Fly ash quality control method for thermal power generating unit

Abstract

The invention discloses a thermal power generating unit fly ash quality control method, and belongs to the technical field of control optimization, and the method comprises the steps: collecting and integrating working condition data from a thermal power generating unit system to form a standardized data set, and building a prediction model based on the standardized data set to predict the key indexes of fly ash in the future, then, a multi-objective optimization engine containing an expert rule and a particle swarm optimization algorithm and a digital twinborn simulation technology are utilized to generate and optimize a control instruction covering electric field operation, boiler combustion organization, graded conveying path switching and flue gas conditioning; and finally, the model is updated through instruction execution and an online incremental learning algorithm to form closed-loop adaptive control. According to the method, the dynamic prediction model is constructed, the problems that a traditional control method depends on artificial experience and response lags behind are solved, the perspectiveness and initiative of quality control are improved, and meanwhile the contradiction that the high-quality fly ash output rate and the system operation energy consumption are difficult to consider at the same time is solved through multi-target collaborative optimization.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of control optimization, and particularly relates to a fly ash quality control method for a thermal power unit. BACKGROUND

[0002] Fly ash is the main solid waste produced by a coal-fired thermal power unit and is an important industrial byproduct for realizing circular economy. With the increasing requirement of downstream industries such as building materials on the performance of raw materials, fine quality control of fly ash to realize high-value resource utilization has become an important direction for energy saving and efficiency improvement and green development of power plants. The core principle is to ensure quality stability and improve economic benefits. However, in the process of fly ash quality control of the thermal power unit, due to the frequent fluctuation of the coal quality entering the furnace, the complex and changeable boiler combustion conditions, and the strong coupling and nonlinear characteristics among the systems such as pulverizing and dust removal, the traditional control method mainly relies on the historical experience of the operation personnel and the lagging adjustment of the after-sampling test results of the fly ash. This method cannot accurately predict the future fly ash quality change trend in real time based on the industrial big data of the unit operation. Especially in the face of coal switching, unit load adjustment and other working condition disturbances, the existing control method is difficult to carry out forward-looking and coordinated regulation, often leading to the key indicators such as unburned carbon content and fineness of fly ash exceeding the high-level standard, causing the resource value of fly ash to be greatly reduced, even becoming solid waste, and bringing economic losses and environmental protection pressure to the power plant. SUMMARY

[0003] In order to overcome the technical problems described in the background, the application provides a fly ash quality control method for a thermal power unit, which solves the problem of relying on artificial experience and response lag of the traditional control method by constructing a dynamic prediction model, improves the forward-looking and initiative of quality control, and solves the contradiction between high-quality fly ash production rate and system operation energy consumption through multi-objective collaborative optimization.

[0004] The technical scheme of the application is as follows: a fly ash quality control method for a thermal power unit, comprising the following steps: S1, collecting historical operation data and real-time working condition data of the thermal power unit, and forming a unified original working condition data package after time alignment and preliminary cleaning; S2, performing data cleaning, integration and standardization on the original working condition data package to generate a standardized data set for model construction; S3, based on the standardized data set, constructing and solidifying a prediction model capable of reflecting the dynamic characteristics of the system, and generating a prediction result of the key quality indicators of the future fly ash through the prediction model; S4, generating control instructions including power plant operation, combustion organization, conveying path and flue gas conditioning according to the prediction result and combining with a preset processing rule; S5, send control instructions to the corresponding executing mechanism in the thermal power generating unit and execute, dynamically adjust the running state of the system; S6, based on the new running state data generated after adjustment, update the internal parameters of the prediction model using online incremental learning algorithm to continuously optimize the prediction accuracy of the model; S7, real-time monitoring of the key response data of the system, and generating a control effect evaluation report containing key performance indicators; S8, when the key indicators in the evaluation report are abnormal, trigger the early warning mechanism and generate the correction control instruction for abnormal disposal; S9, send the correction control instruction to the executing mechanism of the system for rapid intervention and closed-loop iterative adjustment; S10, structurally store the optimal control parameters, system response data and evaluation results, and output to the monitoring platform.

[0005] Further, step S1 includes the following steps: S11, collect historical operation data and real-time working condition data from the distributed control system, fuel management information system, laboratory information management system and environmental protection monitoring system of the thermal power generating unit through open platform communication protocol, application program interface or database direct connection; S12, collect core working condition parameters reflecting the real-time running state of the unit at a high sampling frequency, including boiler and combustion related data, pulverizing system related data and dust removal equipment related data; S13, obtain the quality analysis data of the coal entering the furnace and the quality data of the fly ash after artificial sampling and testing as the target value for subsequent model training at a low sampling frequency; S14, time stamp alignment of high-frequency working condition data and low-frequency quality data, and use Lagrange interpolation method to fill in missing values and eliminate abnormal values; S15, integrate the processed data into a structured pre-processing working condition data package, and pass it to step S2.

[0006] Further, step S2 includes the following steps: S21, obtain the pre-processing working condition data package generated in step S1; S22, normalize the pre-processing working condition data package to eliminate the influence of different physical dimensions, and generate a normalized feature matrix; the normalization processing uses the following formula: ; Wherein, is the normalized data, is the original data value, is the minimum value in the historical data set, is the maximum value in the historical data set; S23, the Pearson correlation coefficient analysis and LASSO regression model based on L1 regularization are combined for feature screening, and key features strongly related to fly ash quality are extracted to generate a feature optimized data set; S24, principal component analysis dimension reduction processing is performed on the feature optimized data set, and principal components with a cumulative variance contribution rate reaching a preset threshold are extracted to generate a standardized data set; S25, verify the data quality and consistency of the standardized data set, and output a data quality verification report.

