and aging; second, the limitation of control logic, based on preset conditions and sequence, lacks adaptive adjustment capability, and may not achieve optimal control when facing complex working conditions or frequent load changes, affecting the service life of the burner; third, the start-up sequence control is relatively fixed and lacks flexibility, and cannot handle special situations during normal operation.

[0006] Therefore, in the prior art, although some gasification furnace control systems have introduced equipment state monitoring or life assessment functions, the control logic still remains in an open-loop or semi-closed-loop mode of "state monitoring → manual intervention → parameter adjustment", without achieving automatic coupling and dynamic response between "life prediction results" and "control strategy generation". Specific problems include:

[0007] 1. Inaccurate life prediction model: existing systems basically rely on the burner life recommendations provided by manufacturers or use simple empirical models to predict remaining life, and cannot obtain accurate burner life prediction results.

[0008] 2. The life prediction result fails to directly participate in the control decision: The existing system mainly uses the life prediction result for alarm or maintenance suggestion, and does not use it as a control variable or an optimization target, so that the control strategy lags behind the equipment degradation process.

[0009] 3. The control strategy lacks an adaptive mechanism: The existing control logic is mainly a fixed threshold or an empirical rule, which cannot dynamically adjust the control parameters according to the burner degradation rate, coal type change and the like, and it is difficult to realize the collaborative optimization of life and efficiency.

[0010] 4. The multi-objective optimization capability is insufficient: The existing system mainly optimizes a single target (such as gasification efficiency), and does not establish a collaborative optimization mechanism for multiple targets such as life, efficiency and safety, so that the control strategy has poor adaptability under complex working conditions. SUMMARY

[0011] Therefore, it is necessary to provide a gas flow bed gasification furnace control method, electronic equipment, storage medium and computer program product in order to solve the technical problem that the existing technology fails to realize the automatic coupling between the burner life prediction and the control strategy of the gas flow bed gasification furnace.

[0012] The present application provides a gas flow bed gasification furnace control method, comprising:

[0013] constructing a burner life prediction model, training the burner life prediction model;

[0014] inputting current data of the gas flow bed gasification furnace into the burner life prediction model to obtain a predicted remaining life of the burner of the gas flow bed gasification furnace;

[0015] constructing an optimization target equation of the gas flow bed gasification furnace;

[0016] adjusting parameters of the optimization target equation according to the predicted remaining life, and controlling the gas flow bed gasification furnace based on the adjusted optimization target equation.

[0017] Further, the construction of the optimization target equation of the gas flow bed gasification furnace comprises:

[0018] constructing an optimization target equation comprising three optimization targets, wherein the first optimization target is to maximize the effective gas yield, the second optimization target is to minimize the burner life decay rate, and the third optimization target is to minimize the specific coal consumption, and the parameters of the optimization target equation include a first weight of the first optimization target, a second weight of the second optimization target and a third weight of the third optimization target.

[0019] Still further, the construction of the optimization target equation comprising three optimization targets comprises:

[0020] The optimization objective equation including three optimization objectives is constructed as: J = ω1 * η + ω2 * v + ω3 * χ + ∑ i λ i penalty term i wherein:

[0021] ω1 is a first weight, ω2 is a second weight, ω3 is a third weight, λ i is a penalty coefficient of the ith constraint term, η is an effective gas yield, v is a burner life attenuation rate, and χ is a specific coal consumption.

[0022] Further, the adjusting the parameters of the optimization objective equation according to the predicted remaining life comprises:

[0023] when the predicted remaining life is greater than or equal to a first percentage of a burner life design value and a confidence of the burner life prediction model is greater than or equal to a first confidence, increasing a first weight and / or a third weight of the optimization objective equation and decreasing a second weight of the optimization objective equation.

[0024] Further, the adjusting the parameters of the optimization objective equation according to the predicted remaining life comprises:

[0025] when the predicted remaining life is less than a second percentage of the burner life design value or the confidence of the burner life prediction model is less than a second confidence, decreasing the first weight and / or the third weight of the optimization objective equation and increasing the second weight of the optimization objective equation.

[0026] Further, the adjusting the parameters of the optimization objective equation according to the predicted remaining life comprises:

[0027] when the predicted remaining life is less than a second percentage of the burner life design value or the confidence of the burner life prediction model is less than a second confidence, decreasing the first weight and / or the third weight of the optimization objective equation and increasing the second weight of the optimization objective equation.

[0028] Further, the method further comprises:

[0029] when the burner life prediction model predicts that the burner life attenuation rate is greater than a rate threshold value, starting a shutdown protection measure of the entrained-flow gasification furnace.

[0030] The present application provides an electronic device, comprising:

[0031] at least one processor; and

[0032] a memory in communication with the at least one processor; wherein

[0033] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the entrained flow gasifier control method as described above.

[0034] The present application provides a storage medium storing computer instructions for performing all steps of the entrained flow gasifier control method as described above when the computer executes the computer instructions.

[0035] The present application provides a computer program product comprising computer programs / instructions which, when executed by a processor, implement the entrained flow gasifier control method as described above.

[0036] The present application predicts the predicted remaining life of the burner of the entrained flow gasifier by constructing a burner life prediction model, and adjusts the parameters of the optimization target equation according to the predicted remaining life, and controls the entrained flow gasifier based on the adjusted optimization target equation. The present application realizes real-time prediction of burner life by constructing a high-precision multi-modal burner life prediction model, breaks through the limitation of single parameter linear adjustment, adjusts the parameters of the optimization target equation according to the predicted remaining life, realizes automatic coupling between burner life prediction and control strategy, and makes the control strategy accurately adapt to the equipment degradation process. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The figure is a work flow chart of an embodiment of the entrained flow gasifier control method of the present application;

[0038] Figure 2 The figure is a work flow chart of another embodiment of the entrained flow gasifier control method of the present application;

[0039] Figure 3 The figure is a schematic diagram of a burner life prediction model of an embodiment of the present application;

[0040] Figure 4 The figure is a work flow chart of an embodiment of the entrained flow gasifier control method of the present application;

[0041] Figure 5 The figure is a schematic diagram of a burner life prediction driving control logic state of the best embodiment of the present application;

[0042] Figure 6 The figure is a schematic diagram of the hardware structure of an electronic device of the present application. DETAILED DESCRIPTION

[0043] The specific embodiments of the present application are further illustrated below with reference to the accompanying drawings. The same reference numbers in different drawings represent the same or similar elements. It should be noted that the words "front", "back", "left", "right", "up", and "down" used in the following description are the directions in the drawings, and the words "inner" and "outer" respectively refer to the directions towards or away from the geometric center of a particular component.

[0044] As shown in Figure 1 The working flow chart of the control method of the gas flow bed gasification furnace according to an embodiment of the present application is shown, which comprises:

[0045] Step S101, constructing a burner life prediction model, training the burner life prediction model;

[0046] Step S102, inputting the current data of the gas flow bed gasification furnace into the burner life prediction model to obtain the predicted remaining life of the burner of the gas flow bed gasification furnace;

[0047] Step S103, constructing an optimization target equation of the gas flow bed gasification furnace;

[0048] Step S104, adjusting the parameters of the optimization target equation according to the predicted remaining life, controlling the gas flow bed gasification furnace based on the adjusted optimization target equation.

[0049] Specifically, the present application can be applied to electronic devices with processing capabilities, such as computers.

[0050] First, step S101 is performed to construct a burner life prediction model and train the burner life prediction model.

[0051] In some embodiments, constructing a burner life prediction model and training the burner life prediction model comprises:

[0052] Obtaining training data of multi-modal characteristic parameters of the burner of the gas flow bed gasification furnace;

[0053] Filtering the multi-modal characteristic parameters to obtain optimal multi-modal characteristic parameters of the burner of the gas flow bed gasification furnace;

[0054] Constructing a burner life prediction model, training the burner life prediction model based on the training data of the optimal multi-modal characteristic parameters, and the burner life prediction model predicting the predicted remaining life of the burner of the gas flow bed gasification furnace according to the data of the optimal multi-modal characteristic parameters.

[0055] Specifically, multi-source data acquisition and characterization construction of the entrained-flow gasifier burner. Collecting data of the entire cycle of the entrained-flow gasifier burner, including medium parameters (composition of carbon-containing raw materials, hardness, etc.), process parameters (oxygen-to-coal ratio, load rate, feed flow rate), burner characteristic parameters (material, structure size, etc.), dynamic sensing data (temperature distribution near the burner, vibration spectrum), environmental parameters (temperature distribution in the gasifier, gasification product indicators, etc.), historical maintenance records (burner replacement cycle, repair measures), etc. as multi-modal feature parameters.

[0056] Then, the degradation sensitive features of the burner are screened and optimized. First, all available data are preprocessed and arranged. Finally, an improved genetic algorithm is applied to screen the optimal multi-modal feature parameters from the multiple multi-modal feature parameters.

[0057] Finally, a machine learning model architecture is designed and trained to predict the burner life.

