Plasma pulverized coal burner ignition operation control method and device based on big data

Abstract

The invention discloses a plasma pulverized coal burner ignition operation control method and device based on big data, and relates to the technical field of ignition operation control. The method comprises the following steps: establishing a staged behavior skeleton of ignition operation of the plasma pulverized coal burner; historical ignition operation data are collected, and a behavior alignment sample set is obtained through staged behavior skeleton alignment; successfully-ignited samples are clustered, and an ignition success behavior track cluster is generated; based on the ignition success behavior track cluster, mapping rules of different coal quality categories and ignition behavior tracks are established, and stage target windows of various coal qualities are generated; selecting a corresponding reference track from an ignition success behavior track cluster according to the current coal quality category, and generating an ignition operation control instruction set in combination with a stage target window; executing the control instruction set and collecting real-time data; the instruction set is dynamically corrected according to the deviation between the real-time data and the reference track, so that the real-time ignition behavior track converges towards the reference track until ignition is completed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ignition operation control, in particular to a plasma coal burner ignition operation control method and device based on big data. BACKGROUND

[0002] The plasma coal burner ignition technology is a key link for modern coal-fired power plants to realize oil-free start-up of the unit, deep peak shaving and energy saving and consumption reduction. Its core goal is to quickly and reliably ignite the coal powder gas flow under complex and variable coal quality and initial state of the boiler, and to realize self-stable combustion, which puts forward very high requirements for the adaptability, accuracy and reliability of the control strategy.

[0003] The traditional plasma coal burner ignition operation control method generally adopts a fixed time sequence control method based on a preset program. Before ignition, a set of fixed time sequence control table is preset for each continuous step of the ignition process according to the test and experience of typical working conditions, and a threshold value is preset for the key controlled quantity in each step. The whole control process will strictly follow the fixed time sequence control table and gradually advance, and the main criterion for stage switching is that the actual measured value of the key controlled quantity reaches the threshold value preset for the stage.

[0004] However, due to the fluctuations of key characteristics such as volatile matter, ash content and calorific value of the actual coal fed into the furnace, and the fact that the initial wall temperature and air distribution state of the boiler are not constant, the fixed time sequence control table and threshold value cannot perceive and respond to these real-time changes, so there is inevitably a mismatch between the control instructions and the actual dynamic needs of the ignition operation, which ultimately leads to the risk of ignition delay and extinguishment during the process, seriously restricting the success rate of ignition. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a plasma coal burner ignition operation control method and device based on big data to solve the problems in the background art.

[0006] To achieve the above purpose, the present application is realized by the following technical scheme: a plasma coal burner ignition operation control method and device based on big data, comprising the following steps: Step S1: establishing a phased behavior skeleton of the plasma coal burner ignition operation; Step S2: collecting historical ignition operation data, aligning the historical ignition operation data based on the phased behavior skeleton, constructing a behavior alignment sample set, performing cluster analysis on the ignition success samples in the behavior alignment sample set, and generating ignition success behavior trajectory clusters; Step S3: based on the ignition success behavior trajectory clusters, establishing a mapping rule of different coal quality categories and ignition behavior trajectories, and generating a phased target window for different coal quality categories; Step S4: Select the corresponding reference trajectory from the ignition success behavior trajectory cluster according to the current coal quality category, and generate an ignition operation control instruction set based on the reference trajectory and the stage target window of the current coal quality category; Step S5: Execute the ignition operation control command set and collect real-time ignition operation data. Based on the deviation between the real-time ignition operation data and the reference trajectory, dynamically correct the ignition operation control command set through the Model Predictive Control (MPC) algorithm, so that the real-time ignition behavior trajectory during the ignition operation converges to the reference trajectory until the ignition operation ends.

[0007] Preferably, the staged behavioral framework for the ignition and operation of a plasma pulverized coal burner is established, including the following steps: A theoretical framework is established to construct a staged behavioral skeleton of plasma pulverized coal burner ignition operation. Based on the fundamental physical criterion of whether a qualitative change occurs in combustion behavior, a complete ignition operation process is abstracted into a pre-activation stage, an initial ignition stage, a flame expansion stage, and a stable combustion transition stage. A behavioral fingerprint that can be objectively quantified by an online monitoring system is defined for each stage.

[0008] Preferably, historical ignition operation data is collected, and the historical ignition operation data is aligned based on the staged behavior skeleton to construct a behavior alignment sample set, including the following steps: Collect historical ignition run datasets ,in Indicates the first Each historical operation record, each record The included time-series sensor data directly corresponds to the monitoring signals specified in the staged behavior skeleton for quantifying the behavioral fingerprints of each stage; simultaneously, each record It must also include a label indicating whether it was ultimately successful or not; Based on the aforementioned staged behavior skeleton, the dataset Each record in Behavioral alignment processing is performed, and behavioral fingerprint indicators for each stage are calculated sequentially over time. When an indicator meets the specified stage end criterion, the end point of that time period is marked as the stage boundary point, and the four stage boundaries are identified sequentially. The temporal feature point sequence within each stage is extracted and spliced ​​into a standardized behavioral trajectory in the order of pre-activation → initial ignition → flame expansion → stable combustion transition. All trajectories constitute a behavior-aligned sample set. A is the sample size.

[0009] Preferably, cluster analysis is performed on the successful ignition samples in the behavior alignment sample set to generate clusters of successful ignition behavior trajectories, including the following steps: From the behavioral homogeneous sample set All samples labeled "successful" were selected to form a subset of successful ignition samples. The similarity is calculated using Dynamic Time Warped Distance (DTW distance). The aim is to find the optimal alignment path between two sequences, and its calculation formula is defined as follows: ; in, Is it the behavioral trajectory? Index in; It is the first Article and Section A successful behavioral trajectory; It is all regular paths that satisfy boundary conditions, continuity, and monotonicity constraints. A set; It is a regular path A pair of matching points on, where It is a trajectory The index of a certain feature vector in the data. It is a trajectory The index of a certain feature vector; It is a trajectory In the Feature vectors at each position; It is the Euclidean distance function that calculates the distance between two feature vectors; Based on this, the K-means algorithm is used to... Cluster analysis is performed to minimize the sum of squared errors within clusters (SSE), and the formula is as follows: ; in, It is the number of clusters; It is the first output of the K-means algorithm. A cluster of trajectories, It is the first A cluster of trajectories The central trajectory; Is the trajectory in the cluster Index within; After the cluster analysis is completed, generate Cluster of successful ignition behavior trajectories .