[0007] Further, step S3 includes the following steps: S31, the standardized data set output by S2 is divided into a training set, a validation set and a test set according to the time sequence; S32, a long short-term memory network model is constructed for the unburned carbon content index; S33, a gradient boosting decision tree model is constructed for the fly ash characteristic particle size index; S34, a gradient boosting decision tree model is constructed for the pozzolanic activity index; S35, the training set is used to train the initial prediction model, and the model hyperparameters are optimized by grid search combined with cross-validation. Finally, the performance of the optimized model is evaluated using the test set, and the final prediction model is solidified; S36, the fly ash production rate is obtained by real-time calculation, which provides decision basis for classification conveying and ash storage capacity management; the fly ash production rate calculation formula is: ; Where, is the theoretical output rate of fly ash per unit time, is the real-time coal quantity, is the ash content of the coal, is the unburned carbon content of fly ash provided by the prediction model in S32; S37, in the real-time operation stage, the working condition data is input into the three solidified final prediction models, and the calculation results of S36 are combined to output a comprehensive prediction result data package containing four key indicators within the next 30 minutes.

[0008] Further, step S4 includes the following steps: S41, obtain the future 30-minute fly ash quality comprehensive prediction result generated in step S3; S42, generate an initial control instruction based on a multi-objective optimization rule engine including a strategy generation layer and a parameter optimization layer; the parameter optimization layer takes a target function as the optimization target, and searches in the feasible region of the control variable by a particle swarm optimization algorithm, and the target function The expression is: ; wherein, is the predicted high-rank fly ash yield, is the predicted ton of ash collection energy consumption, and are weight coefficients; S43, input the initial control instruction into a digital twin model composed of a boiler combustion module and an electrostatic precipitation module for simulation, to generate simulation working condition response data; S44, analyze the deviation of the simulation working condition response data from the preset fly ash quality target parameters, and if the deviation exceeds the allowable range, generate an instruction optimization correction amount; S45, correct the initial control instruction according to the instruction optimization correction amount, and output the structured final control instruction; S46, provide a theoretical basis for the electrostatic precipitator operation optimization instruction in the final control instruction, which is based on the following electrostatic precipitator collection efficiency calculation formula: ; wherein, is the theoretical collection efficiency, is the total area of the dust collection plate, is the flue gas flow, is the dust driving speed.

[0009] Further, step S5 includes the following steps: S51, issue the final control instruction to the thermal power generating unit distributed control system and the PLC controller through the industrial control network; S52, control the high-voltage power supply and the rapping control system of the electrostatic precipitator to optimize the voltage of each electric field and the rapping cycle; S53, send a combustion fine-tuning instruction to the boiler control system to fine-tune the combustion organization parameters of the air-coal ratio and the burner swing angle; S54, control the path switching valve of the pneumatic ash conveying system to switch the classified conveying path according to the predicted fly ash quality; S55, real-time collect equipment state feedback data in the instruction execution process, and generate an execution log data package containing a timestamp and an execution result.

[0010] Further, step S6 includes the following steps: S61, obtain the execution log data package generated in step S5, the new running state data after system adjustment, and the corresponding new fly ash quality test data; S62, take the new test data as the true label, calculate the deviation between it and the model prediction result, and generate a model prediction error analysis report; S63, determine whether the preset model update trigger condition is met based on the model prediction error analysis report; S64, when the trigger condition is met, retrain the prediction model using new data through an online incremental learning algorithm matched with the model type; the model update strategy includes: C1, for a long short-term memory network model, adopt a periodic fine-tuning strategy, and the weight update formula is: ; wherein, is the updated weight matrix, is the current weight matrix, is the fine-tuning learning rate, is the gradient of the loss function to the weight; C2, for a gradient boosting decision tree model, adopt an enhanced learning strategy with residual correction, and the new model prediction logic is: ; wherein, is the updated prediction value, is the prediction value of the old model, is the prediction value of the correction model trained on the new data residual; S65, store the updated prediction model and its version information, training date and other metadata to the model library.

[0011] Further, step S7 includes the following steps: S71, collect key response data after execution of control instructions using a thermal power unit distributed control system acquisition system; S72, calculate key performance indicators, including high-grade fly ash yield, fly ash comprehensive utilization value index, ton of ash collection energy consumption and environmental protection emission compliance rate; S73, generate a control effect evaluation report based on a multi-dimensional KPI threshold comparison algorithm, the report evaluates the economy of the dust removal system by calculating the ton of ash collection energy consumption, and the calculation formula of the ton of ash collection energy consumption is: ; wherein, is the ton of ash collection energy consumption, is the total power of the electrostatic precipitator body operation, is the power of the pneumatic ash conveying system operation, is the actual collection rate of fly ash per unit time; S74, when any key performance indicator exceeds the preset safety or economic threshold range, mark it as an abnormal state in the evaluation report; S75, automatically generate and output the evaluation report to the power plant intelligent environmental protection monitoring platform at a preset period.

[0012] Further, step S8 comprises the following steps: S81, triggering the early warning mechanism when the abnormal state data is marked in the evaluation report of step S7, and the triggering of the early warning mechanism is based on the fly ash quality deterioration index , and the calculation formula is: ; Wherein, is the fly ash quality deterioration index, and are the real-time values of the unburned carbon content and the fineness, respectively, and are the target values thereof, and are weight coefficients; S82, using the Bayesian network to perform root cause diagnosis to reversely infer the most likely fault source, and generate an abnormal reason diagnosis report; S83, based on the abnormal reason diagnosis report, and matching the preset expert rule knowledge base, a revised control instruction draft is generated; S84, the potential impact of the revised control instruction draft is audited through the control strategy safety simulation module, and the final revised control instruction is output after confirming that there is no conflict; S85, the complete early warning triggering, reason diagnosis, instruction generation and execution process are recorded to the system event log.