[0058] Then, step S102 is performed, and the current data of the optimal multi-modal feature parameters are input into the burner life prediction model to obtain the predicted remaining life of the entrained-flow gasifier burner.

[0059] Specifically, the current data of the actual optimal multi-modal feature parameters are obtained and input into the burner life prediction model to obtain the predicted remaining life of the entrained-flow gasifier burner output by the burner life prediction model.

[0060] Then, step S103 is performed to construct an optimization target equation of the entrained-flow gasifier.

[0061] Specifically, an optimization controller is designed, and a target function of a multi-objective optimization strategy is defined. A quantum evolutionary algorithm module is used to quickly search for an optimal solution set in an 80-dimensional parameter space (the calculation time is ≤3 seconds). In addition, a robustness verification unit can also be added to call the HyperStudy software for a million-level random disturbance simulation to ensure that the solution set satisfies the process constraint probability > 99.97%. For the frequent variable load scenario, a "predicted rolling window + feedback correction" dual-mode control strategy is proposed, and the adjustment instruction set is refreshed every 15 seconds to ensure the accuracy of the working condition tracking while avoiding frequent actions of the actuator.

[0062] Finally, step S104 is performed to adjust the parameters of the optimization target equation according to the predicted remaining life, and to control the entrained-flow gasifier based on the adjusted optimization target equation.

[0063] The application predicts the remaining life of the burner of the entrained-flow bed gasifier by constructing a burner life prediction model; meanwhile, according to the predicted remaining life, the parameters of the optimization target equation are adjusted, and the entrained-flow bed gasifier is controlled based on the adjusted optimization target equation. The application realizes real-time prediction of the burner life by constructing a high-precision multi-modal burner life prediction model, breaks through the limitation of single parameter linear adjustment, adjusts the parameters of the optimization target equation according to the predicted remaining life, realizes automatic coupling between the burner life prediction and the control strategy, and makes the control strategy accurately adapt to the equipment degradation process.

[0064] As Figure 2 The application also provides a working flow chart of a control method of an entrained-flow bed gasifier, including the following steps:

[0065] In step S201, a burner life prediction model is constructed, and the burner life prediction model is trained.

[0066] In step S202, current data of the entrained-flow bed gasifier is input into the burner life prediction model, and the predicted remaining life of the burner of the entrained-flow bed gasifier is obtained.

[0067] In step S203, an optimization target equation including three optimization targets is constructed, wherein the first optimization target is to maximize the effective gas yield, the second optimization target is to minimize the burner life decay rate, and the third optimization target is to minimize the specific coal consumption; and the parameters of the optimization target equation include a first weight of the first optimization target, a second weight of the second optimization target, and a third weight of the third optimization target.

[0068] In step S204, when the predicted remaining life is greater than or equal to a first percentage of the burner life design value, and the confidence of the burner life prediction model is greater than or equal to a first confidence, the first weight and / or the third weight of the optimization target equation is increased, and the second weight of the optimization target equation is decreased.

[0069] In step S205, when the predicted remaining life is less than a first percentage of the burner life design value and greater than or equal to a second percentage, and the confidence of the burner life prediction model is greater than or equal to a second confidence, a life benefit function based on the burner life decay rate to calculate the life benefit is constructed, an efficiency benefit function based on the effective gas yield and the specific coal consumption is constructed, the life benefit function and the efficiency benefit function are gambled, the first weight, the second weight and / or the third weight of the optimization target equation is adjusted according to the gambling result, and the second percentage is less than the first percentage.

[0070] Step S206, when the predicted remaining life is less than a second percentage of the burner life design value, or the confidence of the burner life prediction model is less than a second confidence, the first weight and / or the third weight of the optimization objective equation are reduced, and the second weight of the optimization objective equation is increased.

[0071] Step S207, when the burner life prediction model predicts that the burner life decay rate is greater than a rate threshold, a shutdown protection measure of the entrained-flow gasifier is started.

[0072] The present application aims at the main technical defects existing in the existing gasifier control strategy in the management of burner life: ① relying on fixed threshold and manual experience adjustment, unable to dynamically respond to the change of burner degradation rate under complex working conditions such as coal type switching and load fluctuation, lacking accurate burner life prediction ability; ② the life prediction model and the control system are independent of each other, and the parameter adjustment lags behind the material performance degradation process; ③ under the guidance of a single optimization target, process optimization often sacrifices burner life to improve efficiency, and a multi-objective collaborative optimization mechanism has not been established. Therefore, the present application proposes an intelligent control method based on a burner life prediction model, the core innovation of which is to build a full-closed-loop control chain of “dynamic perception→intelligent decision→precise execution”. The present application develops a multi-modal data fusion feature engineering to real-time and synchronously process heterogeneous data streams such as process parameters (oxygen-coal ratio, slag layer thickness), material performance (ultrasonic detection value of residual coating thickness), and environmental factors (chloride ion concentration of cooling water), and establishes a high-fidelity life dynamic perception model; secondly, based on the “RUL prediction driven three-stage control logic” and the life prediction model, an adaptive control strategy generation mechanism is established, taking the output of the life prediction model (such as remaining life percentage RUL%, life decay rate ΔRUL / Δt, prediction confidence Conf, etc.) as the input variable of the control strategy generation, and realizing the deep coupling of the life prediction model and the control system. That is, when the remaining life (RUL) is ≥75% and Conf>0.9, the efficiency priority mode is executed; when RUL is between 50% and 75% and Conf>0.7, the economic balance algorithm is activated; when RUL<50% or Conf<0.7, the life extension protection strategy is started; when ΔRUL / Δt exceeds the set threshold, the emergency load reduction or oxygen-coal ratio adjustment protection measures are triggered; the contradiction of target conflict in the existing PID regulation is solved; finally, a multi-objective collaborative optimization engine based on Pareto frontier is developed, and through the constraint weighting method, the indicators such as syngas effective gas content (η), specific coal consumption (χ), and burner life decay rate (v) are integrated into the same decision space, realizing self-consistent control under complex working conditions.

[0073] Specifically, first, step S201 is executed to build a burner life prediction model, and the burner life prediction model is trained.

[0074] In some embodiments, the construction of the burner life prediction model, the training of the burner life prediction model comprises:

[0075] Obtaining training data of multi-modal characteristic parameters of an entrained flow gasifier burner;

[0076] Screening the multi-modal characteristic parameters to obtain optimal multi-modal characteristic parameters of the entrained flow gasifier burner;

[0077] Constructing a burner life prediction model, training the burner life prediction model based on the training data of the optimal multi-modal characteristic parameters, the burner life prediction model predicting the predicted remaining life of the entrained flow gasifier burner according to the data of the optimal multi-modal characteristic parameters, the burner life prediction model comprising:

[0078] A neural network branch, a time series sensor data processing branch, a semantic feature branch, a cross attention mechanism module, a weight module, and a burner remaining life prediction module, wherein:

[0079] The neural network branch inputs static or quasi-static characteristic parameters in the optimal multi-modal characteristic parameters and outputs a static feature vector;

[0080] The time series sensor data processing branch inputs time series sensor data and outputs a time series feature vector;

[0081] The semantic feature branch inputs maintenance record text and outputs a semantic feature vector;

[0082] The output of the time series sensor data processing branch is the query value input to the cross attention mechanism module, the output of the semantic feature branch is the keyword and key value input to the cross attention mechanism module, and the cross attention mechanism module outputs a context feature vector;

[0083] The weight module inputs operating condition context and outputs weights of the neural network branch, the time series sensor data processing branch, the semantic feature branch, and the cross attention mechanism module;

[0084] The outputs of the neural network branch, the time series sensor data processing branch, the semantic feature branch, and the cross attention mechanism module are multiplied by the corresponding weights to form a fusion feature vector, which is input to the burner remaining life prediction module, and the burner remaining life prediction module outputs the predicted remaining life of the entrained flow gasifier burner.

[0085] Specifically, first, multi-source data acquisition and characterization construction of the gas flow bed gasification furnace burner are performed. The multi-modal characteristic parameters of the gas flow bed gasification furnace burner are collected, including medium parameters (composition, hardness, etc. of carbon-containing raw materials), process parameters (oxygen-coal ratio, load rate, feed flow rate), burner characteristic parameters (material, structure size, etc.), dynamic sensing data (temperature distribution near the burner, vibration spectrum), environmental parameters (temperature distribution in the gasification furnace, gasification product indicators, etc.), historical maintenance records (burner replacement cycle, repair measures), etc.

[0086] Then, the multi-modal characteristic parameters are screened to obtain the optimal multi-modal characteristic parameters of the gas flow bed gasification furnace burner.