[0010] Preferably, based on the cluster of successful ignition behavior trajectories, a mapping rule is established between different coal types and ignition behavior trajectories, including the following steps: The coal quality is classified and characterized in an engineering manner, and a coal quality feature vector is defined. Each component This represents a coal quality characteristic that can be obtained through engineering. The number of coal quality characteristics is set, and classification threshold rules are preset for each characteristic. A collection of categories m is the category index; From the successful sample subset of step S2 In the process, all coal quality category labels were filtered out. sample subset Subsequently, statistical sample subset In the process, for each sample belonging to its trajectory cluster, the sample within each trajectory cluster under that coal quality category is calculated. ( The proportion of the quantity in ) Next, select the quantity percentage. Largest trajectory cluster As a coal quality category The mapping target is determined by the number of trajectory clusters. If multiple trajectory clusters have the same proportion and are all at their maximum values, the trajectory cluster with the smallest sum of squared DTW distances among the samples within that cluster is selected, thus forming a one-to-one mapping rule. .

[0011] Preferably, generating stage target windows for different coal quality categories includes the following steps: For each coal quality category, a suitable stage target window is generated. The stage target window defines the flexible value range that various behavioral fingerprint features should reach at different combustion stages. and its mapped successful trajectory cluster Considering all samples in this cluster at stage and behavioral fingerprint features Therefore, The target window for a given period can be defined as an interval centered at the mean and expanded by a factor of several standard deviations.

[0012] in, Coal quality category In the stage For behavioral fingerprint features The lower bound of the target window; Coal quality category In the stage For behavioral fingerprint features The upper bound of the target window; Coal quality category Corresponding success trajectory cluster In the process, all samples are in the stage Behavioral fingerprint characteristics The arithmetic mean; Coal quality category Corresponding success trajectory cluster In the process, all samples are in the stage Behavioral fingerprint characteristics Standard deviation; It is the window width coefficient, which is a positive real number.

[0013] Preferably, selecting a corresponding reference trajectory from the cluster of successful ignition behavior trajectories based on the current coal quality category includes the following steps: Based on the current coal quality, a unique reference trajectory is determined, and... The current coal quality category is indicated by rules based on industry experience thresholds; before ignition and operation begin, the mapping rules established in step S3 are applied. From the set of ignition success behavior trajectories In the selection process, the trajectory cluster that best matches the current coal quality is chosen. From this cluster Select its most representative center trajectory As a reference trajectory for this ignition .

[0014] Preferably, based on the reference trajectory and the current coal quality category's stage target window, an ignition operation control instruction set is generated, including the following steps: Reference trajectory Compared with the current coal quality Phase Target Window Based on the phased behavioral framework, a structured ignition operation control instruction set is generated, dividing the process into four stages: pre-activation, initial ignition, flame propagation, and stable combustion transition. A specific control rhythm is generated for each stage, forming the ignition operation control instruction set for this ignition. : ; in, The pre-activation phase controls the rhythm; It is about controlling the pace during the initial ignition stage; It controls the rhythm during the flame expansion phase; It is the controlled rhythm during the stable combustion transition phase.

[0015] Preferably, the ignition operation control command set is dynamically corrected using the Model Predictive Control (MPC) algorithm, including the following steps: To dynamically eliminate bias, a Model Predictive Control (MPC) algorithm is introduced. This algorithm performs a certain function at each sampling time. Based on the current state A simplified predictive model describing the dynamic characteristics of the combustion process for future finite time domains. Predicting system behavior within a step: ; in, Current sampling time; This is the prediction time domain, representing the number of steps to predict forward; It is the control time domain, representing the number of steps of the control action to be optimized (usually...). ); At any moment Predicted future Behavioral fingerprint vector of the step; It is the ideal behavioral fingerprint vector of the reference trajectory at the corresponding future moment; It is the control increment vector to be optimized, that is, the control command for the current ignition operation. The adjustment amount; These are the weight matrices for tracking error and control increment, respectively.

[0016] The plasma pulverized coal burner ignition and operation control device based on big data includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the above method.

[0017] This invention provides a method for controlling the ignition and operation of a plasma pulverized coal burner based on big data, involving machine learning and deep learning technologies, which has the following beneficial effects: (1) This AI-based power distribution box regulation method automatically summarizes and classifies historical successful ignition samples by applying the K-means clustering algorithm. It can efficiently and objectively extract several representative clusters of successful ignition behavior trajectories from a large number of high-dimensional behavior alignment sample sets. This process successfully transforms discrete successful operations that depend on individual experience into a structured and quantifiable knowledge base of typical successful patterns. It realizes automated and intelligent knowledge mining of massive historical ignition operation data and makes implicit operational experience explicit into clear and reusable successful path templates.

[0018] (2) This AI-based power distribution box regulation method introduces the Model Predictive Control (MPC) algorithm and uses it as the core strategy for trajectory following control, constructing a feedforward-feedback composite control system based on prediction and optimization. The algorithm uses a reference trajectory as the tracking target and, within each control cycle, calculates the control command sequence that makes the future predicted trajectory most closely approximate the ideal trajectory by solving the optimal control problem in the finite time domain. This achieves precise coordinated control of multivariable, strongly coupled, and nonlinear complex ignition processes, overcoming the limitations of traditional sequential control in handling multivariable coordination and dynamic optimization.

[0019] (3) The power regulation method of the distribution box based on artificial intelligence introduces a rolling optimization mechanism. This mechanism restarts the optimization process based on the latest real-time measurement data at each sampling moment, rolls the optimization domain forward, and only implements the optimal control quantity at the current moment, ensuring that the control strategy always matches the current actual combustion state. This enables the entire ignition control process to have dynamic self-correction and continuous optimization capabilities, so that it can maintain high-precision trajectory tracking and optimal economy and safety under changing working conditions, realizing the fundamental transformation from static program control to dynamic intelligent optimization. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Fig. 1 This is a flowchart of the steps of the plasma pulverized coal burner ignition and operation control method based on big data proposed in this invention; Fig. 2 This is a step hierarchy diagram of the successful ignition behavior trajectory cluster obtained in the big data-based plasma pulverized coal burner ignition operation control method proposed in this invention; Fig. 3 This is a step-by-step diagram of the ignition operation process in the big data-based plasma pulverized coal burner ignition operation control method proposed in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figs. 1-3 This invention provides a technical solution: a method for controlling the ignition and operation of a plasma pulverized coal burner based on big data.

[0024] Step S1: Establish the staged behavioral framework for the ignition and operation of the plasma pulverized coal burner.

[0025] The core of this step is to establish a theoretical framework for the staged behavior skeleton of plasma pulverized coal burner ignition operation, in order to standardize the description of the ignition operation process of plasma pulverized coal burner. This staged behavior skeleton abandons the traditional approach of dividing stages by fixed time length, and instead, based on the fundamental physical criterion of whether a qualitative change occurs in combustion behavior, it abstracts a complete ignition operation process into a pre-activation stage, an initial ignition stage, a flame expansion stage, and a stable combustion transition stage, and defines a behavioral fingerprint for each stage that can be objectively quantified by the online monitoring system, ensuring that the entire control process has a clear discrimination benchmark.