[0013] Further, step S9 comprises the following steps: S91, the final revised control instruction generated in step S8 is issued to the actuator module of the distributed control system of the thermal power generating unit; S92, the actuator module schedules and executes the revised control task according to the preset safety and process priority; S93, through the closed-loop adjustment of the quick response, until the monitoring data shows that the fly ash quality deterioration index is restored to within the threshold value, and real-time operation data of the intervention process is generated; S94, verifying whether the real-time operation data of the intervention process meets the expectation, so as to evaluate the effectiveness of the revised control instruction, and outputting a revised intervention effect report; S95, correlating and analyzing the revised intervention effect report and the abnormal reason diagnosis report, which is used to optimize the expert rule knowledge base and the revised instruction generation strategy in step S8.

[0014] The present application has the following beneficial effects due to the adoption of the above-mentioned technology.

[0015] 1. In the process of quality control of fly ash in thermal power units, the application can accurately predict the unburned carbon content, particle size distribution, and volcanic ash activity of fly ash in the future period by constructing a dynamic prediction model based on multi-source heterogeneous industrial big data, deeply mining and extracting features from massive historical and real-time data, solving the core problem of unstable fly ash quality caused by frequent fluctuations in coal quality and changes in unit operating conditions, and significantly improving the foresight and initiative of quality control, providing a solid foundation for the high-value resource utilization of fly ash.

[0016] 2. In the process of control strategy generation and optimization, by introducing a multi-objective optimization engine combining expert rule base and meta-heuristic algorithm, and using digital twin model for closed-loop simulation and pre-evaluation of initial control instructions, the application can optimize the output rate of high-quality fly ash and the energy consumption per ton of fly ash collection, solve the contradiction between quality and efficiency in traditional single control strategy, avoid the blindness and trial-and-error cost of control parameter adjustment, ensure the scientificity and economy of control instructions, and maximize the system operation efficiency.

[0017] 3. In the process of long-term operation and adaptive optimization of the system, the application uses online incremental learning algorithm to periodically fine-tune and correct the parameters and structure of the prediction model according to the latest running data generated after system adjustment, solves the problem of continuous decline in prediction accuracy caused by system characteristics drift due to factors such as equipment aging and environmental changes, ensures the long-term effectiveness and robustness of the control algorithm, and enables the entire control system to have the ability of continuous learning and autonomous evolution, thus being able to adapt to changing production conditions for a long time. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of the application.

[0019] Figure 2 is a flowchart of the original operating condition data acquisition and processing of the application.

[0020] Figure 3 is a flowchart of the construction and use of the prediction model reflecting the dynamic characteristics of the dust removal equipment of the thermal power unit of the application.

[0021] Figure 4 is a flowchart of the conversion of control instructions according to the prediction results of the application.

[0022] Figure 5 is a flowchart of the key performance indicator monitoring of the application. DETAILED DESCRIPTION

[0023] Example 1: As Figure 1As shown, the present application provides a method for controlling the quality of fly ash in a thermal power unit, comprising the following steps: S1, collecting historical operation data and real-time working condition data of the thermal power unit, and forming a unified original working condition data package after time alignment and preliminary cleaning; S2, data cleaning, integration and standardization of the original working condition data package to generate a standardized data set for model construction; S3, based on the standardized data set, a prediction model capable of reflecting the dynamic characteristics of the system is constructed and solidified, and the prediction results of the future fly ash key quality indicators are generated through the prediction model; S4, according to the prediction results, and combining with the preset processing rules, control instructions including power plant operation, combustion organization, conveying path and flue gas conditioning are generated; S5, the control instructions are sent to the corresponding execution mechanism in the thermal power unit and executed, and the running state of the system is dynamically adjusted; S6, based on the new running state data generated after adjustment, the internal parameters of the prediction model are updated using an online incremental learning algorithm to continuously optimize the prediction accuracy of the model; S7, real-time monitoring of the key response data of the system, and generating a control effect evaluation report containing key performance indicators; S8, when the key indicators in the evaluation report are abnormal, triggering the early warning mechanism and generating a correction control instruction for abnormal disposal; S9, the correction control instruction is sent to the execution mechanism of the system for rapid intervention and closed-loop iterative adjustment; S10, the optimal control parameters, system response data and evaluation results are stored in a structured manner, and output to the monitoring platform.

[0024] S1 includes the following steps: S11, through the open platform communication protocol, application program interface or database direct connection, the historical operation data and real-time working condition data are collected from the distributed control system, fuel management information system, laboratory information management system and environmental protection monitoring system of the thermal power unit; S12, through the interface of the distributed control system of the thermal power unit, core working condition parameters reflecting the real-time running state of the unit are collected at a sampling frequency of 1 second per time, including unit load, main steam pressure, main steam temperature, total coal supply, total primary air volume, total secondary air volume, furnace outlet oxygen content, furnace characteristic temperature, each layer burner swing angle These boiler and combustion related data also include each mill running current, mill outlet temperature, primary air powder mixture concentration, these related data of the pulverizing system, also include the inlet flue gas temperature of the electrostatic precipitator, the inlet flue gas flow, the primary side voltage and the primary side current of each electric field, and the dust removal equipment related data; S13, collecting detailed quality analysis data of the coal entering the furnace at a batch-per-day frequency through an interface of the fuel management information system and the laboratory information management system, including ash content, moisture content, volatile matter content, Hardgrove grindability index, and content of main oxides including silicon dioxide, aluminum oxide, and iron trioxide; and collecting fly ash quality data after manual sampling and testing at a shift-per-hour frequency through an interface of the laboratory information management system as target values for subsequent model training, including unburned carbon content of fly ash in each ash bucket, characteristic particle size measured by a laser particle size analyzer, and volcanic ash activity index measured according to the activity index test method in Appendix C of the national standard GB / T18736-2017; S14, time stamping the high-frequency operating condition data collected in S12 and the low-frequency quality data collected in S13, that is, associating the time stamps of the low-frequency coal quality and ash quality data to all high-frequency operating condition data in the effective time period, and simultaneously performing preliminary cleaning on the integrated data set, filling a small amount of missing values by using Lagrange interpolation method, and eliminating obvious abnormal values caused by sensor failure; S15, integrating the historical data and real-time data processed above, which contain complete operating condition characteristics and corresponding fly ash quality labels, into a structured "preprocessed operating condition data package" and delivering it to step S2 for subsequent standardization and feature engineering.