[0087] In some embodiments, the screening of the multi-modal characteristic parameters to obtain the optimal multi-modal characteristic parameters of the gas flow bed gasification furnace burner comprises:

[0088] The multi-modal characteristics are encoded into chromosomes, and each gene represents a characteristic variable and its weight;

[0089] The initial population is initialized, and the population includes multiple chromosomes;

[0090] The following operations are iteratively performed:

[0091] The prediction error and the simplicity of the feature set are taken as double objective functions, and the fitness value of each chromosome is calculated by weighted summation;

[0092] If the fitness value converges or reaches the maximum number of iterations, the optimal feature subset is output as the optimal multi-modal characteristic parameter of the gas flow bed gasification furnace burner, and the iteration is ended, otherwise, a chromosome is selected from the population as a to-be-operated chromosome based on the fitness;

[0093] The to-be-operated chromosome is operated to generate a child population through single-point crossover operation;

[0094] The next iteration is performed on the child population.

[0095] Specifically, the burner degradation sensitive features are screened and optimized. First, all available data are preprocessed and arranged. For example, for discrete data such as medium parameters and burner feature parameters, the laboratory test results are introduced into the burner life prediction model database through the data interface. The data to be saved include, but are not limited to, the industrial analysis of carbon-containing materials, the particle size distribution of carbon-containing materials, the hardness, the sulfur content, the material of the burner, the diameter of the burner channel, the angle of the burner nozzle, etc. For process parameters, dynamic sensing data, and environmental parameters, the real-time data collected by sensors and DCS systems are transmitted to the burner life prediction model database. The data to be saved include, but are not limited to, the oxygen-coal ratio, the load rate, the feed flow rate (solid flow rate and conveying gas flow rate for dry pulverized coal gasification technology, and coal slurry flow rate and coal slurry concentration for coal slurry gasification technology), oxygen flow rate, gasifier temperature, syngas composition, burner cooling water flow rate, burner temperature distribution, and vibration frequency. For historical maintenance records and other multi-modal data, NLP techniques (such as BERT fine-tuning) can be used to extract key events (such as “2023-05-12 burner A bad spot repair welding, duration 120 min”) from work order logs and encode them into semantic vectors. Alternatively, the quantification weights of maintenance types (cleaning, coating repair) on life repair effects (such as 10% life recovery for spot welding and 30% life recovery for coating repair) can be defined to construct a maintenance effect matrix based on time decay (such as: maintenance effect exponentially decreases with time). After collecting and arranging the relevant databases, redundant parameters (such as weak influence of environmental temperature and humidity on high-temperature burners) are removed using Spearman rank correlation analysis and expert experience library (such as corrosion mechanism prior knowledge), and the prediction index range is narrowed.

[0096] Finally, the improved genetic algorithm (with constraint condition population initialization) is applied to encode the multi-modal features (the above-mentioned parameters) into chromosomes, with each gene representing a feature variable and its weight. A chromosome is a combination of a “feature subset + importance of each feature”. The “feature variable” indicates which features are selected in this chromosome, and the “weight” indicates the relative importance of these features in this chromosome. By encoding the weight into the gene, the genetic algorithm can optimize both the “feature selection” and the “weight adjustment” dimensions, improving the prediction accuracy while considering the feature simplicity.

[0097] Then, the prediction error and the feature set simplicity (number of features) are used as dual objective functions to calculate the fitness value by weighted summation. The high fitness individuals are retained using the tournament selection strategy. The single-point crossover operation is used to generate offspring, enhancing the diversity of feature combinations. The iteration is stopped when the fitness value converges or the maximum number of iterations is reached, and the optimal feature subset is output, completing the screening of high-contribution feature combinations (such as hot spot area change rate, load rate, and feed flow rate).

[0098] wherein, for the i-th chromosome, its fitness F i The following weighted sum formula is used for calculation:

[0099]

[0100] wherein, E i : the normalized prediction error (mean squared error or MAE, normalized to [0, 1]) of the i-th chromosome on the validation set for the corresponding feature subset; S i : the normalized value of the number of selected features of the chromosome, S i =(number of selected features) / (total number of features), also in [0, 1]; w1, w2: user preset weight coefficients, satisfying w1+w2=1 (for example, w1=0.7, w2=0.3).

[0101] Finally, a burner life prediction model is constructed, and the training of the burner life prediction model is performed based on the training data of the optimal multi-modal feature parameters, the burner life prediction model predicts the predicted remaining life of the gas flow bed gasifier burner according to the data of the optimal multi-modal feature parameters, and the burner life prediction model comprises:

[0102] a neural network branch, a time series sensor data processing branch, a semantic feature branch, a cross attention mechanism module, a weight module, and a burner remaining life prediction module, wherein:

[0103] The neural network branch inputs the static or quasi-static feature parameters in the optimal multi-modal feature parameters, and outputs a static feature vector;

[0104] The time series sensor data processing branch inputs the time series sensor data, and outputs a time series feature vector;

[0105] The semantic feature branch inputs the maintenance record text, and outputs a semantic feature vector;

[0106] The output of the time series sensor data processing branch is used as the query value input into the cross attention mechanism module, the output of the semantic feature branch is used as the keyword and key value input into the cross attention mechanism module, and the cross attention mechanism module outputs a context feature vector;

[0107] The weight module inputs the working condition context, and outputs the weights of the neural network branch, the time series sensor data processing branch, the semantic feature branch, and the cross attention mechanism module;

[0108] The outputs of the neural network branch, the time series sensor data processing branch, the semantic feature branch, and the cross attention mechanism module are multiplied by the corresponding weights, and then spliced to form a fusion feature vector, which is input into the burner remaining life prediction module, and the burner remaining life prediction module outputs the predicted remaining life of the gas flow bed gasifier burner.

[0109] Specifically, the machine learning model architecture design and training are performed. To achieve accurate prediction of the service life of the entrained-flow gasifier burner, the embodiment constructs a multi-modal fusion deep learning architecture composed of multiple sub-models that process different types of input data and are jointly trained and optimized through shared bottom layer feature expression and multi-task learning mechanism. Specifically as follows:

[0110] 1. Model structure and function division

[0111] The architecture includes the following core sub-models, each of which processes data of a specific modality and outputs an intermediate feature vector for subsequent fusion and prediction:

[0112] 1) Neural network branch, preferably DNN (Deep Neural Network) branch:

[0113] Input: static or quasi-static features such as process parameters (e.g. oxygen-to-coal ratio, load rate), coal properties (e.g. sulfur content, ash content), burner structure parameters (e.g. nozzle diameter), etc.

[0114] Output: fixed-dimension static feature vector (e.g. 128 dimensions) for representing the influence of process and material properties on burner degradation. For example, including gasifier oxygen-to-coal ratio, load rate, coal ash content, coal sulfur content, burner nozzle size, etc. to form a feature vector, which after normalization becomes a static feature vector [0.67, 0.12, -0.05, -0.12, 0.03…].

[0115] 2) Time series sensor data processing branch, preferably LSTM (Long Short-Term Memory Network) or Transformer branch:

[0116] Input: time series sensor data (e.g. temperature, vibration, cooling water flow, etc.)

[0117] Output: time series feature vector (e.g. 128 dimensions) for capturing the dynamic evolution trend during burner degradation. For example, time series data measured by process instruments, including gasifier temperature, burner vibration frequency, burner cooling water flow, gasification product flow, etc. to form a feature vector, which after normalization becomes a time series feature vector [0.82, -0.24, 0.17, 0.21, …].

[0118] 3) Semantic feature branch, preferably BERT fine-tuning model (text encoder):

[0119] Input: maintenance record text (e.g. "2023-05-12 Burner A repair welding repair")

[0120] Output: Semantic feature vector (e.g. 128 dimensions) to encode maintenance event type, frequency and its impact on the remaining life. For example, maintenance record text, parking record text, etc. to extract maintenance method, repair duration, post-repair status, etc. and translate into vector encoding using BERT, normalized as semantic feature vector [0.45, 0.03, 0.60,...].

[0121] 4) Cross-Attention module:

[0122] Input: Time series features from LSTM / Transformer as Query, maintenance semantic features from BERT output as Key / Value;

[0123] Output: Fused context feature vector to capture the relevance of "performing a certain type of maintenance operation at a certain degradation stage". For example, under the trend of "gasifier time series temperature rising", the comprehensive impact of "coating repair" on the remaining life, get the attention weight, weighted sum of each semantic with attention weight, form a new multi-dimensional vector, normalized as context feature vector [0.71, 0.09, -0.18,...].

[0124] 5) Weight module, preferably using a learnable gating network (Gating Network):

[0125] Input: Current operating context (e.g. coal type, load level);

[0126] Output: Weight coefficients of each modal branch (e.g. DNN weight 0.3, LSTM weight 0.5, BERT weight 0.2) to dynamically adjust the contribution of each modal in the final prediction.

[0127] 6) Burner remaining life prediction module (RUL), which is a regression task.

[0128] Input of the burner remaining life prediction module: fused multi-modal feature vector (concatenated after weighted by the branch output weight module), specifically the static feature vector of the neural network branch, the time series feature vector output by the time series sensor data processing branch, the semantic feature vector output by the semantic feature branch, and the context feature vector output by the cross-attention mechanism module are multiplied by the corresponding weights output by the weight module to form a large feature vector as the fused feature vector.

