Method and system for matching raw material characteristics to biomass gasification coupled with coal-fired power generation

Abstract

The present application relates to the field of coal-fired power generation, and discloses a raw material characteristic adaptation biomass gasification coupled coal-fired power generation proportioning method and system, which collects biomass and coal raw material online and laboratory analysis data, maps and discretizes the characteristic coordinate system to construct a raw material characteristic phase space grid, fuses gasification and combustion process monitoring data to construct a working condition response Markov chain, generates a proportioning feedforward control function set through multi-step forward simulation and multi-objective optimization, and embeds an automatic control system to realize dynamic adjustment of fuel proportioning and combustion conditions; the present application solves the problems of loss of high-dimensional characteristic topological correlation of raw materials, difficulty of static models to cope with time-varying and sudden changes of raw materials, and lack of foresight of control decisions in traditional methods, realizes precise optimization and rapid response of fuel proportioning under data driving, and guarantees stable, efficient and environmentally friendly combustion.

Description

Technical Field

[0001] This invention relates to the field of coal-fired power generation technology, and in particular to a method and system for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation. Background Technology

[0002] Biomass gasification coupled with coal-fired power generation technology is an energy utilization method that combines biomass energy with coal-fired power generation. It aims to balance renewable energy utilization with the stability of the power generation system, achieving clean and efficient energy conversion. Existing fuel blending methods primarily determine the blending ratio of biomass and coal based on basic industrial analysis data of the raw materials or long-term operational experience. This is then combined with conventional control loops in the gasifier and boiler. By monitoring operating parameters such as furnace temperature, flue gas oxygen content, and exhaust gas temperature, the fuel feed rate is adjusted to maintain basic stability in the gasification and combustion processes. This type of blending technology has a relatively fixed operating procedure, relying on preset blending rules or a single calorific value index, and is suitable for operating scenarios with concentrated raw material sources and minimal fluctuations in raw material properties.

[0003] However, existing fuel blending methods have significant limitations. Industrial analysis indicators of biomass feedstocks, such as volatile matter, moisture, and ash content, are easily affected by regional differences, seasonal variations, and storage conditions, resulting in dynamic fluctuations. These indicators exhibit a strong nonlinear coupling relationship with the composition of combustible gases generated during gasification, tar production, and combustion reaction characteristics. Furthermore, during co-combustion with pulverized coal, they are further modulated by the fixed carbon content and particulate characteristics of the pulverized coal, forming a complex multivariate interaction. Existing blending strategies fail to effectively identify this complex coupling relationship and lack targeted online adaptation mechanisms. They rely solely on lagging operating parameters such as furnace temperature and oxygen content for feedback adjustment, failing to dynamically reconstruct the optimal blending ratio based on changes in feedstock characteristics before fuel is introduced into the furnace. This leads to unstable calorific value of gasification products, uneven combustion reactions within the boiler, incomplete fuel combustion, increased unburned carbon content, and furnace temperature fluctuations, affecting the uniformity of heat exchange on the heating surfaces. Long-term operation exacerbates localized equipment wear, and in severe cases, fluctuations exceeding the control range may trigger system protection shutdowns. A power plant experienced a change in the characteristics of its biomass raw materials due to changes in the storage environment. The fuel ratio was not adjusted in time, resulting in fluctuations in the composition of the combustible gas generated by gasification, a decrease in boiler combustion stability, and signs of high-temperature corrosion in local areas of the furnace. The plant was forced to reduce the unit load and shut down for inspection, resulting in power generation losses and increased maintenance costs. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, the present invention provides a method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation, comprising:

[0005] Online and laboratory analysis data of biomass and coal feedstocks were collected, a unified feature coordinate system was constructed and discretized, and a phase space grid for characterizing the characteristics of feedstocks with batch and time variations was generated.