[0025] Step S2 includes the following steps: S21, obtaining the preprocessed operating condition data package generated in step S1; S22, performing normalization processing on the preprocessed operating condition data package to generate a normalized feature matrix; the normalization processing uses the following formula: ; wherein, is the normalized data, the value domain of which is distributed between [0, 1], eliminating the influence of different physical dimensions, is the original data value, representing the instantaneous measurement value of a specific operating condition parameter, is the minimum value of the corresponding operating condition parameter in the historical data set, is the maximum value of the corresponding operating condition parameter in the historical data set, so that operating condition parameters with different units and orders of magnitude such as boiler load, flue gas temperature, and electric field voltage are processed by dimensionless processing, providing a comparable data basis for subsequent comprehensive analysis of multi-dimensional features and model training; S23, a plurality of feature selection algorithms are combined for screening, Pearson correlation coefficient analysis is used to eliminate weakly correlated features with a correlation with the target variable below 0.1, and a LASSO regression model based on L1 regularization is used for embedded feature selection to punish the feature coefficients that contribute little to the target variable to zero, thus realizing automatic feature dimension reduction, extracting key features in the normalized feature matrix that are strongly correlated with the fly ash quality, and generating a feature optimized dataset; S24, performing principal component analysis dimension reduction processing on the feature optimized dataset, extracting principal components with a cumulative variance contribution rate of 95% as core variables, eliminating data redundancy and extracting core variables, and generating a standardized dataset; S25, verifying the data quality and consistency of the standardized dataset, and outputting a data quality verification report.

[0026] Step S3 includes the following steps: S31, the standardized dataset output by step S2 is divided into a training set, a validation set and a test set according to a time sequence in a ratio of 8:1:1, for subsequent model training, tuning and evaluation; S32, for the index of unburned carbon content which has strong time sequence dependence, a long short-term memory network model is constructed, wherein the input features are boiler load, air-coal ratio, furnace characteristic temperature, furnace outlet oxygen content and volatile matter of the coal fed into the furnace, the model includes an input layer of 5 input features, two hidden layers (128 neurons per layer), a Dropout layer (dropout rate p=0.2) and an output layer, and the model outputs the predicted unburned carbon content in the next 30 minutes; S33, for the fly ash characteristic particle size D50, a gradient boosting decision tree model is constructed, the input features are boiler load, current of each coal mill, concentration of primary air-pulverized coal mixture, furnace characteristic temperature and Hardgrove grindability index of the coal fed into the furnace, and the model outputs the predicted particle size D50 value in the next 30 minutes; S34, for the pozzolanic activity index, a gradient boosting decision tree model is constructed, the input features are furnace characteristic temperature, boiler outlet oxygen content and main oxide content in the ash fed into the furnace, and the model outputs the predicted pozzolanic activity index in the next 30 minutes; S35, the three initial prediction models constructed in S32, S33 and S34 are respectively trained using the training set, and in the training process, the validation set is used, and the grid search combined with cross-validation is used to optimize the hyperparameters of the long short-term memory network model and the key hyperparameters such as the number and depth of the gradient boosting decision tree, the test set is used to perform final performance evaluation on the optimized models, the model precision evaluation data including root mean square error and mean absolute error are generated, and after confirming that the average absolute percentage error of the model is below the preset precision requirement of 0.5%, the final prediction model is solidified; S36, the fly ash production rate is calculated in real time by a formula, and the fly ash production rate calculation formula is: ; wherein, is the theoretical output rate of fly ash per unit of time, and provides a decision basis for hierarchical conveying and ash storage capacity management, is the real-time coal quantity fed into the furnace, which is obtained from a distributed control system of a thermal power unit, is the ash content of the coal fed into the furnace, which is obtained from a fuel management system, is the unburned carbon content in the fly ash, which is provided by the prediction model in step S32, and thus the theoretical fly ash production rate in the next 30 minutes is calculated; S37, in the real-time operation stage, the working condition data at the current time and in the past period of time are input into the three final prediction models solidified in S35, and combined with the calculation of S36, a comprehensive prediction result data package containing the four key indicators of unburned carbon content, particle size distribution D50, volcanic ash activity index and fly ash production rate in the next 30 minutes is finally output, and is transmitted to step S4 for control decision.