[0129] Output of the burner remaining life prediction module: Burner remaining life (RUL) in hours.

[0130] In one embodiment, the burner life prediction model further comprises a maintenance type classification module and a degradation stage classification module, wherein:

[0131] The maintenance type classification module inputs the fusion feature vector and outputs a maintenance type label;

[0132] The degradation stage classification module inputs the fusion feature vector and outputs a degradation stage label.

[0133] Specifically, a multi-task learning framework and a sharing mechanism are adopted.

[0134] The embodiment adopts a multi-task learning (Multi-Task Learning) framework, and the main task and the auxiliary task share the bottom layer feature expression, which is specifically as follows:

[0135] The main task is a burner remaining useful life prediction module, which is used for burner remaining useful life prediction (RUL) and is a regression task.

[0136] The first auxiliary task is a maintenance type classification module, which is used for maintenance type classification (such as cleaning, repair welding, coating repair, etc.) and is a multi-classification task.

[0137] The second auxiliary task is a degradation stage classification module, which is used for degradation stage classification (such as normal period, early warning period, critical period) and is a multi-classification task.

[0138] The meaning of sharing the bottom layer feature expression is that the feature vectors output by branches such as neural network branch (DNN), time sequence sensor data processing branch (LSTM) and semantic feature branch (BERT) are fused and used as the common input of the life prediction module, the maintenance type classification module and the degradation stage classification module, so as to avoid repeated modeling, improve the training efficiency and the generalization ability.

[0139] Input-output relationship and life prediction mechanism:

[0140] Input of the life prediction model: the fused multi-modal feature vector (which is spliced after being weighted by the Gating Network and output by the branch);

[0141] Output of the life prediction model: burner remaining useful life (RUL), unit: hour;

[0142] Input of the auxiliary task: the same fused feature vector;

[0143] Output of the auxiliary task: maintenance type label or degradation stage label.

[0144] The classification result of the auxiliary task is not used as the input of the life prediction, but is shared with the main task for feature expression, so as to improve the understanding ability of the model to the degradation mode through joint training, thereby improving the prediction accuracy of the main task.

[0145] For example, Figure 3The model architecture of the best embodiment of the present application is shown, including: a neural network branch 31 (preferably a DNN branch), a time series sensor data processing branch 32 (preferably an LSTM / Transformer branch), a semantic feature branch 33 (preferably a BERT fine-tuning model), a cross-attention mechanism module 34, a weight module 35 (preferably a learnable gating network model), a life prediction model 36, a maintenance type classification model 37, and a degradation stage classification model 38.

[0146] Loss function design and model updating mechanism:

[0147] To coordinate multi-task training and prevent a certain modality or task from dominating model updating, the embodiment designs the following loss function:

[0148] Main task loss: L MSE , specifically mean square error loss (MSE), used for regression life prediction;

[0149] First auxiliary task loss: L CE1 , specifically cross-entropy loss (Cross-Entropy), used for maintenance type classification task;

[0150] Second auxiliary task loss: L CE2 , specifically cross-entropy loss (Cross-Entropy), used for degradation stage classification task;

[0151] Modality balance loss (Modality Balance Loss): L balance , specifically L2 norm constraint on the feature vectors output by each modality branch, to prevent a certain modality feature from dominating the fusion process;

[0152] Total loss function:

[0153] L total = λ1L MSE + λ2L CE1 + λ3L CE2 + λ4L balance

[0154] Where λ1, λ2, λ3 and λ4 are adjustable weight coefficients for balancing the contributions of each task and the regularization term.

[0155] Model training and updating method:

[0156] Training mode: end-to-end training, all sub-modules / branches, including: neural network branch (DNN), time series sensor data processing branch (LSTM), semantic feature branch (BERT), cross-attention mechanism branch / module (Cross-Attention) and weight module (Gating Network) participate in back propagation together, and the total loss function is optimized uniformly;

[0157] Prevent single modality from dominating updates: through modal balance loss and dynamic adjustment of modality weights by gating network, ensure stable learning of the model under multi-modal input;

[0158] Model update mechanism: in the online deployment stage, use incremental learning method (such as updating only the fully connected layer) to fine-tune the model locally, and adapt to new working conditions.

[0159] This embodiment quantifies the coupling effect of coal type, process, and maintenance event on burner life through cross-modal feature engineering and cross-attention mechanism, breaks through the prediction accuracy, and at the same time, classifies the maintenance type and degradation stage while predicting the burner life.

[0160] Then step S202 is performed, and the current data of the entrained-flow bed gasifier is input into the burner life prediction model to obtain the predicted remaining life of the burner of the entrained-flow bed gasifier.

[0161] In some embodiments, further comprising: when the update condition is met, retraining the burner remaining life prediction module, the maintenance type classification module, and the degradation stage classification module.

[0162] Specifically, dynamic working condition adaptation and online update of the burner life prediction model are performed. Unsupervised clustering (such as t-SNE) is used to identify coal switching or load fluctuation scenarios, and the model threshold is adjusted in real time.

[0163] In some embodiments, based on the predicted remaining life of the burner of the entrained-flow bed gasifier, the burner health index of the burner of the entrained-flow bed gasifier is calculated, and whether to trigger an early warning is determined based on the burner health index and a burner health index threshold (HI threshold).

[0164] Specifically, the burner health index is calculated as: wherein HI is the burner health index, RUL pred is the predicted remaining life, RUL design is the designed life of the burner.

[0165] Then, whether to trigger an early warning is determined based on the burner health index and a burner health index threshold (HI threshold).

[0166] The burner health index (HI) is used to divide the degradation stages (such as normal period, early warning period, critical period) of the burner.

[0167] Among them, the real-time adjustment model threshold value includes adjusting: the HI threshold value for triggering different levels of early warning (such as HI<0.7 triggering early warning, HI<0.4 triggering critical alarm); the confidence interval threshold value for judging whether the current prediction result is reliable (such as triggering model updating when the confidence interval width exceeds the set threshold value).

[0168] The specific adjustment method is: when the system detects the working condition change (such as coal switching, load fluctuation) through unsupervised clustering (such as t-SNE), the HI threshold value under the current working condition is dynamically adjusted according to the HI distribution under the same working condition in the historical data.

[0169] Specifically, when t-SNE detects a new working condition (such as high-sulfur coal), the system does not immediately retrain the model, but judges (1) finds the historical HI distribution of the same working condition from the historical database; (2) adjusts the "early warning / alarm threshold value" with the statistical value (such as the 5% quantile) of these historical HI (for example, adjusts the HI alarm line from 0.7 to 0.6). For example: the burner degradation is faster under the high-sulfur coal working condition, and the HI threshold value will be adjusted accordingly (such as from 0.7 to 0.6) to adapt to the new degradation rate.

[0170] In one of the embodiments, the satisfaction of the updating condition includes:

[0171] If the prediction error of the burner life prediction model exceeds the set threshold value, it is judged that the updating condition is satisfied; or

[0172] If the working condition clustering result of the entrained-flow gasifier changes, it is judged that the updating condition is satisfied; or

[0173] If the difference between the current data distribution input into the burner life prediction model and the training data distribution is greater than the difference threshold value, it is judged that the updating condition is satisfied.

[0174] Specifically, based on the Flink stream processing engine, the real-time data is divided according to the time window, and the following indicators are continuously monitored:

[0175] 1) The prediction error exceeds the set threshold value (such as MAE>10%);

[0176] 2) The working condition clustering result changes (such as coal switching);

[0177] 3) The new data distribution and the training data distribution have a large difference (judged by KL divergence or PSI index).

[0178] When any of the above conditions is satisfied, local retraining is triggered.

[0179] The local retraining is limited to the parameters of the fully connected layer after the fusion layer, i.e., the remaining life prediction module of the burner, the maintenance type classification module, and the degradation stage classification module; the backbone feature extraction network such as the neural network branch (DNN), the time series sensor data processing branch (LSTM), the semantic feature branch (BERT), the cross-attention mechanism branch / module (Cross-Attention), and the weight module (Gating Network) is not updated, so as to save the computing resources.

[0180] In one of the embodiments, the satisfaction of the updating condition comprises:

[0181] A confidence level is set, and the upper limit and the lower limit of the confidence interval corresponding to the running time of the current burner are calculated;

[0182] According to the predicted remaining life output by the burner life prediction model, the burner health index is calculated as: wherein HI is the burner health index, RUL pred is the predicted remaining life, RUL design is the designed life of the burner;

[0183] If the burner health index falls outside the confidence interval, it is judged that the updating condition is satisfied.

[0184] Specifically, the burner health index (HI) is defined as:

[0185]

[0186] wherein RUL pred is the remaining life (hours) predicted by the model; RUL design is the designed life of the burner (such as 2000 hours). When HI is 1, the burner is in a brand-new state; when HI is 0, the burner life is exhausted and needs to be replaced immediately; when HI is in the interval of 0 to 1, it represents the degradation degree of the burner, and the smaller the value, the more serious the degradation.