[0026] The phased behavior skeleton includes the following four phases and their boundary criteria: For the pre-activation stage, this stage begins with plasma arc ignition and aims to establish a stable recirculation zone and preheat the pulverized coal. The end of this stage (i.e. the boundary to enter the next stage) is determined by the following criteria: a stable thermal recirculation structure has been formed inside the burner, the pulverized coal particles have begun to be effectively heated and release volatiles, and the coupling between plasma energy and primary air-powder airflow has reached a dynamic equilibrium.

[0027] For the initial ignition stage, this stage is the qualitative change point where pulverized coal changes from preheating to continuous combustion. The end criterion is that one or more stable high-temperature fire cores that can be self-sustaining and do not experience intermittent interruptions are formed in the monitoring area near the burner outlet.

[0028] The flame propagation stage marks the spread of the flame from a local fire core to the entire combustion zone. Its termination is determined by the stable adhesion of the flame root at the burner outlet, and the occurrence of a continuous, monotonous increase in key parameters characterizing combustion intensity without significant fluctuations.

[0029] The stable combustion transition phase aims to gradually remove external auxiliary energy to achieve self-sustaining combustion. Its termination (i.e., the completion of the entire ignition process) is determined by the flame maintaining stable combustion without signs of backfire or flameout during the gradual reduction and even shutdown of plasma power.

[0030] To ensure that the stage boundary criteria can be directly identified and reused, specific and quantifiable behavioral fingerprints must be defined for each stage. These behavioral fingerprints are characterized by the characteristics of online monitoring signals collected by specific sensors deployed at key locations. Behavioral fingerprint quantification during the pre-activation phase involves monitoring the spectral characteristics of pressure fluctuation signals using pressure sensors deployed at the burner outlet. When the spectral energy is concentrated in the 50-200Hz band and lasts for ≥15s, it indicates that the recirculation structure is stabilizing. Simultaneously, the temperature change rate (temperature rise gradient) is monitored using armored thermocouples deployed at the burner outlet. When the temperature rise gradient is maintained in the 5-15℃ / s range and lasts for ≥10s, it indicates that the thermal activation trend of pulverized coal is established. Additionally, the ratio of plasma power output to primary air-coal flow rate is monitored, and the coupling coefficient is calculated. When the fluctuation range of this coupling coefficient is ≤±8% and lasts for ≥10s, it indicates that the coupling between plasma energy and primary air-coal flow reaches dynamic equilibrium. Behavioral fingerprint quantification of the initial ignition stage is achieved by acquiring flame images, light intensity signals, and fire core temperature values ​​using an industrial camera and infrared flame detector positioned at the burner outlet. When bright spots with brightness fluctuations ≤ ±5% and area fluctuations ≤ ±10% are identified in the image, and the detector signal changes from pulsating to a sustained high level (≥8000 lux), and the fire core temperature is ≥1200℃ for a duration ≥10s, it indicates that a sustainable high-temperature fire core has formed. Behavioral fingerprint quantification during the flame expansion phase involves real-time analysis of the overall flame morphology using industrial cameras installed inside the furnace. The analysis is based on the observation that the time-series curves of key parameters such as flame area and brightness exhibit a stable, monotonically increasing trend, and that the flame area meets the 5-second average growth rate requirement. For stable flame expansion, the average annual growth rate of brightness over 5 seconds must be 2% / second, and this growth must continue for at least 20 seconds. This is further verified using radiant heat flux meters installed on the furnace sidewalls. When monitoring data shows a continuous increase in radiation intensity, meeting the requirement of an average annual growth rate of ≥2% / second over 5 seconds, and also lasting for at least 20 seconds, this provides strong evidence of increased combustion intensity. Finally, flame stability also requires examining the root adhesion. The position of the flame root is identified and tracked using a high-precision industrial camera. When the distance between the flame root and the burner outlet face fluctuates within ±20 mm, and this stable state lasts for at least 15 seconds, it indicates that the flame root has achieved stable adhesion at the burner outlet.

[0031] Behavioral fingerprint quantification during the stable combustion transition phase involves comprehensive monitoring of all the aforementioned signals (pressure spectrum, temperature gradient, flame image features, and radiation intensity). If the fluctuation amplitude of these key signals remains within the preset safety threshold range and lasts for ≥30s as the plasma power gradually decreases at a gradient of 5kW / min until it is shut off, this will be used as a criterion for the flame's self-sustaining capability.

[0032] It should be noted that the preset security threshold range is as follows: Pressure spectrum fluctuation threshold: the main frequency fluctuation amplitude of the pressure spectrum is ≤ ±10Hz; temperature fluctuation threshold: the temperature fluctuation amplitude monitored by the armored thermocouple is ≤ ±10℃; flame image feature fluctuation threshold: the flame brightness fluctuation amplitude is ≤ ±8%, and the area fluctuation amplitude is ≤ ±10%; radiation intensity fluctuation threshold: the radiation intensity fluctuation amplitude is ≤ ±5%; plasma power adjustment threshold: it is gradually reduced in a gradient of 5kW / min until it is turned off, and the above signal fluctuations still maintain the threshold and the duration is ≥30s.

[0033] This step abandons the traditional fixed-time division model and divides the ignition process into four stages based on the qualitative changes in combustion behavior: pre-activation, initial ignition, flame propagation, and stable combustion transition. A quantifiable behavioral fingerprint (based on sensor signals) and clear boundary criteria are defined for each stage, forming a staged behavioral framework that replaces the vague stage divisions relying on human experience. This provides a unified behavioral analysis benchmark for subsequent steps, ensuring a clear structural basis for historical data alignment and trajectory clustering. It avoids data confusion caused by inconsistent stage divisions and lays the foundation for extracting successful ignition patterns and establishing control logic.

[0034] Step S2: Collect historical ignition operation data, align the historical ignition operation data based on the staged behavior skeleton, construct a behavior alignment sample set, perform cluster analysis on the successful ignition samples in the behavior alignment sample set, and generate a cluster of successful ignition behavior trajectories.

[0035] This step aims to align and reconstruct historical data based on the behavioral fingerprints and corresponding monitoring signals clearly defined in the phased behavioral framework established in step S1, and then summarize typical ideal behavioral patterns from all successful ignition records.

[0036] First, collect historical ignition operation datasets. ,in Indicates the first This historical operation record This represents the number of historical records. Each record... The included time-series sensor data directly corresponds to the monitoring signals specified in the staged behavior skeleton for quantifying the behavioral fingerprints of each stage, with a sampling frequency ≥10Hz, a single record data missing rate ≤5%, and abnormal data directly removed; simultaneously, each record It must also include a label indicating whether it was ultimately successful or not, with the success or failure label determined according to the S1 stable combustion transition stage criteria.