[0027] Step S4 includes the following steps: S41, obtaining the fly ash quality prediction result in the next 30 minutes generated by step S3, including particle size distribution, unburned carbon content, volcanic ash activity index and fly ash production rate; S42, generating an initial control instruction set based on a multi-objective optimization rule engine, wherein the rule engine at least includes a strategy generation layer based on an expert rule base and a parameter optimization layer based on a meta-heuristic algorithm, wherein the strategy generation layer determines the adjustment direction and approximate range for each control variable according to the prediction result obtained in S41, forms a preliminary strategy combination, and the parameter optimization layer takes the adjustment range given by the strategy generation layer as the search space, and takes the objective function J as the optimization target, ; wherein, is the predicted high-grade fly ash output rate, for example, the unburned carbon content less than 3% is high-grade fly ash, is the predicted ton of ash collection energy consumption; and are weight coefficients that can be adjusted by the operator according to market demand and energy saving targets, and , for example, set and 0.7 and 0.3, respectively, indicating that more emphasis is placed on improving the output of high-quality fly ash, and the particle swarm optimization algorithm is used to search in the feasible region of each electric field voltage, burner swing angle, etc. Control variables to find the control parameter combination that minimizes the objective function J, forming the initial control instruction, wherein the particle swarm optimization algorithm parameters are set to a population size of 100, a maximum number of iterations of 200, an inertia weight of 0.8, an individual learning factor c1 of 1.5, and a social learning factor c2 of 1.5; wherein the preset quality control expert rule knowledge base includes the following rules: R1, if the predicted unburned carbon content exceeds 4.5%, trigger the enhanced burnout strategy, initially set the upper secondary air opening to increase by 2-5%, and the burner swing angle to adjust upward by 3-6 degrees; R2, if the predicted dust particle size D50 decreases and the unburned carbon decreases, causing the specific resistance to increase, trigger the anti-corona reversal strategy, initially set the final electric field voltage to decrease by 5-8 kV, and trigger the anti-wall sticking strategy, initially set the vibration period to shorten by 10-15%; R3, if the system monitors the coal quality change and predicts that the dust specific resistance will exceed , trigger the quality adjustment injection strategy, initially set the flue gas conditioning agent SO3 injection amount to a basic value calculated based on the predicted specific resistance; S43, simulate the initial control instruction with the digital twin model to generate simulation condition response data, wherein the digital twin model is a hybrid model including a boiler combustion module and an electrostatic precipitator module, the boiler combustion module is a lumped parameter model based on thermodynamic equilibrium and chemical reaction kinetics, inputs the combustion fine-tuning instruction such as the wind-coal ratio and the burner swing angle change, and outputs the change trend of the furnace temperature field and the flue gas composition, while the core of the electrostatic precipitator module is the electric precipitator capture efficiency calculation formula described in step S46, wherein the key parameter dust driving speed is no longer a constant, but is dynamically calculated by a small neural network with two hidden layers, 16 neurons per layer, and ReLU activation function according to the predicted fly ash particle size, unburned carbon content, and input electric field voltage. By inputting the initial control instruction into the digital twin model, the dynamic response of the main state parameters of the system within the next 30 minutes can be quickly simulated to generate simulation condition response data; S44, analyze the deviation of the simulation condition response data from the preset fly ash quality target parameter of unburned carbon content less than 4.0%, if the deviation exceeds the allowed range of plus or minus 5%, generate an instruction optimization correction amount; S45, according to the instruction optimization correction amount corrects the initial control instruction, and outputs the final control instruction, the final control instruction is a structured instruction set, at least including boiler combustion organization fine-tuning instruction, electric precipitator operation optimization instruction, hierarchical conveying path switching instruction and flue gas conditioning agent injection amount adjustment instruction, for example, the boiler combustion organization fine-tuning instruction adopts the key value of {“burner swing angle_A layer”: “+5 degrees”} and {“secondary air damper_C layer”: “+3%”}, the electric precipitator operation optimization instruction adopts the key value of {“electric field voltage_3 zone”: “58kV”}, {“electric field voltage_4 zone”: “52kV”} and {“rapping cycle_all”: “15 minutes”}, the hierarchical conveying path switching instruction is generated according to a multi-dimensional quality grading threshold matrix; for example, if the prediction result satisfies {‘unburned carbon content’ < 3.0%, ‘activity index’ > 85%}, the instruction is the key value of {“conveying path”: “1# premium ash storage”}; otherwise, if {‘unburned carbon content’ < 5.0%} is satisfied, the instruction is the key value of {“conveying path”: “2# ordinary ash storage”}, and the rest of the case is the key value of {“conveying path”: “3# waste storage”}, and the flue gas conditioning agent injection amount adjustment instruction adopts the key value of {“SO3 injection pump frequency”: “35Hz”}; S46, the theoretical collection efficiency of the electrostatic precipitator is calculated, which provides a theoretical basis for the optimization of electric field voltage and current, and the calculation formula of the collection efficiency of the electric precipitator is: ; Wherein, is the theoretical collection efficiency of the electric precipitator for dust, which is the core index for evaluating the performance of the electric precipitator, is the total area of the dust collection plate, which is a fixed design parameter of the equipment, is the flue gas flow through the electric field, which is extracted from the real-time flue gas flow data, is the dust driving speed, which is positively correlated with dust particle size, electric field strength, flue gas characteristics and other factors, and is a key variable for control optimization; thus, the electric field voltage is adjusted by the control system to change the dust driving speed , so as to achieve the target collection efficiency provides a core physical model support, ensuring the scientificity and effectiveness of the control instruction.

[0028] Step S5 includes the following steps: S51, the final control instruction is sent to the distributed control system of the thermal power generating unit and the PLC controller execution module through the industrial control network; S52, the execution module controls the high-voltage power supply and the rapping control system of the electrostatic precipitator, and optimizes each electric field voltage and rapping cycle; S53, the execution module sends the combustion fine-tuning instruction to the boiler control system to fine-tune the combustion organization parameters such as air-coal ratio and burner swing angle. S54, the executive module controls the path switching valve of the pneumatic ash conveying system, and switches the grading conveying path according to the predicted fly ash quality; S55, real-time acquisition of equipment state feedback data in the instruction execution process, generation of execution log data packet containing timestamp and execution result.