[0187] Based on the RUL output by the model, the remaining life stage is divided (such as normal period→warning period→critical period), and the confidence interval can be calibrated in combination with the Weibull distribution model. The Weibull distribution is a statistical model commonly used for life analysis, which is suitable for describing the failure time distribution of the burner. Its probability density function is:

[0188]

[0189] wherein k is the shape parameter; λ is the size parameter, and t is the running time of the burner. The confidence interval calibration process is:

[0190] Step 1: Based on historical burner failure data, fit the Weibull distribution, estimate parameters k and λ;

[0191] Step 2: According to the running time t of the current burner, calculate its survival probability P(T>t);

[0192] Step 3: Set the confidence level (such as 95%), use the same historical failure sample, estimate the parameter uncertainty of Weibull parameters (k, λ) by maximum likelihood or Bayesian method, and then calculate the confidence interval of P(T>t), get the confidence interval corresponding to the confidence level of the running time of the current burner The upper and lower limits of the confidence interval;

[0193] Step 4: Compare the RUL predicted by the model with the confidence interval of the Weibull distribution. If the predicted value is outside the confidence interval, trigger model retraining or alarm.

[0194] Then step S203 is performed, and an optimization objective equation including three optimization objectives is constructed, wherein a first optimization objective is to maximize effective gas yield, a second optimization objective is to minimize burner life decay rate, and a third optimization objective is to minimize specific coal consumption. The parameters of the optimization objective equation include: a first weight of the first optimization objective, a second weight of the second optimization objective, and a third weight of the third optimization objective.

[0195] Multi-objective control strategy dynamic generation. Design an optimization controller to convert the multi-objective optimization problem into a solution system composed of three interrelated modules:

[0196] Adaptive weight allocation network, define the objective function of the multi-objective optimization strategy, including maximizing effective gas yield η, minimizing burner life decay rate v and minimizing specific coal consumption χ (mass of coal consumed per 1000 Nm3 of (CO+H2) produced), specifically:

[0197]

[0198] The "constraint weighting method" is used to convert multiple optimization objectives into a single objective function J through weighted summation, and process constraints (such as temperature, pressure, oxygen-to-coal ratio limits) are considered during optimization. Specifically:

[0199] According to the real-time economic demand, automatically adjust the weight ratio between life and efficiency targets, for example, when the burner life is high, increase ω1, when the burner life is low, increase ω2;

[0200] Quantum evolutionary algorithm module, quickly search for the optimal solution set in an 80-dimensional parameter space (calculation time ≤ 3 seconds);

[0201] Robustness verification unit, call HyperStudy software to carry out million level random disturbance simulation, ensure that the solution set meets the process constraint probability > 99.97%. For the frequent variable load scene, the "prediction rolling window + feedback correction" dual-mode control strategy is proposed, and the adjustment instruction set is refreshed every 15 seconds, which can ensure the tracking accuracy of the working condition while avoiding frequent action of the actuator.

[0202] Then, the strategy decision scheme is generated and the parameters are optimized. The "life-control mapping function" and the "RUL prediction driven three-stage control logic" are constructed, that is, according to the predicted remaining life and the model confidence, steps S204-S207 are executed respectively. Specifically:

[0203] When the predicted remaining life is greater than or equal to the first percentage of the design value of the burner life, and the confidence of the burner life prediction model is greater than or equal to the first confidence, step S204 is executed, the first weight and / or the third weight of the optimization objective equation are increased, and the second weight of the optimization objective equation is reduced.

[0204] Specifically, when the remaining life of the burner is greater than the first percentage of the design value, and the confidence of the model is greater than the first confidence, the burner is in good condition, and the control target is to maximize the effective gas yield (maximize the effective gas (CO+H2) component). Through the fuzzy control algorithm, a fuzzy PID controller is used, and the target weight of the multi-objective optimization function J is set, such as "efficiency priority" weight factor ω1=0.8 and "life protection" weight factor ω2=0.2. The variables such as oxygen-coal ratio, coal slurry concentration, and gasifier load are controlled, and under the premise of ensuring that the burner temperature gradient ΔT / Δt is less than a set threshold (25℃ / min), the oxygen-coal ratio is gradually increased to the upper limit of the process, while ensuring that the raw material quantity and the oxygen-coal ratio fluctuate small, and the system is maintained stable.

[0205] The first percentage is preferably 75%, and the first confidence is 0.9.

[0206] When the predicted remaining life is less than the first percentage of the design value of the burner life and greater than or equal to the second percentage, and the confidence of the burner life prediction model is greater than or equal to the second confidence, step S205 is executed, a life benefit function based on the calculation of the life benefit of the burner life decay rate is constructed, an efficiency benefit function based on the calculation of the effective gas yield and the specific coal consumption is constructed, and the life benefit function and the efficiency benefit function are gambled. According to the game result, the first weight and the second weight and / or the third weight of the optimization objective equation are adjusted, and the second percentage is less than the first percentage.

[0207] When the remaining life of the burner is less than 75% of the design value but greater than or equal to 50% of the design value, and the confidence of the model is greater than 0.7, the control target is changed, and the burner life (i.e., the safety and stable operation target of the gasifier) and the gasification efficiency (the economic index) need to be considered, and the life benefit and efficiency benefit functions are constructed.

[0208] The second percentage is preferably 50%, and the second confidence is preferably 0.7.

[0209] In some embodiments, the life benefit function takes the burner life decay rate as the core variable, which is quantified as the amount of remaining life of the burner per unit time. Its specific form is as follows:

[0210]

[0211] where v is the burner life decay rate (unit: % / hour), which is output in real time by the life prediction model; β1 is the first life benefit function weight coefficient, and β2 is the second life benefit function weight coefficient, which are calibrated through historical data to determine the first life benefit function weight coefficient and the second life benefit function weight coefficient (for example, β1 = 100, β2 = 0.8), for balancing the linear inverse proportion of the decay rate and the quadratic penalty term.

[0212] In the life benefit formula, the first term represents the direct benefit of delaying life decay (the smaller v is, the higher the benefit is); the second term is a quadratic penalty for high decay rate, to prevent rapid depletion of life in extreme working conditions.

[0213] In some embodiments, the efficiency benefit function integrates the effective gas yield and specific coal consumption two economic indicators, and the specific form is as follows:

[0214] E benefit = α1η - α2χ

[0215] where η is the effective gas yield (unit: %), that is, the volume fraction of (CO+H) in the syngas; χ is the specific coal consumption (unit: kg / (1000Nm 3 (CO+H2)), that is, the mass of coal consumed per unit of effective gas produced. α1 is the first efficiency benefit function weight coefficient, and α2 is the second efficiency benefit function weight coefficient (for example, α1 = 0.6, α2 = 0.4), reflecting the relative importance of efficiency and energy consumption.

[0216] In the formula, the effective gas yield needs to be maximized, and the specific coal consumption needs to be minimized, and a weight adjustment mechanism based on game is adopted.

[0217] Specifically, when the remaining life is in the middle interval (e.g., 50%~75% of the design life) and the model confidence is high, the weights (ω1, ω2, ω3) of the optimization objective equation need to be dynamically adjusted through Nash equilibrium game. The specific steps are as follows:

[0218] 1) Construct a game model

[0219] The two parties of the game model are the life benefit L benefit and the efficiency benefit E benefit , wherein the goal of the life benefit is to maximize the life benefit, and the goal of the efficiency benefit is to maximize the efficiency benefit.

[0220] The two parties conduct a game by adjusting the weight distribution strategy (ω2 and ω1+ω3). For example:

[0221] Life side strategy: Choose the value range of ω2 (life weight) [0.3, 0.8].

[0222] Efficiency side strategy: Choose the value range of ω1+ω3 (efficiency+energy consumption weight) [0.7, 0.2] (total is 1).

[0223] 2) Construct a benefit matrix

[0224] Generate a benefit matrix according to historical data, for example:

[0225] Lifetime benefit weight Efficiency benefit weight [[ L benefit ]]> E benefit ]]> 0.3 0.7 60 85 0.5 0.5 75 75 … … … … 0.8 0.2 90 40

[0226] 3) Nash equilibrium solution

[0227] Use the Lemke-Howson algorithm to calculate the Nash equilibrium point, that is, the optimal weight combination that both parties cannot obtain higher benefits by changing their strategies unilaterally. For example:

[0228] Calculate ω2=0.6 (life weight) and ω1+ω3=0.4 (efficiency+energy consumption weight); substitute this weight combination into the optimization objective equation J to guide real-time control.

[0229] 4) Dynamic update

[0230] Recalculate the benefit matrix at fixed intervals (e.g., 15~30 minutes / time) to adapt to coal switching or load fluctuations.