[0037] Based on the aforementioned staged behavior skeleton, the dataset Each record in Behavioral alignment processing is performed. Specifically, a 10-second sliding time window and a 1-second step size are used to calculate the behavioral fingerprint indicators for each stage defined in S1 in chronological order. When an indicator meets the stage end criterion (including the corresponding duration threshold) specified in S1, the end point of that time period is marked as the stage boundary point, and the four stage boundaries are identified sequentially. The temporal feature point sequence within each stage is extracted and spliced ​​into a standardized behavioral trajectory in the order of pre-activation → initial ignition → flame expansion → stable combustion transition. This trajectory is formed by sequentially connecting the behavioral fingerprint feature vectors corresponding to four stages, i.e. ,in For the first Record in The temporal feature point subsequence within the stage, each feature point in the subsequence contains All quantitative indicators defined for this stage (such as pressure spectrum dominant frequency, temperature rise gradient, and energy coupling coefficient for pre-activation stage feature points) ensure that the indicator dimensions of all feature points are consistent. It is a pre-activated temporal feature point sequence; It is a sequence of temporal characteristic points of the initial ignition; It is a sequence of temporal feature points of flame propagation; It is a sequence of temporal characteristic points of steady-state combustion transition. All trajectories constitute a behavior-aligned sample set. A is the sample size.

[0038] Next, from the behavior alignment sample set All samples labeled "successful" were selected to form a subset of successful ignition samples. To discover inherent patterns of success, cluster analysis is needed. First, any two successful behavioral trajectories need to be measured. and The similarity between them is calculated using dynamic time warping distance, as the duration of each stage in different operations can be flexibly adjusted.

[0039] First, align the behavior sample set. All time-series feature points are subjected to min-max normalization for the same index. The normalization formula is as follows: ,in , These are the minimum and maximum values ​​of the indicator among all successful sample time-series feature points, respectively; after normalization, the value range of all indicators is unified as follows: Then, perform subsequent DTW distance calculations; DTW distance The aim is to find the optimal alignment path between two sequences to minimize the cumulative distance, and its calculation formula is defined as follows: ; in, Is it the behavioral trajectory? Index in; It is the first Article and Section A successful behavioral trajectory; It is all regular paths that satisfy boundary conditions, continuity, and monotonicity constraints. A set; It is a regular path A pair of matching points on, where It is a trajectory The index of a certain feature vector in the data. It is a trajectory The index of a certain feature vector; It is a trajectory In the The feature vector at each location (corresponding to the fingerprint of a certain stage); It is the Euclidean distance function that calculates the distance between two feature vectors; It should be noted that, The specific expression for Euclidean distance is: ; in, For a single time-series feature point, the index dimension is... Indicates the first Trajectory number The first feature point Item indicator value; Based on this, the K-means algorithm is used to... The goal of cluster analysis is to minimize the sum of squared errors within clusters (SSE), as shown in the following formula: ; in, It is the number of clusters (the number of trajectory clusters); It is the first output of the K-means algorithm. A cluster of trajectories, It is the first A cluster of trajectories The central trajectory; Is the trajectory in the cluster The formula clusters similar trajectories into the same cluster and finds the central trajectory of each cluster such that the sum of the squared DTW distances of all trajectories in the cluster to the central trajectory is minimized.

[0040] Number of clusters The determination is based on the elbow rule combined with the contour coefficient method, and the process is as follows: calculate Plot the sum of squared intra-cluster errors (SSE) from 2 to 10. Find the elbow point on the curve where the slope changes abruptly. Range; calculate each within this range Choose the contour coefficient with the largest value. As the final number of clusters; the initial center trajectory from Random selection Each time series trajectory is unique, avoiding initial value deviation.

[0041] Finally, after the cluster analysis is completed, the following is generated: Cluster of successful ignition behavior trajectories Each trajectory cluster The center trajectory This represents the ideal behavioral path for this type of successful model.

[0042] It should be noted that the center trajectory The center trajectory is calculated using the DTW centroid averaging method. This process typically employs an iterative optimization algorithm, and its core update formula is as follows: The first Feature vectors at each position Its updated value is: ; in, Trajectory within the cluster With the current center trajectory The optimal regularized path obtained through DTW calculation. It is the first of all normalized paths mapped to the central trajectory. A set of feature vectors at each position. It is an indicator function; in the sense of DTW distance metric, it calculates an average sequence that minimizes the sum of its DTW distances to all sequences within the cluster, i.e., the center trajectory; the termination condition for iteratively updating the center trajectory is the average DTW distance between the center trajectories of two adjacent iterations. The iteration will stop when the number of iterations reaches 50.

[0043] This step generates Cluster of successful ignition behavior trajectories It serves as the direct data foundation for S3's adaptive mapping of coal quality and is also the source for S4 to select the reference trajectory for the current ignition. It makes the implicit experience of operators explicit and standardized, providing a reusable and ideal template for subsequent ignition control.

[0044] Step S3: Based on the ignition success behavior trajectory cluster, establish mapping rules between different coal types and ignition behavior trajectories, and generate stage target windows for different coal types.

[0045] This step follows the ignition success behavior trajectory cluster generated in step S2, aiming to address the critical issue of control strategy failure caused by differences in coal quality characteristics. By establishing a mapping relationship between coal quality categories and success behavior trajectory clusters, and by pre-setting dynamic compliance standards for behavior fingerprints at each stage for different coal qualities, the control strategy gains intelligent adaptability to changes in coal quality.

[0046] First, coal quality is classified and characterized using engineering methods. A coal quality feature vector is defined. Each component This represents a coal quality characteristic that can be obtained through engineering. Quantities representing coal quality characteristics, such as: dry ash-free volatile matter. Received base ash Received low-grade heat of heat Softening temperature and coal powder fineness Based on these characteristics, and combined with industry experience and operational data, classification threshold rules are preset for each characteristic (e.g., according to...). Coal is classified into high, medium, and low volatile matter coals, and coal quality is further classified into... A collection of categories , where m is the category index.

[0047] The subset of successful ignition samples generated in step S2 Each sample carries a corresponding coal quality category label (this label comes from the coal quality testing data entering the furnace in historical operation records, and corresponds to the coal quality category classified in this step). (Complete match); and all samples have been clustered into the corresponding ignition success behavior trajectory cluster. In this process, a three-dimensional correlation is formed between the sample, coal quality category, and trajectory cluster. For each coal quality category ( To establish a mapping relationship, follow these steps: From the successful sample subset of step S2 In the process, all coal quality category labels were filtered out. sample subset Subsequently, statistical sample subset In the process, for each sample belonging to its trajectory cluster, the sample within each trajectory cluster under that coal quality category is calculated. ( The proportion of the quantity in ) The calculation formula is:

[0048] Next, select the percentage of quantity. Largest trajectory cluster As a coal quality category The mapping target is determined by the number of trajectory clusters. If multiple trajectory clusters have the same proportion and are all at their maximum values, the trajectory cluster with the smallest sum of squared DTW distances among the samples within the cluster (i.e., the trajectory cluster with the most stable behavior pattern within the cluster) is selected, forming a one-to-one mapping rule. That is, coal quality category The optimal fire behavior pattern for adaptation is trajectory clusters. The central trajectory.