[0029] Step S6 includes the following steps: S61, obtaining the execution log data packet generated in step S5, the new running state data of the system adjustment, and the fly ash quality data obtained from the laboratory information management system corresponding to the new data timestamp; S62, taking the new actual test data as the real label, calculating the deviation between the prediction result of the model at the corresponding time and the prediction result, and generating a model prediction error analysis report containing the root mean square error and the average absolute percentage error; S63, based on the model prediction error analysis report, judge whether the preset model update trigger condition is met, and select the corresponding update strategy according to the model type, wherein the trigger condition is set to the week average MAPE of any model compared with the last week increases by more than 5% or the prediction error exceeds the allowed range of 10% for three consecutive days; S64, when the trigger condition is met, the prediction model is retrained by using the new data of the past 7 days through the online incremental learning algorithm matched with the model type, and the updated prediction model is generated; The model update strategy is: C1, for the long short-term memory network model in S32, a periodic fine-tuning strategy is adopted, that is, the existing model weight is used as the starting point of training, the data set containing the data of the last 30 days including the new data is used, and the retraining is performed with a small fine-tuning learning rate η=0.0001, and the model weight update formula is: ; Wherein, is the updated model weight matrix, representing the state of the model after learning new knowledge, represents the weight matrix of the current model, represents the fine-tuning learning rate, which is a hyperparameter controlling the step size of each update, and determines the speed of the model learning new data, represents the loss function of the weight gradient, which represents the change direction of the prediction error with respect to the weight parameter, and the retraining is performed for 5-10 rounds to prevent overfitting of the model to new data, while retaining the knowledge learned from historical data; C2. For the gradient boosting decision tree model in step S33, step S34 adopts an enhanced learning strategy with residual correction, that is, first input the new data into the old gradient boosting decision tree model , to obtain the predicted value ; calculate the residual error between the true value of the new data and the predicted value ; then train a correction model: take the operating condition features of the new data as input, and take the residual error as the target label, train a new gradient boosting decision tree model with a smaller scale, containing only 10-20 decision trees, and finally update the model to be the combination of the old model and the residual correction model, and the prediction logic is: ; wherein is the input operating condition data, so that the new model specially corrects the prediction deviation in the current operating condition on the basis of retaining the ability of the old model; S65, store the updated prediction model and its new version information, training date, and metadata such as the data set period used in the corresponding model library for subsequent deployment and version tracing.

[0030] Step S7 includes the following steps: S71, use the key response data after the execution of the control instruction collected by the distributed control system of the thermal power unit to collect system; S72, calculate the key performance indicators, including the high-grade fly ash yield, the comprehensive utilization value index of fly ash, the energy consumption per ton of fly ash collected, and the environmental protection emission compliance rate; S73, generate a control effect evaluation report according to the energy consumption per ton of fly ash collected based on the multi-dimensional key performance indicator threshold comparison algorithm, and the calculation formula of the energy consumption per ton of fly ash collected is: ; wherein, : the energy consumption per ton of fly ash collected is the core indicator for evaluating the economy of the dust collection system; : the total power of the electrostatic precipitator body, including the power consumption of the high-voltage power supply, the vibration motor, etc., represents the running power of the pneumatic ash conveying system, is the actual fly ash collection rate per unit time, which is calculated according to the balance degree of the collection amount of each ash bucket and the ash discharge record, so that the energy cost paid for obtaining unit mass of fly ash is quantified; through the energy consumption per ton of fly ash collected, it can be evaluated whether the control strategy can guarantee the quality of fly ash while achieving overall energy saving and consumption reduction of the system operation; ​S74, when any key performance indicator exceeds the preset safety or economic threshold range, mark it as an abnormal state in the evaluation report; S75, automatically generate and output the evaluation report to the power plant intelligent environmental protection monitoring platform at a preset period of every half hour.

[0031] Step S8 includes the following steps: S81, when the evaluation report of step S7 marks abnormal state data, trigger the early warning mechanism, and the early warning trigger condition is the fly ash quality deterioration index Determination: ; Wherein, represents the fly ash quality deterioration index, which triggers the early warning when its value exceeds the preset threshold, represents the real-time predicted value or measured value of the unburned carbon content of fly ash, represents the target control value of the unburned carbon content, represents the real-time predicted value or measured value of the fineness of fly ash; represents the target control value of the fineness of fly ash, and is a weight coefficient, which is set according to the influence degree of unburned carbon and fineness on the comprehensive utilization value of fly ash under different working conditions, so that the multi-dimensional quality deviation is converted into a single early warning trigger basis, making the early warning mechanism more sensitive and accurate; S82, identify the potential source through the root cause diagnosis algorithm, preferably use Bayesian network for diagnosis, so that by pre-constructing the conditional probability relationship between the reason nodes such as coal quality mutation, mill fault, burner blockage, electric field flashover and the result node of fly ash quality abnormality, by inputting real-time abnormal phenomena, the most possible fault source can be inferred in reverse, such as coal quality mutation, insufficient output of mill, failure of a specific electric field, etc., to generate an abnormal reason diagnosis report; S83, based on the abnormal reason diagnosis report, and matching the preset expert rule knowledge base, generate a revised control instruction draft; S84, review the potential impact of the revised control instruction draft through the control strategy safety simulation module, and output the final revised control instruction after confirming that there is no conflict, the instruction includes starting the emergency fly ash discharge program, triggering the electric field intensification vibration, switching the abnormal quality fly ash to the waste bin, etc. S85, record the complete early warning trigger, reason diagnosis, instruction generation and execution process to the system event log for traceability analysis.