[0231] When the predicted remaining life is less than the second percentage of the design value of the burner life, or the confidence of the burner life prediction model is less than the second confidence, perform step S206, reduce the first weight and / or the third weight of the optimization objective equation, and increase the second weight of the optimization objective equation.

[0232] Specifically, when the remaining life of the burner is less than 50% of the design value, or the confidence of the model is less than 0.7, it indicates that the burner life is low and the prediction model cannot accurately predict, and the control target is to minimize the life decay rate, reduce the first weight and / or the third weight of the optimization objective equation, and increase the second weight of the optimization objective equation.

[0233] In some embodiments, when the predicted remaining life is less than a second percentage of the design value of the burner life, or the confidence of the burner life prediction model is less than a second confidence, a reinforcement learning compensation strategy (such as DQN or PPO) is triggered to match the planned shutdown cycle, an agent is trained according to historical operation data to learn an optimal life extension control strategy, and rewards or penalties are obtained according to the life decay.

[0234] When the burner life prediction model predicts that the burner life decay rate is greater than the rate threshold, step S207 is performed to start the shutdown protection measures of the entrained flow gasifier.

[0235] Specifically, when the burner life prediction driving model predicts that the burner life decay rate exceeds the threshold (greater than 0.1% / hour), it indicates that the burner may have suffered irreversible damage and is rapidly degrading, and emergency shutdown protection measures need to be started to ensure system safety.

[0236] The rate threshold is preferably 0.1% / hour.

[0237] wherein the burner life decay rate v refers to the percentage of decay of the remaining life (RUL) of the burner per unit time, and the calculation formula is:

[0238]

[0239] wherein RUL t-1 is the remaining life (hours) output by the burner life prediction model at time t-1, and RUL t is the remaining life output by the burner life prediction model at time t; RUL design is the design life of the burner (a fixed value, for example, 2000 hours); and Δt is the time interval between time t-1 and time t.

[0240] The degradation rate is indirectly calculated from the RUL sequence output by the life prediction model in real time. The life prediction module outputs the current RUL value every Δt, and the system caches the last two RUL values, which are substituted into the above formula to obtain ΔRUL / Δt, i.e., the burner life decay rate v.

[0241] The embodiment solves three technical bottlenecks of the prior art: ① The problem that the traditional gas flow bed gasifier burner control relies on fixed threshold and manual experience adjustment is broken through, the real-time prediction of the burner life is realized by constructing a high-precision multi-modal burner life prediction model (referring to the content of the 051th patent proposed by the right holder); ② The limitation of single parameter linear regulation is broken through, the life prediction result is deeply embedded into the control strategy by constructing a “life-control mapping function” and a “RUL prediction driven three-stage control logic”, and the dynamic balance of life and efficiency is realized; ③ The limitation of single optimization target orientation of the traditional system control is expanded, a multi-objective collaborative optimization engine based on the Pareto frontier is developed, indexes such as effective gas content (η), specific coal consumption (χ) and burner life decay rate (v) are integrated into the same decision space, and self-consistent control under complex working conditions is realized. The above technical innovations make the method especially suitable for industrialized scenes with high-sulfur coal and large fluctuation of inferior coal, and provide a new generation of technical platform for intelligent upgrading of the gas flow bed gasifier.

[0242] As Figure 4 The working flow chart of the gas flow bed gasifier control method of the best embodiment of the application comprises:

[0243] Step S401, setting a sensor acquisition layer. In addition to the flow, temperature, pressure and other measuring points regularly arranged in the gasifier, a group of heat flux density sensors near the burner is arranged.

[0244] Step S402, setting an edge computing layer. All sensor data related to the burner life of the gasifier are uploaded to the edge computing server, and the sensor data are pretreated and feature extracted in turn

[0245] Step S403, setting a burner life prediction layer. The multi-modal data are jointly driven for feature engineering, the cross-modal key variables are screened by constructing a heat flow coupling parameter combined with an improved genetic algorithm, the full-dimensional data fusion of process-material-environment is realized, the burner life prediction model based on machine learning and deep learning algorithm is constructed, and the current burner state and future life are predicted according to the real-time sensor operation data.

[0246] Step S404, strategy decision and parameter optimization.

[0247] Specifically, a “life-control mapping function” is constructed, the output of the life prediction model (such as the remaining life percentage RUL%, the life decay rate ΔRUL / Δt, the prediction confidence Conf, etc.) is taken as the input variable of the control strategy generation, and the following mapping relationship is constructed:

[0248] Control strategy = f(RUL%, ΔRUL / Δt, Conf, current working condition parameter)

[0249] Wherein, the function f is trained by a fuzzy neural network (FNN) or a deep reinforcement learning (DRL) model, which can automatically adjust the weight distribution of the control target according to the real-time life state. For example:

[0250] When RUL% ≥ 75% and Conf > 0.9, the system prioritizes optimizing gasification efficiency;

[0251] When RUL% < 50% or Conf < 0.7, the system prioritizes executing life extension protection strategies;

[0252] When RUL is between 50% and 75% and Conf > 0.7, activate the economic balance algorithm;

[0253] When ΔRUL / Δt exceeds the set threshold (> 0.1% / hour), trigger emergency load reduction or adjust oxygen-coal ratio protection measures.

[0254] That is, on the basis of the "RUL (remaining life) prediction-driven three-stage control logic" of the burner life, the real-time influence of the burner life prediction model on decision-making is introduced, the game-type real-time optimization of the burner life (i.e. safety, stability) and the gasification efficiency (economy) is carried out, the optimal dynamic optimization strategy is automatically selected, and the control scheme and parameters are optimized according to the multi-objective prediction model

[0255] Step S405, execute the control layer. According to the control scheme and the regulation method, the dynamic adjustment of the actuator is carried out through the DCS system, the superposition suppression type regulation algorithm is developed, the composition fluctuation problem of the synthesis gas in the parameter adjustment process is solved through the feedforward compensation mechanism (such as oxygen flow regulation, synchronous correction of lock hopper deslagging period, and ensuring that the gasification chamber pressure fluctuation is less than 7kPa), and the real-time data is fed back to the strategy decision and parameter optimization layer when adjusting step by step. The safety and stability state of the system is judged, and after the state is stable, the execution is continued until the control target of the decision layer is reached.

[0256] As Figure 5 The burner life prediction-driven control logic state diagram of the best embodiment of the present application is shown. The burner life prediction-driven control logic state involved in the present application is divided into four parts, including:

[0257] Through the high-precision burner life prediction model, the output of the life prediction model (such as the remaining life percentage RUL%, the life attenuation rate ΔRUL / Δt, the prediction confidence Conf, etc.) is taken as the input variable of the adaptive control strategy generation, forming a three-stage coupled dynamic control strategy, which is:

[0258] Strategy S501, when the remaining life of the burner is greater than 75% of the design value, and the confidence of the model is greater than 0.9, the burner is in good condition, and the control target is to maximize the production efficiency (maximize the effective gas (CO+H2) component), through the fuzzy control algorithm, control the oxygen-coal ratio, coal slurry concentration, gasifier load and other variables, while ensuring that the raw material quantity and oxygen-coal ratio fluctuate less, and maintaining system stability;

[0259] Strategy S502, when the remaining life of the burner is less than 75% of the design value, but greater than or equal to 50% of the design value, and the confidence of the model is greater than 0.7, the control target changes, and the burner life (i.e. the safety and stability of the gasifier) and the gasification efficiency (economic index) need to be considered, a Nash equilibrium game model is constructed, and a dynamic monitoring and real-time optimization adjustment mode is used to periodically detect and adjust the gasifier operation.

[0260] Strategy S503, when the remaining life of the burner is less than 50% of the design value, or the confidence of the model is less than 0.7, it means that the burner life is low and the prediction model cannot accurately predict, the control target is to minimize the life decay rate, and the reinforcement learning compensation strategy is triggered to match the planned shutdown period, ensuring that the burner degrades according to the plan while ensuring system efficiency.

[0261] Strategy S504, when the burner life prediction driven model predicts that the burner life decay rate breaks through the threshold (greater than 0.1% per hour), it means that the burner may have suffered irreversible damage and is rapidly degrading, and emergency shutdown protection measures need to be taken to ensure system safety.

[0262] The specific steps include:

[0263] Real-time acquisition and fusion of multi-modal data. The present application deploys a multi-dimensional sensor network on the burner body and adjacent process pipelines, which includes eight broad-spectrum high-frequency heat flow sensors (sampling rate ≥ 100 Hz) and two sets of in-situ infrared thermal imagers (thermal sensitivity ≤ 0.03℃), forming a ring-shaped monitoring array. Time synchronization (clock synchronization error < 0.8 μs) of sensor data is achieved through OPC UA protocol, and a data space mapping model based on process physical coordinates is established. To solve the problem of data loss under harsh conditions of the gasifier, a dual-channel redundant transmission architecture is designed, and a generative adversarial network (GAN) is used to reconstruct high-fidelity local data that fails to transmit, ensuring that the original data integrity rate is > 99%. In addition, offline detection data (such as coal ash viscosity-temperature curve) from the coal quality laboratory is integrated with online sensor data to form a cross-modal feature engineering, an improved chi-square test algorithm is used to screen key correlation factors (such as "sulfur content-coating wear acceleration index"), and a feature library for life prediction is constructed.