[0049] Next, a suitable stage target window is generated for each coal quality category. The stage target window defines the flexible value range that various behavioral fingerprint features should achieve at different combustion stages. Its purpose is to fine-tune the achievement thresholds for each stage based on coal quality characteristics while reusing the overall shape of the successful trajectory cluster. For each coal quality category... and its mapped successful trajectory cluster Considering all samples in this cluster at stage fingerprint characteristics of a certain behavior Therefore, The target window for a given period can be defined as an interval centered at the mean and expanded by a factor of several standard deviations. ; in, Coal quality category In the stage For behavioral fingerprint features The lower bound of the target window; Coal quality category In the stage For behavioral fingerprint features The upper bound of the target window; Coal quality category Corresponding success trajectory cluster In the process, all samples are in the stage Behavioral fingerprint characteristics The arithmetic mean; Coal quality category Corresponding success trajectory cluster In the process, all samples are in the stage Behavioral fingerprint characteristics Standard deviation; This is the window width coefficient, a positive real number. The value of the window width coefficient α needs to be preset based on the coal quality characteristics and the differences in the ignition stage. For the pre-activation and initial ignition stages, α=2, covering 95% of the samples; for the flame expansion and stable combustion transition stages, α=1.5 to improve control accuracy; for low-volatile coals, the α value is reduced by 0.3 in each stage to match their difficult-to-burn characteristics; for high-volatile coals, the α value is increased by 0.2 in each stage to match their flammable characteristics. This formula represents the coal quality category. In the stage Targeting behavioral fingerprint features A dynamic range of permissible values ​​(target window) is calculated, so that the characteristic values ​​of the vast majority of successful samples for this coal quality fall within this range, thereby providing a quantifiable and coal quality-matched criterion for real-time control.

[0050] Finally, through the above process, this step outputs two core results: first, the mapping rules from coal quality categories to successful trajectory clusters; and second, the stage target windows for different coal quality categories. These two together constitute a coal quality prior knowledge base, enabling the control system to select the optimal reference trajectory based on the actual coal quality before ignition, and to use the dynamic target window as the criterion for behavior compliance during ignition, thereby achieving stable and adaptive ignition control for multiple coal sources.

[0051] Step S4: Select the corresponding reference trajectory from the ignition success behavior trajectory cluster according to the current coal quality category, and generate an ignition operation control instruction set based on the reference trajectory and the stage target window of the current coal quality category.

[0052] This step is the core of the methodology's shift from cognitive modeling to control execution, aiming to transform static knowledge into dynamic strategies before the ignition operation begins. It takes the coal quality stage target window output from step S3 and utilizes the ignition success behavior trajectory cluster established in step S2 to generate a personalized and executable action plan for the upcoming single ignition operation.

[0053] First, a unique reference trajectory is determined based on the current coal quality, then... The current coal quality category is indicated by rules based on industry experience thresholds; before ignition and operation begin, the mapping rules established in step S3 are applied. From the set of ignition success behavior trajectories In the selection process, the trajectory cluster that best matches the current coal quality is chosen. Subsequently, from this cluster Select its most representative center trajectory As a reference trajectory for this ignition This trajectory is an ideal evolutionary path formed by sequentially connecting the target behavior fingerprint feature vectors of four stages.

[0054] Secondly, the reference trajectory will be Compared with the current coal quality Phase Target Window Based on the phased behavioral framework, a structured ignition operation control instruction set is generated. The process is divided into four stages: pre-activation, initial ignition, flame propagation, and stable combustion transition. A specific control rhythm is generated for each stage, forming the ignition operation control instruction set for this ignition. : ; in, The pre-activation phase controls the rhythm; It is about controlling the pace during the initial ignition stage; It controls the rhythm during the flame expansion phase; It is the controlled rhythm during the stable combustion transition phase; It should be noted that the control rhythm at each stage The generation of [something] must simultaneously adhere to two principles: the goal-oriented principle and the constraint satisfaction principle. For the goal-oriented principle, the design of the control rhythm aims to drive the system state along a reference trajectory. The ideal path evolution is shown; for the constraint satisfaction principle, the parameters controlling the rhythm (such as the plasma power preset curve, primary air volume adjustment rhythm) must ensure that the real-time behavioral fingerprint can enter and stabilize in the trajectory cluster most suitable for the current coal quality. Phase Target Window The interior is the core design principle.

[0055] Controlling the rhythm of each stage The content design directly serves the physical goals and behavioral instructions defined in step S1 for different stages: Controlling the pace of the pre-activation phase Plasma initial power , by reference trajectory The energy coupling characteristic value of the pre-activation stage is determined, and the fluctuation range of the coupling coefficient must fall within the target window of the current coal quality stage. Within the corresponding coupling coefficient threshold range; power progression rate ,according to calculate, The duration of the pre-activation phase of the reference trajectory, The power threshold at the end of the pre-activation phase must be met, and during the power increment process, the dominant frequency of the burner outlet pressure spectrum and the temperature rise gradient must be ensured to fall synchronously within the specified range. Internal; Initial preset value of primary air volume The stability characteristic value of the recirculation structure during the pre-activation stage of the reference trajectory is determined to ensure the matching between the airflow velocity and the plasma arc, providing a basis for the formation of the hot recirculation zone; the airflow fine-tuning step size The preset value is 0.5% of the rated primary air volume. After each adjustment, it is necessary to verify that the spectral energy concentration characteristics of the pressure fluctuation signal meet the requirements. Requirements: Avoid sudden changes in airflow that could disrupt recirculation stability; Controlling the pace during the initial ignition stage Plasma power maintenance value , by reference trajectory The characteristic values ​​of the fire core temperature during the initial ignition stage are determined, and the temperature characteristics of the fire core region fall within the stage target window. Internal; power fine-tuning rate ,according to calculate, The duration of the initial ignition phase is used as a reference trajectory. This is the power fine-tuning threshold at the end of the initial ignition stage. After adjustment, the infrared detector signal must be kept stable in the middle range between the upper and lower boundaries of the target window; primary air volume preset value. The oxygen concentration matching characteristics during the initial ignition stage of the reference trajectory are used to ensure a suitable oxygen concentration environment for fire core formation; the air volume adjustment step size is determined by these characteristics. The preset value is 1% of the rated primary air volume. After each adjustment, it is necessary to verify that the fluctuation rate of the flame bright spot area falls within the specified range. Within the corresponding feature threshold; Controlling the rhythm during the flame expansion phase Plasma power retention value , by reference trajectory The characteristic value of combustion intensity during the flame propagation stage is determined, and the furnace radiation intensity growth rate falls within the stage target window. Internal; power adjustment threshold , default to When the flame area growth rate is lower than the corresponding characteristic value of the reference trajectory for three consecutive sampling periods, the trigger power is increased by 5%. After adjustment, it is necessary to ensure that the adhesion characteristics at the flame root are stable; primary air volume preset value. Determined by the flame morphology guidance characteristics during the flame expansion stage of the reference trajectory, it is used to propel the flame from the local fire core to the entire combustion zone; airflow adjustment gradient ,according to calculate, The duration of the flame propagation phase is used as a reference trajectory. The target airflow value is the airflow rate at the end of the flame propagation phase, and the airflow adjustment process must ensure that the monotonically increasing characteristic of the flame brightness time-series curve conforms to the target value. Require.