[0032] Step S9 includes the following steps: S91, issue the final revised control instruction generated in step S8 to the actuator module of the distributed control system of the thermal power generating unit; S92, the actuator module schedules and executes the correction control task according to the preset safety and process priority, for example, switching the waste bin > intensifying the vibration > requesting load reduction; S93, through the rapid response of closed-loop adjustment, until the monitoring data shows that the fly ash quality degradation index is restored to within the threshold, and the real-time operation data of the intervention process is generated; S94, verify whether the real-time operation data of the intervention process meets the expectation, evaluate the effectiveness of the correction control instruction, and output the correction intervention effect report; S95, correlate and analyze the correction intervention effect report and the abnormal reason diagnosis report, which are used to optimize the expert rule knowledge base and the correction instruction generation strategy in step S8.

[0033] Step S10 includes the following steps: S101, the optimal control parameter set, the system response data packet, the model evaluation report and the correction intervention report are associated and integrated by using the foreign key, to form a closed-loop event data file; S102, the closed-loop event data file is written into the InfluxDB database and the PostgreSQL database of the power plant intelligent environmental protection monitoring platform in a structured format; S103, on the user interface of the power plant intelligent environmental protection monitoring platform, a fly ash quality real-time board, a control benefit analysis chart and a system operation history tracking interface are generated; S104, the historical data is regularly archived and backed up, to provide historical data set support for subsequent long-term evaluation of system performance and iteration of algorithm model.

Claims

1. A method for controlling the quality of fly ash from a coal-fired power plant, characterized by, Comprise the following steps: S1, collecting historical operation data and real-time working condition data of the thermal power generating unit, and forming a unified original working condition data package after time alignment and preliminary cleaning; S2, data cleaning, integration and standardization are performed on the original working condition data package to generate a standardized data set for model construction; S3, based on the standardized data set, a prediction model capable of reflecting the dynamic characteristics of the system is constructed and solidified, and a prediction result of future fly ash key quality indicators is generated through the prediction model; S4, according to the prediction result, and combining with a preset processing rule, a control instruction including power plant operation, combustion organization, conveying path and flue gas conditioning is generated; S5, the control instruction is sent to the corresponding executing mechanism in the thermal power generating unit and executed to dynamically adjust the operating state of the system; S6, based on the new operating state data generated after adjustment, an online incremental learning algorithm is used to update the internal parameters of the prediction model to continuously optimize the prediction accuracy of the model; S7, real-time monitoring of the key response data of the system is performed, and a control effect evaluation report containing key performance indicators is generated; S8, when the key indicators in the evaluation report are abnormal, a warning mechanism is triggered and a correction control instruction for abnormal treatment is generated; S9, the correction control instruction is sent to the executing mechanism of the system for rapid intervention and closed-loop iterative adjustment; S10, the optimal control parameters, system response data and evaluation results are stored in a structured manner and output to a monitoring platform.

2. A method for controlling the quality of fly ash of a thermal power unit according to claim 1, characterized in that: Step S1 includes the following steps: S11, collect historical operation data and real-time working condition data from the distributed control system, fuel management information system, laboratory information management system and environmental protection monitoring system of the thermal power generating unit through open platform communication protocol, application program interface or database direct connection; S12, collect core working condition parameters reflecting the real-time operating state of the unit at a high sampling frequency, including boiler and combustion related data, milling system related data and dust removal equipment related data; S13, obtain the quality analysis data of the coal entering the furnace and the fly ash quality data after artificial sampling and testing as the target value for subsequent model training at a low sampling frequency; S14, align the time stamps of the high-frequency working condition data and the low-frequency quality data, and use the Lagrange interpolation method to fill in the missing values and remove the abnormal values; S15, integrate the processed data into a structured pre-processing working condition data package and pass it to step S2.

3. The method for controlling the quality of fly ash of a thermal power unit according to claim 1, characterized in that: Step S2 includes the following steps: S21, obtain the pre-processing working condition data package generated in step S1; S22, normalize the pre-processing working condition data package to eliminate the influence of different physical dimensions and generate a normalized feature matrix; the normalization processing uses the following formula: ； wherein, is the normalized data, is the raw data value, is the minimum value in the historical data set, is the maximum value in the historical data set; S23, use Pearson correlation coefficient analysis and LASSO regression model based on L1 regularization for combined feature screening to extract key features strongly related to fly ash quality and generate a feature optimization data set; S24, perform principal component analysis dimension reduction processing on the feature optimization data set to extract principal components with a cumulative variance contribution rate reaching a preset threshold to generate a standardized data set; S25, verify the data quality and consistency of the standardized data set, and output a data quality verification report.

4. The method for controlling the quality of fly ash of a thermal power unit according to claim 1, characterized in that: Step S3 includes the following steps: S31, divide the standardized data set output by S2 into a training set, a validation set, and a test set according to a time sequence; S32, for the unburned carbon content index, a long short-term memory network model is constructed; S33, for the fly ash characteristic particle size index, a gradient boosting decision tree model is constructed; S34, for the index of the fly ash activity index, a gradient boosting decision tree model is constructed; S35, the training set is used to train the initial prediction model, and the model hyperparameters are optimized by using the grid search combined with the cross-validation method, and finally the performance of the optimized model is evaluated by using the test set, and the final prediction model is solidified; S36, the fly ash production rate is obtained by real-time calculation, which provides decision basis for classified conveying and ash storage capacity management; the fly ash production rate calculation formula is: ； wherein, is the theoretical output rate of fly ash per unit time, is the real-time amount of coal fed into the furnace, is the ash content of the coal fed into the furnace, is the unburned carbon content of fly ash provided by the prediction model in S32; S37, in the real-time running stage, the working condition data is input into the three solidified final prediction models, and the calculation results of S36 are combined to output a comprehensive prediction result data package containing four key indexes in the future 30 minutes.