[0006] By integrating online monitoring data of gasification products and boiler combustion, a Markov chain of operating condition response is constructed along the temporal evolution path of the feedstock characteristic phase space grid;

[0007] Based on the Markov chain of operating condition response, forward simulation of state transition is performed under different mixing assumptions to generate a set of feedforward control functions that provide fuel ratio and auxiliary control quantities for each state.

[0008] The set of feedforward control functions for biomass ratio is embedded in the fuel feeding and boiler automatic control system. The corresponding function is called according to the real-time identified status to perform dynamic adjustment of the biomass and coal ratio and combustion conditions.

[0009] Furthermore, the steps for generating the phase space mesh for raw material properties include:

[0010] Online detection devices are installed at the measuring points of the conveyor belt, feeding hopper and pulverizing system to obtain industrial analysis indicators of biomass and coal, and record the corresponding timestamps and feeding point location information.

[0011] The industrial analysis indicators in the raw material characteristic data source are mapped to a multi-dimensional feature coordinate system, with each dimension corresponding to the key raw material characteristics. Then, intervals are divided on each dimension according to standards or based on the statistical distribution of historical data, forming a raw material characteristic phase space grid on the multi-dimensional feature coordinate system.

[0012] The statistical information of each phase space grid cell of the raw material characteristics is maintained in the phase space grid to comprehensively characterize the distribution of biomass and coal raw material characteristics in phase space and their migration trajectory over time.

[0013] Furthermore, the steps for obtaining industrial analytical indicators for biomass and coal include:

[0014] Online moisture meters and near-infrared spectrometers were installed at the measuring points on the conveyor belts; weighing belts and rapid ash analyzers were installed at the measuring points on the feeding hoppers; and particle size analyzers and industrial cameras were installed at the measuring points on the powder making system.

[0015] Data on raw material industrial analysis indicators are collected in real time through various online detection devices;

[0016] The output data of all online detection devices are bound to the timestamps generated by a unified clock source, and the material loading point location number is marked to form a raw material online detection data package; at the same time, standard industrial analysis is performed on the biomass and coal raw materials before mixing in the laboratory on a daily or batch-by-batch basis to form a laboratory calibration data package; the raw material online detection data package and the laboratory calibration data package together constitute the raw material characteristic data source.

[0017] Furthermore, the steps for constructing the condition-response Markov chain include:

[0018] Data on the gasification and combustion processes are collected synchronously at the gasifier outlet and key locations in the furnace, and aligned with the raw material collection timestamp to form a process monitoring data stream.

[0019] Based on the raw material characteristic phase space grid cell into which the raw material falls at each moment, the gasification and combustion response at that moment are statistically included in the corresponding cell, thus upgrading the raw material characteristic phase space grid into a raw material-response joint distribution carrier.

[0020] The movement trajectory of raw material properties on the raw material property phase space grid is tracked in chronological order. Each visited raw material property phase space grid cell is regarded as a state. The jump between raw material property phase space grid cells falling into adjacent time moments is recorded as state transition. The transition frequency and condition response change between each state pair are statistically obtained to construct the working condition response Markov chain.

[0021] Furthermore, the specific steps for constructing the working condition response Markov chain include:

[0022] Traverse all elements in the phase space grid of raw material properties, filter out elements with sample counters greater than 0, define each element that meets the conditions as a working condition response Markov chain state node, and create a state attribute record for each state node.

[0023] Based on the timestamp order, the cells that fall into two adjacent time points are paired, the state transitions are recorded, the frequency of all directed transitions is counted, and a state transition frequency matrix is ​​constructed.

[0024] The state transition frequency matrix is ​​normalized row by row to obtain the state transition probability matrix. The condition response change vector is calculated. The set of state nodes, the state transition probability matrix, the set of condition response change vectors, and the set of state attribute records are encapsulated to form the overall object of the condition response Markov chain.

[0025] Furthermore, the steps for generating the set of proportioning feedforward control functions include:

[0026] The biomass blending ratio and coal powder fineness parameters are regarded as control variables, and a set of candidate control combinations are set for each state node in the working condition response Markov chain.