[0264] Step two: sensor data edge computing. Based on the edge computing server, the collected multi-modal data is pre-processed, such as performing data denoising (wavelet threshold filtering), time alignment (PTP clock synchronization), missing value compensation (Kriging interpolation), etc.

[0265] Step three: remaining life prediction and credibility evaluation. Using industrial operation data, burner structure parameters, burner maintenance records and other multi-modal data, a burner life prediction model framework is constructed, a multi-modal fusion deep learning model is established, and multi-modal, multi-scale and online analysis of burner life is realized. Using the constructed model, online prediction and credibility evaluation are performed. Credibility evaluation can be performed by expert evaluation of the burner state during shutdown maintenance, and the burner state is determined according to the guidance of engineers and manufacturers, and the prediction results of the model are compared.

[0266] Step four: dynamic identification of burner degradation state. Based on the step response of the system controller and the burner life prediction model, a burner degradation state and control system control dynamic identification based on physical field-model coupling is constructed, and a control system dynamic response equation is established.

[0267] Step five: dynamic generation of multi-objective control strategy. Design an optimal controller to convert the multi-objective optimization problem into a solution system composed of three interrelated modules:

[0268] Adaptive weight allocation network, define the objective function of the multi-objective optimization strategy, including maximizing the effective gas yield η, minimizing the burner life decay rate v and minimizing the specific coal consumption χ (the mass of coal consumed per 1000 Nm3 of (CO+H2) produced), specifically:

[0269]

[0270] The "constraint weighting method" is used to convert multiple optimization objectives into a single objective function J through weighted summation, and process constraints (such as temperature, pressure, oxygen-coal ratio limits, etc.) are considered during optimization. Specifically:

[0271] According to the real-time economic demand, the weight ratio between life and efficiency targets is automatically adjusted, for example, when the burner life is high, ω1 is increased, and when the burner life is low, ω2 is increased;

[0272] Quantum evolutionary algorithm module, quickly search for the optimal solution set in an 80-dimensional parameter space (calculation time ≤ 3 seconds);

[0273] Robustness verification unit, call HyperStudy software to carry out million-level random disturbance simulation, ensure that the solution set meets the process constraint probability > 99.97%. For the frequent variable load scene, the "prediction rolling window + feedback correction" dual-mode control strategy is proposed, and the adjustment instruction set is refreshed every 15 seconds, which ensures the tracking accuracy of the working condition while avoiding frequent action of the actuator.

[0274] Step six: strategy decision scheme generation and parameter optimization. Construct "life-control mapping function" and "RUL prediction driven three-stage control logic", that is,

[0275] When the remaining life of the burner is greater than 75% of the design value, and the confidence of the model Conf is greater than 0.9, the burner is in good condition, and the control target is to maximize the effective gas yield (maximize the effective gas (CO+H2) component). Through fuzzy control algorithm, adopt fuzzy PID controller, set the target weight of multi-objective optimization function J, such as "efficiency priority" weight factor ω1=0.8, "life protection" weight factor ω2=0.2. Control oxygen-coal ratio, coal slurry concentration, gasifier load and other variables, on the premise of ensuring that the burner temperature gradient ΔT / Δt<set threshold (25 ℃ / min), gradually increase the oxygen-coal ratio to the upper limit of the process, while ensuring that the raw material quantity and oxygen-coal ratio fluctuate small, maintain system stability;

[0276] When the remaining life of the burner is less than 75% of the design value, but greater than or equal to 50% of the design value, and the confidence of the model Conf is greater than 0.7, the control target changes, and the burner life (i.e. the safety and stability of the gasifier) and the gasification efficiency (economic index) need to be considered. Construct life benefit and efficiency benefit functions, which are:

[0277] U life =f(ΔRUL / Δt, historical maintenance cost)

[0278] U eff =f(η,χ)

[0279] Start Nash equilibrium game model (such as Lemke-Howson algorithm), real-time calculation of a set of optimal control parameter combinations between "life benefit" and "efficiency benefit" two game parties, which cannot obtain higher benefits by changing the strategy of any one party, and periodically detect and adjust the gasifier operation;

[0280] When the remaining life of the burner is less than 50% of the design value, or the confidence of the model Conf is less than 0.7, it indicates that the burner life is low and the prediction model cannot accurately predict, the control target is to minimize the life decay rate, and the reinforcement learning compensation strategy (such as DQN or PPO) is triggered to match the planned shutdown cycle, the agent is trained according to the historical operation data to learn the optimal life extension control strategy, and the reward or punishment is obtained according to the life decay.

[0281] When the burner life prediction driving model predicts that the burner life decay rate breaks through the threshold (greater than 0.1% per hour), it indicates that the burner may have irreversible damage and is rapidly degrading, and emergency shutdown protection measures need to be taken to ensure system safety.

[0282] Step seven: closed-loop feedback and model evolution. Based on the dynamic response equation completed in step four and the strategy decision scheme in step six, a closed-loop feedback regulation and model updating system is established: (1) the control scheme and system operation parameters are proposed based on the model prediction results, and the multi-objective control strategy is used to realize the automatic adjustment of the system; (2) during the automatic regulation and control of the system, the operation parameters and system product index results are fed back to the parameter optimization model to optimize the adjustment parameters and ensure the safety and stability of the system; (3) the factory-level case accumulation layer is established in the MES system to archive typical working condition adjustment schemes and build an industrial knowledge graph with a sample size of 300GB; (4) the cloud deep optimization layer calls the supercomputing center resources (1000+ computing nodes) every quarter to implement comprehensive model evolution; (5) the time for a single prediction of the burner life model is 1 day, during which the burner life model result is automatically called to generate a strategy decision scheme to guide the automatic regulation and control of the system; (6) the overall update and optimization period of the burner life model is about 40-45 days, which is about half of the regular burner life cycle.

[0283] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0284] Application example:

[0285] Intelligent control system for gas flow bed gasifier burner

[0286] Implementation device configuration

[0287] Gasifier: multi-nozzle opposed type (4 groups of burners), design pressure 6.5MPa, temperature 1350-1500℃

[0288] Data acquisition system:

[0289] Eight-channel armored thermocouple (0°-360° annular distribution, accuracy ±0.5℃)

[0290] Coal quality online analyzer (laser-induced breakdown spectroscopy technology, LiBS)

[0291] Edge computing platform: Huawei Atlas 500Pro intelligent edge server (AI computing power 16TOPS)

[0292] Control actuator: ABB electric regulating valve (positioning accuracy ±0.1%), Fusheng high-pressure coal slurry pump (flow accuracy ±0.8%)

[0293] Specific implementation steps

[0294] Step one: Real-time fusion of multi-modal data

[0295] Deploy a data acquisition network on #3 burner (diameter φ350mm, CoCrAlY coating thickness 1.2mm) and its adjacent process pipe section:

[0296] 1. Eight-channel thermocouple measures the temperature field in the diameter direction of the burner throat (ΔT_max=78℃), and the data is transmitted through the 5G-MEC edge gateway (delay ≤8ms);

[0297] 2. Coal quality laboratory off-line determination of the calorific value of the coal fed on the same day (22.1MJ / kg) and the ash melting point (FT=1320℃), to generate the sulfur content-coating wear correlation factor (0.87);

[0298] Technical advantage: Data loss rate reduced from 13.5% of traditional systems to 0.7%, and no missing values in all feature dimensions during 960h continuous operation;

[0299] Step two: Sensor data edge computing

[0300] 1. Data preprocessing:

[0301] (1) Use db4 wavelet packet to denoise the temperature signal (eliminate fan vibration interference)

[0302] (2) Apply Kriging interpolation to complete the missing section of short-wave infrared data transmission (confidence ≥97%)

[0303] 2. Feature engineering: Generate the following key indicators:

[0304] (1) Burner eccentricity index (BEI) = (Tm_max-Tm_min) / Tm_avg x 100 (BEI measured under certain working conditions =5.3%)

[0305] (2) Coal ash deposition rate = LiBS detected ash content x air velocity correction coefficient

[0306] Step three: Residual life prediction and evaluation

[0307] 1. Predictive model construction:

[0308] (1) Establish a 5-layer residual fully connected network (ReLU activation, Dropout rate 0.25);

[0309] Input features: 26 dimensions (including implicit layer cross terms such as coal sulfur content x cooling water pH);

[0310] (2) Online prediction:

[0311] When the load rises to 105%, the model predicts RUL = 1420h (confidence 92.3%), and through actual maintenance, it is found that the average thinning of the coating is 0.15mm, with an error rate of 7.8% (satisfying the engineering error tolerance);

[0312] Step four: Burner degradation dynamic identification

[0313] 1. System identification experiment:

[0314] Step-up load 7% (from 85% to 92%), collect response curve CTC (characteristic time constant), and establish identification to obtain burner dynamic equation.