5. A method for fly ash quality control of a thermal power unit according to claim 1, characterized in that: Step S4 includes the following steps: S41, obtain the future 30-minute fly ash quality comprehensive prediction result generated in step S3; S42. Generate initial control instructions based on a multi-objective optimization rule engine that includes a policy generation layer and a parameter optimization layer; the parameter optimization layer uses an objective function... To optimize the objective function, a particle swarm optimization algorithm is used to search within the feasible region of the control variables. The expression is: ； wherein, is the predicted high-rank fly ash yield, is the predicted ton of ash collection energy consumption, and is the weight coefficient; S43, input the initial control instruction into a digital twin model composed of a boiler combustion module and an electrostatic dust removal module for simulation to generate simulation working condition response data; S44, analyze the deviation of the simulation working condition response data from the preset fly ash quality target parameter, if the deviation exceeds the allowed range, generate an instruction optimization correction amount; S45, correct the initial control instruction according to the instruction optimization correction amount, and output a structured final control instruction; S46, provide a theoretical basis for the electrostatic precipitator operation optimization instruction in the final control instruction, which is based on the following electrostatic precipitator capture efficiency calculation formula: ； wherein, is the theoretical collection efficiency, is the total area of the dust collection electrode plate, is the flue gas flow rate, is the dust driving-in speed.

6. A method for fly ash quality control of a thermal power unit according to claim 1, characterized in that: Step S5 includes the following steps: S51, issue the final control instruction to the distributed control system and PLC controller of the thermal power generating unit through the industrial control network; S52, control the high-voltage power supply and the rapping control system of the electrostatic precipitator to optimize the voltage of each electric field and the rapping cycle; S53, send a combustion fine-tuning instruction to the boiler control system to fine-tune the combustion organization parameters of the wind-coal ratio and the burner swing angle; S54, control the path switching valve of the pneumatic ash conveying system to switch the classified conveying path according to the predicted fly ash quality; S55, real-time acquisition of equipment state feedback data in the process of executing the instruction, and generation of an execution log data package containing a time stamp and execution results.

7. A method for fly ash quality control of a thermal power unit according to claim 1, characterized in that: Step S6 includes the following steps: S61, obtain the execution log data package generated in step S5, the new running state data adjusted by the system, and the new fly ash quality test data corresponding thereto; S62, take the new test data as the true label, calculate the deviation between the new test data and the model prediction result, and generate a model prediction error analysis report; S63, based on the model prediction error analysis report, determine whether the preset model update trigger condition is met; S64, when the trigger condition is met, retrain the prediction model with new data by online incremental learning algorithm matching the model type; the model update strategy includes: C1, for long short-term memory network model, adopt periodic fine-tuning strategy, and the weight update formula is: ； wherein, is the updated weight matrix, is the current weight matrix, is the fine-tuning learning rate, is the gradient of the loss function with respect to the weights; C2, for gradient boosting decision tree model, adopt enhanced learning strategy with residual correction, and the new model prediction logic is: ； wherein, is the updated prediction value, is the prediction value of the old model, is the prediction value of the correction model trained on the new data residual; S65, store the updated prediction model and its version information, training date and other metadata to the model library.

8. A method for fly ash quality control of a thermal power unit according to claim 1, characterized in that: Step S7 includes the following steps: S71, collect key response data after the execution of control instructions by the distributed control system of thermal power generating unit; S72, calculate key performance indicators, including high-grade fly ash yield, fly ash comprehensive utilization value index, ton of ash collection energy consumption and environmental protection emission compliance rate; S73, generate control effect evaluation report based on multi-dimensional KPI threshold comparison algorithm, which evaluates the economy of dust removal system by calculating ton of ash collection energy consumption, and the calculation formula of ton of ash collection energy consumption is: ； wherein, is the energy consumption for fly ash collection, is the total power for ESP body operation, is the power for pneumatic ash conveying system operation, is the actual fly ash collection rate per unit time; S74, when any key performance indicator exceeds the preset safety or economic threshold range, mark it as an abnormal state in the evaluation report; S75, automatically generate and output the evaluation report to the intelligent environmental protection monitoring platform of power plant at preset period.

9. A method for fly ash quality control of a thermal power unit according to claim 1, characterized in that: Step S8 includes the following steps: S81, when the abnormal state data is marked in the evaluation report of S7 step, triggering the early warning mechanism, the triggering of the early warning mechanism is based on the fly ash quality deterioration index The calculation formula is: ； wherein, is a fly ash quality deterioration index, and are real-time values of unburned carbon content and fineness, respectively, and are target values thereof, and are weight coefficients; S82, use Bayesian network for root cause diagnosis to infer the most likely fault source in reverse and generate abnormal cause diagnosis report; S83, generate revised control instruction draft based on the abnormal cause diagnosis report and matching preset expert rule knowledge base; S84, review the potential impact of the revised control instruction draft by the control strategy safety simulation module, and output the final revised control instruction after confirming no conflict; S85, record the whole process of early warning trigger, cause diagnosis, instruction generation and execution to system event log.

10. A method for fly ash quality control of a thermal power unit according to claim 1, characterized in that: Step S9 includes the following steps: S91, issue the final revised control instruction generated in step S8 to the actuator module of the distributed control system of thermal power generating unit; S92, the actuator module schedules and executes the revised control task according to the preset safety and process priority; S93, through fast response closed-loop adjustment, until the fly ash quality deterioration index is restored to within the threshold value, and real-time running data of the intervention process is generated; S94, verify whether the real-time running data of the intervention process meets the expectation to evaluate the effectiveness of the revised control instruction, and output the revised intervention effect report; S95, correlate and analyze the revised intervention effect report and the abnormal cause diagnosis report to optimize the expert rule knowledge base and revised instruction generation strategy in step S8.

Images