[0056] Controlling the pace during the stable combustion transition phase Initial value of plasma power This is equal to the power value at the end of the flame propagation phase, and it must satisfy the condition that all behavioral fingerprint features fall within the phase target window at the initial moment. Internal; power descent gradient ,according to calculate, To ensure the duration of the stable combustion transition phase of the reference trajectory, it is necessary to guarantee that the pressure spectrum, temperature fluctuations, and other characteristics after each adjustment do not exceed the specified limits during the power reduction process. Threshold range; Primary air volume adaptation value Determined by the combustion self-sustaining characteristics during the stable combustion transition phase of the reference trajectory, it is used to compensate for the energy gap after the plasma power decreases; airflow adjustment rate The default setting is to adjust every 5 seconds, with each adjustment being [amount missing]. The adjustment should be 0.5%, and the fluctuation range of radiation intensity needs to be verified to remain within a certain range after the adjustment. Within the corresponding safe range, the plasma power will decrease to 0 and the flame will burn stably.

[0057] Next, define the ignition operation control instruction set. The dynamic execution logic—intra-phase rhythm control—ensure that the progression of the control process strictly adheres to the phase boundary criteria defined in the framework. The phase switching permission function `ProceedIf` is defined. as follows: ; in, It represents the current stage of the ignition process, and its value belongs to the set. ; It is the current stage of operation immediately following ignition. The subsequent stages; At a certain point in time The current behavior fingerprint feature vector is calculated from real-time monitoring signals; It is for coal quality The stage target window at the current stage of ignition operation is a multi-dimensional interval that defines the allowable value range of each characteristic component. To maintain the time threshold, the value is consistent with the duration of the stage boundary criterion corresponding to step S1.

[0058] This step matches the optimal reference trajectory from the cluster of successful ignition behavior trajectories generated in S2, based on the coal quality type of the current ignition scenario. Combining the stage target window for this coal quality type, control rhythms are formulated for each of the four ignition stages, specifying preset values ​​for key parameters such as plasma power and primary air volume. This forms a set of directly executable ignition operation control instructions, while also providing a reference benchmark for subsequent deviation calculations, ensuring that the correction process does not deviate from the direction of the successful ignition trajectory.

[0059] Step S5: Execute the ignition operation control command set and collect real-time ignition operation data. Based on the deviation between the real-time ignition operation data and the reference trajectory, dynamically correct the ignition operation control command set through the Model Predictive Control (MPC) algorithm, so that the real-time ignition behavior trajectory during the ignition operation converges to the reference trajectory until the ignition operation ends.

[0060] This step is the final closed-loop execution and adaptive control stage of the entire method. Its core task is to implement the ignition control command set generated in step S4 during real-time ignition operation. The initial execution framework is established, and a Model Predictive Control (MPC) algorithm is introduced. A rolling optimization mechanism is used to correct control commands in real time, ensuring that the actual combustion process closely follows the reference trajectory. Precise tracking enables adaptive optimization throughout the transition from pre-activation to stable combustion.

[0061] First, after ignition and startup, the instruction set is executed sequentially. Controlling the rhythm of the current stage Simultaneously, real-time ignition operation data is continuously collected via a sensor network. Based on the quantization rules defined in step S1, real-time behavioral fingerprint features are calculated from this real-time data. .Will Ideal fingerprint features corresponding to the stage in the reference trajectory Compare and calculate the deviation. .

[0062] Next, to dynamically eliminate bias, the Model Predictive Control (MPC) algorithm is introduced. This algorithm operates at each sampling time... Based on the current state A simplified predictive model describing the dynamic characteristics of the combustion process for future finite time domains. The algorithm predicts the system behavior within a step. The core of the algorithm is to solve a rolling optimization optimal control problem, whose objective function is... The aim is to minimize the deviation between the predicted trajectory and the reference trajectory, while considering the smoothness of the control input: ; in, Current sampling time; This is the prediction time domain, representing the number of steps to predict forward; It is the control time domain, representing the number of steps of the control action to be optimized (usually...). ); At any moment Predicted future Behavioral fingerprint vector of the step; It is the ideal behavioral fingerprint vector of the reference trajectory at the corresponding future moment; It is the control increment vector to be optimized, that is, the control command for the current ignition operation. The adjustment amount; These are the weight matrices for tracking error and control increment, respectively. The tracking error weight matrix... This is a diagonal matrix, where the diagonal elements are the tracking weights of the fingerprint features for each row, representing the initial ignition stage. The diagonal elements are set to 1.0, and the other elements are set to 0.5; control the incremental weight matrix. It is a diagonal matrix, with diagonal elements representing smoothing weights that control the increment during the stable combustion transition phase. The diagonal element is set to 1.0, and the other stages are set to 0.3; prediction time domain With control time domain Adjustments are made based on the dynamic characteristics of the ignition phase, including the pre-activation and initial ignition phases. , Flame expansion and stable combustion transition stage , Time-domain units and sampling period Consistency is ensured to guarantee the main dynamic processes in the prediction coverage phase; this objective function defines the optimization criteria for MPC. Part One: Penalizing Prediction Behavior Fingerprint Deviation from the reference value at the corresponding time point ensures tracking accuracy; the second part is the penalty control increment. To prevent drastic changes in (such as plasma power adjustment and damper opening) and ensure the smoothness of control actions, so as to avoid impact on the burner.