[0315] Actual measurement data: The maximum erosion rate during coal type switching is 0.038mm / h, with a deviation of 7.8% from the calculated value 0.041mm / h.

[0316] Step five: Multi-objective control strategy generation

[0317] 1. Quantum evolutionary algorithm solution:

[0318] Decision variables: oxygen-coal ratio, coal slurry concentration, synthetic gas compressor speed, etc. 15 items; time-consuming 2.6 seconds to generate Pareto optimal solution set.

[0319] Economic benefit comparison: After adopting this strategy, the coal consumption per ton of olefin is reduced by 1.8%, and the annual raw material cost is saved by about 5.6 million yuan.

[0320] Step six: Parameter coordination optimization and closed-loop feedback

[0321] 1. Three-stage control execution and emergency plan:

[0322] Mode A (RUL≥75% and Conf>0.9): Fuzzy PID control oxygen-coal ratio (fluctuation rate ≤±0.3%);

[0323] Mode B (50%≤RUL<75% and Conf>0.7): Nash game to solve efficiency-life balance point (reduction rate control at 2-4%);

[0324] Mode C (RUL < 50% or Conf < 0.7): Implement reinforcement learning compensation strategy (12% reduction in production rate, 93% match rate in repair cycle);

[0325] Mode D (ARUL / At > 0.1% / hour): Start emergency parking protection strategy.

[0326] 2. Implementation effect:

[0327]

[0328] 3. Model evolution effect: After 42 days of first round training, the RUL prediction error is reduced from 14.5% to 9.1%

[0329] This embodiment realizes:

[0330] 1. Accurate life management: Break through the traditional experience maintenance mode, realize individualized regulation and control of the remaining life of key components;

[0331] 2. Value balance optimization: First achieve dynamic optimization of coal conversion rate and material loss rate at the control layer;

[0332] 3. Continuous evolution of knowledge: Form an industrial knowledge graph, continuously expand effective working condition cases, and support rapid deployment of new product lines.

[0333] As Figure 6 The hardware structure schematic diagram of the electronic equipment provided by the application is shown in the figure, which comprises:

[0334] at least one processor 601; and,

[0335] a memory 602 in communication connection with the at least one processor 601; wherein,

[0336] The memory 602 stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the gas flow bed gasification furnace control method as described above.

[0337] Figure 6 Take the processor 601 as an example.

[0338] The electronic equipment can further comprise an input device 603 and a display device 604.

[0339] The processor 601, the memory 602, the input device 603 and the display device 604 can be connected through a bus or other means, and in the figure, connection through a bus is taken as an example.

[0340] The memory 602, as a non-volatile computer readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions / modules corresponding to the gas flow bed gasification furnace control method in the embodiments of the present application, for example, the method flow shown in the above formula (1). The processor 601 executes various function applications and data processing by running the non-volatile software programs, instructions and modules stored in the memory 602, that is, implements the gas flow bed gasification furnace control method in the above embodiments. Figure 1 、 Figure 2 The processor 601 executes various function applications and data processing by running the non-volatile software programs, instructions and modules stored in the memory 602, that is, implements the gas flow bed gasification furnace control method in the above embodiments.

[0341] The memory 602 can include a program storage area and a data storage area, wherein the program storage area can store an operating system and application programs required by at least one function; the data storage area can store data created according to the use of the gas flow bed gasification furnace control method, etc. In addition, the memory 602 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 602 can optionally include a memory remotely arranged with respect to the processor 601, and these remote memories can be connected to the device executing the gas flow bed gasification furnace control method through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0342] The input device 603 can receive input user clicks and generate signal inputs related to user settings and function controls of the gas flow bed gasification furnace control method. The display device 604 can include a display screen and other display equipment.

[0343] When the one or more modules are stored in the memory 602 and are run by the one or more processors 601, the gas flow bed gasification furnace control method in any of the above method embodiments is executed.

[0344] The present application predicts the predicted remaining life of the gas flow bed gasification furnace burner by constructing a burner life prediction model; at the same time, according to the predicted remaining life, the parameters of the optimization target equation are adjusted, and based on the adjusted optimization target equation, the gas flow bed gasification furnace is controlled. The present application realizes real-time prediction of burner life by constructing a high-precision multi-modal burner life prediction model, breaks through the limitation of single parameter linear adjustment, adjusts the parameters of the optimization target equation according to the predicted remaining life, realizes the automatic coupling between the burner life prediction and the control strategy, and makes the control strategy accurately adapt to the equipment degradation process.

[0345] An embodiment of the present application provides a storage medium storing computer instructions, when a computer executes the computer instructions, all steps of the gas flow bed gasification furnace control method as described above are executed.

[0346] In the context of the present disclosure, the storage medium can be a tangible medium which can contain or store the program for use by or in connection with the instruction execution system, apparatus, or device. The storage medium can be a machine-readable signal medium or a machine-readable storage medium. Alternatively, the storage medium can be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0347] An embodiment of the present application provides a computer program product comprising computer programs / instructions which, when executed by a processor, implement the control method of the gas flow bed gasifier as described above.

[0348] The above-described embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. An entrained flow gasifier control method, characterized by, The method comprises: constructing a burner life prediction model, training the burner life prediction model; inputting current data of the entrained-flow gasifier into the burner life prediction model to obtain a predicted remaining life of the burner of the entrained-flow gasifier; constructing an optimization objective equation of the entrained-flow gasifier; adjusting parameters of the optimization objective equation according to the predicted remaining life, and controlling the entrained-flow gasifier based on the adjusted optimization objective equation.

2. The control method of a gas flow bed gasification furnace according to claim 1, characterized by, The construction of the optimization objective equation of the entrained-flow gasifier comprises: constructing an optimization objective equation comprising three optimization objectives, wherein the first optimization objective is to maximize the effective gas yield, the second optimization objective is to minimize the burner life decay rate, and the third optimization objective is to minimize the specific coal consumption, and the parameters of the optimization objective equation comprise a first weight of the first optimization objective, a second weight of the second optimization objective, and a third weight of the third optimization objective.

3. The control method of a gas flow bed gasification furnace according to claim 2, characterized by, The construction of the optimization objective equation comprising three optimization objectives comprises: The optimization objective equation including three optimization objectives is constructed as: J = ω1*η + ω2*v + ω3*χ + ∑ i λ i * constraint term i wherein: ω1 is a first weight, ω2 is a second weight, ω3 is a third weight, λ i is a penalty coefficient of the ith constraint term, η is an effective gas production rate, v is a burner life attenuation rate, and χ is a specific coal consumption.

4. The control method of a gas flow bed gasification furnace according to claim 2, characterized by, The adjustment of the parameters of the optimization objective equation according to the predicted remaining life comprises: when the predicted remaining life is greater than or equal to a first percentage of the design value of the burner life, and the confidence of the burner life prediction model is greater than or equal to a first confidence, increasing the first weight and / or the third weight of the optimization objective equation, and decreasing the second weight of the optimization objective equation.

5. The control method of a gas flow bed gasification furnace according to claim 2, characterized by, The adjustment of the parameters of the optimization objective equation according to the predicted remaining life comprises: when the predicted remaining life is less than a first percentage of the design value of the burner life and greater than or equal to a second percentage, and the confidence of the burner life prediction model is greater than or equal to a second confidence, constructing a life benefit function based on the calculation of the life benefit based on the burner life decay rate, constructing an efficiency benefit function based on the calculation of the effective gas yield and the specific coal consumption, performing a game between the life benefit function and the efficiency benefit function, and adjusting the first weight, the second weight, and / or the third weight of the optimization objective equation according to the game result, wherein the second percentage is less than the first percentage.

6. The control method of a gas flow bed gasification furnace according to claim 2, characterized by, The adjustment of the parameters of the optimization objective equation according to the predicted remaining life comprises: when the predicted remaining life is less than a second percentage of the design value of the burner life, or the confidence of the burner life prediction model is less than a second confidence, decreasing the first weight and / or the third weight of the optimization objective equation, and increasing the second weight of the optimization objective equation.

7. The control method of a gas flow bed gasification furnace according to claim 2, characterized by, Further comprising: when the burner life prediction model predicts that the burner life decay rate is greater than a rate threshold, starting a shutdown protection measure of the entrained-flow gasifier.

8. An electronic device, comprising: The method comprises: at least one processor; and a memory in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for controlling the entrained-flow gasifier according to any one of claims 1 to 7.

9. A storage medium, characterized by The storage medium stores computer instructions, and when the computer executes the computer instructions, all steps of the method for controlling the entrained-flow gasifier according to any one of claims 1 to 7 are performed.

10. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instructions, when executed by the processor, implement the control method of the entrained-flow gasifier according to any one of claims 1 to 7.

Images