[0027] For each candidate control combination, the transition probability of the Markov chain of the operating condition response is used to perform multi-step forward simulation on the chain to predict the state sequence visited within a finite number of steps and the corresponding fluctuations in gas calorific value, tar enrichment trend and boiler load change.

[0028] The forward simulation results are comprehensively and quantitatively evaluated with safety constraints, efficiency targets and environmental protection indicators. A set of optimal control strategies is selected for each state node, and a set of proportional feedforward control functions is constructed.

[0029] Furthermore, the steps for performing multi-step forward simulation include:

[0030] Define an upper limit for the number of forward simulation steps. For each candidate control combination at each state node, start an independent forward simulation process and initialize the current state, gas calorific value, tar content and boiler load.

[0031] The simulation process follows the random walk rule of Markov chains. In each step, the next state node is sampled and determined. The relevant parameters are updated using the conditional response change vector. The transition and update process is repeated until the upper limit of the number of steps is reached or the state of absorption is entered. The simulation trajectory is recorded.

[0032] For each candidate control combination, Monte Carlo simulations are repeated multiple times. The simulation results are aggregated, relevant statistics are calculated, and stored in the forward simulation results database.

[0033] Furthermore, the steps for dynamically adjusting the biomass-to-coal ratio and combustion conditions include:

[0034] During the operation phase, online characteristic data of biomass and coal are continuously collected in real time to locate the current raw material characteristics and identify the corresponding state number in the working condition Markov chain of the working condition response.

[0035] The control system selects a matching function from the set of proportioning feedforward control functions based on the current state number, and calculates the biomass blending ratio, feed rate setpoint, and air distribution ratio to be used at the current moment.

[0036] The control quantity is used as a feedforward setting input to the boiler combustion control loop, and works in conjunction with the furnace temperature and oxygen feedback signals to achieve coordinated regulation of feedforward and feedback, and the ratio strategy is switched synchronously when the state changes.

[0037] Furthermore, the steps to achieve coordinated regulation of feedforward and feedback include:

[0038] The fuel main control loop adopts a cascade control structure to match the biomass and pulverized coal feed rates; the air supply control loop adopts cross-limit control logic to ensure that the total air supply volume matches the fuel quantity; the induced draft control loop maintains a slight negative pressure in the furnace and performs proportional feedforward compensation based on the air supply volume.

[0039] The feedback correction is calculated based on the furnace temperature and oxygen content feedback signals and then added to the feedforward setpoint.

[0040] When a sudden change in the raw material batch leads to a state switch, the control system quickly switches the control strategy, adopts a setpoint ramp transition, and reconstructs the model parameters of the feedback compensation module online to achieve feedforward and feedback synergy to suppress disturbances.

[0041] A method and system for adapting feedstock characteristics to biomass gasification coupled with coal-fired power generation, used to implement the aforementioned method for adapting feedstock characteristics to biomass gasification coupled with coal-fired power generation, the system comprising:

[0042] Raw material property data acquisition and phase space grid construction module: used to collect online and laboratory analysis data of biomass and coal raw materials, construct a unified feature coordinate system and discretize it to generate raw material property phase space grid;

[0043] Operating condition response Markov chain construction module: used to integrate gasification products and boiler combustion online monitoring data to construct an operating condition response Markov chain along the temporal evolution path of the feedstock characteristic phase space grid;

[0044] Proportion feedforward control function set generation module: used for forward simulation and multi-objective optimization based on the Markov chain of operating condition response, to generate proportion feedforward control function set;

[0045] Control execution and coordinated adjustment module: It is used to embed the set of feedforward control functions for biomass ratio into the automatic control system, call the corresponding function according to the real-time status, and combine feedback signals to realize the dynamic adjustment of biomass and coal ratio and combustion conditions.