[0063] Then, the rolling optimization mechanism of the MPC algorithm comes into play. At time... Solving the above optimization problem yields the optimal control increment sequence starting from the current time step. However, only the first control increment in the sequence is implemented. That is, controlling the pace of the current stage of execution. Fine-tuning is performed to obtain updated real-time control commands. The next sampling time is then reached. At that time, the state is recalculated based on the latest collected real-time data, the entire prediction time domain is rolled forward one step, and the optimization solution is repeated. This cycle of "execution-measurement-re-optimization" is called rolling optimization.

[0064] It should be noted that the rolling optimization mechanism is the core of the adaptive convergence optimization within the implementation phase of the Model Predictive Control (MPC) algorithm. Its core lies in ensuring the dynamic accuracy of the control strategy through a closed-loop cycle of "execution-measurement-re-optimization." This mechanism first implements feedback correction, meaning each optimization is based on the latest combustion state, thus promptly overcoming unknown disturbances such as coal quality fluctuations and measurement noise. Second, it possesses look-ahead optimization capabilities, considering not only the current tracking deviation but also optimizing the control strategy for future multiple steps, avoiding short-sighted behavior and enabling the entire trajectory to converge smoothly and stably to the reference trajectory. Finally, the mechanism incorporates constraint handling functionality, explicitly adding engineering physical constraints (such as upper and lower limits) to control variables like plasma power and primary air volume, as well as key state variables like combustion temperature, during the optimization process. This ensures that while pursuing optimal performance, the safety and feasibility of the entire ignition operation process are strictly guaranteed.

[0065] Finally, the above execution and rolling optimization process continues. Current stage behavior fingerprint. Enter the pre-set stage target window for this coal quality. Then, based on the phase switching permission function ProceedIf defined in step S4... The control process automatically advances to the next stage in the skeleton, and the MPC controller also switches to tracking the reference trajectory of the next stage. This cycle continues, and the termination criterion for the stable combustion transition phase is: after the plasma power drops to 0kW, Each characteristic component falls within 30s If there are no signs of backfire or flameout, ignition is considered successful. MPC optimization is then stopped, and the control flow is terminated, marking the end of the ignition process. Through the closed-loop implementation of step S5, the entire method achieves a complete transformation from static knowledge to dynamic intelligent execution.

[0066] This technical solution provides a big data-based ignition and operation control method for plasma pulverized coal burners. By establishing a data-driven intelligent control framework, it effectively solves the industry problem that traditional fixed-sequence control strategies are difficult to adapt to changes in coal quality and operating conditions.

[0067] First, by establishing a phased behavioral framework based on qualitative changes in combustion behavior, the continuous ignition process is deconstructed into four standard stages: pre-activation, initial ignition, flame propagation, and stable combustion transition. A quantifiable behavioral fingerprint is defined for each stage, providing a unified coordinate system for subsequent data analysis and control. Second, innovatively, dynamic time warping distance and K-means clustering algorithms are used to align and mine patterns from historical successful ignition data, generating clusters of successful ignition behavior trajectories representing different success patterns, transforming discrete experience into a structured knowledge base. Based on this, the scheme establishes mapping rules between coal quality categories and behavioral trajectories, and generates coal-adaptive stage target windows, enabling the control strategy to adapt to fuel characteristics. In practical implementation, Model Predictive Control (MPC) is used as the core means to achieve trajectory tracking. A rolling optimization mechanism adjusts control commands in real time to ensure that the actual combustion process accurately tracks the reference trajectory selected based on the current coal quality.

[0068] This method achieves a complete closed loop from historical data mining and coal quality adaptation to real-time optimization control through five organically linked steps. Its technical value lies in upgrading ignition control from the traditional "parameter preset" mode to a "process optimization" mode. Through data-driven approaches, it significantly improves ignition success rate, stability, and economy, providing effective technical support for the flexible and safe operation of coal-fired power units.

[0069] This invention also protects a plasma pulverized coal burner ignition and operation control device based on big data, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0070] This technical solution achieves intelligent power regulation through a multi-level technical architecture. First, based on wavelet transform and the LightGBM machine learning framework, raw electrical parameter waveform data is transformed into a sequence of electrical behavioral events with clear semantics, completing the leap from signal perception to behavioral understanding. Then, a scene-aware model is constructed using knowledge graph technology, combined with the ARIMA time series prediction algorithm, to achieve dynamic reasoning of load priorities and proactive identification of potential conflict risks. Finally, model predictive control theory is employed, with a yielding mechanism at its core, to generate a set of flexible regulation instructions at the equipment level, and reinforcement learning technology is used to achieve continuous self-optimization of control parameters. This method innovatively establishes a fully intelligent closed loop encompassing perception, cognition, prediction, and execution, effectively solving the problems of lag and rigid regulation in traditional distribution boxes. Its technical value lies in introducing semantic understanding, scene adaptation, and predictive control into the field of power distribution. Through dynamic priority management and flexible regulation strategies, it significantly improves user experience while ensuring safe operation, providing core technical support for building an adaptive smart grid. The entire approach demonstrates the advanced nature and engineering feasibility of artificial intelligence technology in power system applications, possessing significant theoretical value and broad application prospects.

[0071] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for ignition and operation control of a plasma pulverized coal burner based on big data, characterized in that, Includes the following steps: Step S1: Establish the staged behavioral framework for the ignition and operation of the plasma pulverized coal burner; Step S2: Collect historical ignition operation data, align the historical ignition operation data based on the staged behavior skeleton, construct a behavior alignment sample set, perform cluster analysis on the successful ignition samples in the behavior alignment sample set, and generate a cluster of successful ignition behavior trajectories. Step S3: Based on the ignition success behavior trajectory cluster, establish mapping rules between different coal types and ignition behavior trajectories, and generate stage target windows for different coal types; Step S4: Select the corresponding reference trajectory from the ignition success behavior trajectory cluster according to the current coal quality category, and generate an ignition operation control instruction set based on the reference trajectory and the stage target window of the current coal quality category; Step S5: Execute the ignition operation control command set and collect real-time ignition operation data. Based on the deviation between the real-time ignition operation data and the reference trajectory, dynamically correct the ignition operation control command set through the Model Predictive Control (MPC) algorithm, so that the real-time ignition behavior trajectory during the ignition operation converges to the reference trajectory until the ignition operation ends.

2. The method for ignition and operation control of a plasma pulverized coal burner based on big data as described in claim 1, characterized in that, Establish a phased behavioral framework for the ignition and operation of a plasma pulverized coal burner, including the following steps: A theoretical framework is established to construct a staged behavioral skeleton of plasma pulverized coal burner ignition operation. Based on the fundamental physical criterion of whether a qualitative change occurs in combustion behavior, a complete ignition operation process is abstracted into a pre-activation stage, an initial ignition stage, a flame expansion stage, and a stable combustion transition stage. A behavioral fingerprint that can be objectively quantified by an online monitoring system is defined for each stage.

3. The method for ignition and operation control of a plasma pulverized coal burner based on big data as described in claim 2, characterized in that, Collect historical ignition operation data, align the historical ignition operation data based on the staged behavior skeleton, and construct a behavior alignment sample set, including the following steps: Collect historical ignition run datasets ,in Indicates the first Each historical operation record, each record The included time-series sensor data directly corresponds to the monitoring signals specified in the staged behavior skeleton for quantifying the behavioral fingerprints of each stage; simultaneously, each record It must also include a label indicating whether it was ultimately successful or not; Based on the aforementioned staged behavior skeleton, the dataset Each record in Behavioral alignment processing is performed, and behavioral fingerprint indicators for each stage are calculated sequentially over time. When an indicator meets the specified stage end criterion, the end point of that time period is marked as the stage boundary point, and the four stage boundaries are identified sequentially. The temporal feature point sequence within each stage is extracted and spliced ​​into a standardized behavioral trajectory in the order of pre-activation → initial ignition → flame expansion → stable combustion transition. All trajectories constitute a behavior alignment sample set. A is the sample size.

4. The ignition and operation control method for a plasma pulverized coal burner based on big data according to claim 3, characterized in that, Cluster analysis is performed on the successful ignition samples in the behavior alignment sample set to generate clusters of successful ignition behavior trajectories, including the following steps: From the behavioral homogeneous sample set All samples labeled "successful" were selected to form a subset of successful ignition samples. The similarity is calculated using dynamic time-warped distance. distance The aim is to find the optimal alignment path between two sequences, and its calculation formula is defined as follows: ； in, Is it the behavioral trajectory? Index in; It is the first Article and Section A successful behavioral trajectory; It is all regular paths that satisfy boundary conditions, continuity, and monotonicity constraints. A set; It is a regular path A pair of matching points on, where It is a trajectory The index of a certain feature vector in the data. It is a trajectory The index of a certain feature vector; It is a trajectory In the Feature vectors at each position; It is the Euclidean distance function that calculates the distance between two feature vectors; Based on this, adopt Algorithm pair Perform cluster analysis to minimize the sum of squared errors within clusters. Its formula is: ； in, It is the number of clusters; yes The first output of the algorithm A cluster of trajectories, It is the first A cluster of trajectories The central trajectory; Is the trajectory in the cluster In-line index; After the cluster analysis is completed, generate Cluster of successful ignition behavior trajectories .

5. The ignition and operation control method for a plasma pulverized coal burner based on big data according to claim 4, characterized in that, Based on the aforementioned successful ignition behavior trajectory cluster, a mapping rule is established between different coal types and ignition behavior trajectories, including the following steps: The coal quality is classified and characterized in an engineering manner, and a coal quality feature vector is defined. Each component This represents a coal quality characteristic that can be obtained through engineering. To determine the number of coal quality characteristics, a classification threshold rule is preset for each characteristic, and the coal quality is divided into categories. A collection of categories m is the category index; From the successful sample subset of step S2 In the process, all coal quality category labels were filtered out. sample subset Subsequently, statistical sample subset In the process, for each sample belonging to its trajectory cluster, the sample within each trajectory cluster under that coal quality category is calculated. ( The proportion of the quantity in ) ; Next, select the percentage of quantity. Largest trajectory cluster As a coal quality category The mapping target is determined; if multiple trajectory clusters have the same proportion and are all at their maximum values, the trajectory cluster with the smallest sum of squared DTW distances among the samples within that cluster is selected; thus forming a one-to-one mapping rule. .

6. The method for ignition and operation control of a plasma pulverized coal burner based on big data according to claim 5, characterized in that, Generating stage target windows for different coal quality categories includes the following steps: For each coal quality category, a suitable stage target window is generated. The stage target window defines the flexible value range that various behavioral fingerprint features should reach at different combustion stages. and its mapped successful trajectory cluster Considering all samples in this cluster at stage and behavioral fingerprint features Therefore, The target window for a given period can be defined as an interval centered at the mean and expanded by a factor of several standard deviations. ； in, Coal quality category In the stage For behavioral fingerprint features The lower bound of the target window; Coal quality category In the stage For behavioral fingerprint features The upper bound of the target window; Coal quality category Corresponding success trajectory cluster In the process, all samples are in the stage Behavioral fingerprint characteristics The arithmetic mean; Coal quality category Corresponding success trajectory cluster In the process, all samples are in the stage Behavioral fingerprint characteristics Standard deviation; It is the window width coefficient, which is a positive real number.

7. The method for ignition and operation control of a plasma pulverized coal burner based on big data as described in claim 6, characterized in that, Selecting a corresponding reference trajectory from the cluster of successful ignition behavior trajectories based on the current coal quality category includes the following steps: Based on the current coal quality, a unique reference trajectory is determined, and... The current coal quality category is indicated by rules based on industry experience thresholds; before ignition and operation begin, the mapping rules established in step S3 are applied. From the set of ignition success behavior trajectories In the selection process, the trajectory cluster that best matches the current coal quality is chosen. From this cluster Select its most representative center trajectory As a reference trajectory for this ignition .

8. The method for ignition and operation control of a plasma pulverized coal burner based on big data according to claim 7, characterized in that, Based on the reference trajectory and the current coal quality category's stage target window, an ignition operation control instruction set is generated, including the following steps: Reference trajectory Compared with the current coal quality Phase Target Window Based on the phased behavioral framework, a structured ignition operation control instruction set is generated, dividing the process into four stages: pre-activation, initial ignition, flame propagation, and stable combustion transition. A specific control rhythm is generated for each stage, forming the ignition operation control instruction set for this ignition. : ； in, The pre-activation phase controls the rhythm; It is about controlling the pace during the initial ignition stage; It controls the rhythm during the flame expansion phase; It is the controlled rhythm during the stable combustion transition phase.

9. The ignition and operation control method for a plasma pulverized coal burner based on big data according to claim 8, characterized in that, The ignition control command set is dynamically corrected using the Model Predictive Control (MPC) algorithm, including the following steps: To dynamically eliminate bias, a Model Predictive Control (MPC) algorithm is introduced. This algorithm operates at each sampling time. Based on the current state A simplified predictive model describing the dynamic characteristics of the combustion process for future finite time domains. Predicting system behavior within a step: ； in, Current sampling time; This is the prediction time domain, representing the number of steps to predict forward; It is the control time domain, representing the number of steps of the control action to be optimized; At any moment Predicted future The behavioral fingerprint vector of the step; It is the ideal behavioral fingerprint vector of the reference trajectory at the corresponding future moment; It is the control increment vector to be optimized, that is, the control command for the current ignition operation. The adjustment amount; These are the weight matrices for tracking error and control increment, respectively.

10. A plasma pulverized coal burner ignition and operation control device based on big data, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-9.

Images