[0046] Compared to existing technologies, the advantages of this invention are as follows: This invention accurately solves the core problems in traditional biomass gasification coupled with coal-fired power generation, such as the loss of high-dimensional characteristic topological correlations, the difficulty of capturing time-varying nonlinear couplings in static models, the lack of foresight in control decisions, and the lag in responding to raw material mutations. It achieves multiple ingenious effects through data-driven hierarchical modeling and collaborative control. First, the raw material characteristic phase space grid maps discrete detection data into a six-dimensional hypercube structure, preserving the topological correlations of characteristics such as volatile matter and particle size. Furthermore, by adapting to streaming data through sparse storage and incremental updates, it accurately captures batch mutations and time migration trajectories of raw materials, avoiding the shortcomings of traditional two-dimensional storage that masks extreme operating conditions. Second, the operating condition response Markov chain dynamically binds the raw material state with the gasification combustion response. Without explicitly fitting high-dimensional functions, it implicitly learns the coupling relationship through state transition probabilities and response change vectors, making it more robust to data noise and abnormal samples, and allowing for early prediction of the impact of raw material changes on the furnace. Third, the feedforward control function set, based on Monte Carlo multi-step simulation, selects the optimal strategy that balances long-term safety, economy, and environmental protection, effectively avoiding local optima that are effective in the short term but cause tar accumulation and load oscillation in the long term, thus achieving global ratio optimization. Fourth, the feedforward-feedback collaborative architecture and millisecond-level state recognition eliminate the need for manual parameter tuning when raw materials change abruptly, allowing for feed and air distribution adjustments to be completed within 30 seconds. This shifts the timing of control intervention forward by tens of seconds to minutes, significantly suppressing furnace temperature fluctuations and load drift. Simultaneously, through the collaborative optimization of pulverized coal fineness and blending ratio, it solves the combustion mismatch problem under traditional fixed fineness conditions. Furthermore, the model supports online thermal updates, continuously adapting to changes in raw material characteristics, ensuring the long-term stable, efficient, and environmentally friendly operation of the unit. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0048] Figure 1 A flowchart illustrating the method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation provided by the present invention.

[0049] Figure 2 This is a schematic diagram of the Markov chain state transition in the working condition response of this invention embodiment;

[0050] Figure 3 This is a schematic diagram of the forward simulation and state trajectory in an embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram of feedforward-feedback coordinated control in an embodiment of the present invention;

[0052] Figure 5 The functional module diagram of the biomass gasification coupled coal-fired power generation ratio system adapted to the raw material characteristics provided by the present invention is shown. Detailed Implementation

[0053] 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.

[0054] Example 1:

[0055] Please see Figure 1 As shown, this embodiment provides a method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation, including:

[0056] Step S1: Collect online and laboratory analysis data of biomass and coal feedstocks, construct a unified feature coordinate system and discretize it to generate a phase space grid for characterizing batch and time-varying feedstock properties.

[0057] Further, step S1 includes the following sub-steps:

[0058] Step S11: Online monitoring devices are installed at the measuring points of the conveyor belt, feeding hopper, and pulverizing system to acquire industrial analysis indicators of biomass and coal, and to record the corresponding timestamps and feeding point location information. Further, step S11 includes the following three sub-steps:

[0059] Step S111: Install an online moisture meter and a near-infrared spectrometer at the measuring point on the conveyor belt, install a weighing belt and a rapid ash analyzer at the measuring point on the feeding hopper, and install a particle size analyzer and an industrial camera at the measuring point on the powder making system.

[0060] Step S112: Real-time data acquisition of raw material industrial analysis indicators is achieved through various online detection devices. The online moisture meter uses microwave attenuation or infrared reflection methods to measure the surface and internal moisture content of the raw material in real time, outputting moisture mass percentage data. The near-infrared spectrometer scans the near-infrared absorption spectrum of the raw material and uses a pre-established partial least squares regression model or support vector machine regression model to invert volatile matter content, fixed carbon content, and ash content data online, outputting the dry ash-free mass percentage data of each component. The weighing belt measures the instantaneous mass flow rate of the raw material through a combination of weighing and speed sensors, outputting flow rate data in tons per hour. The rapid ash analyzer uses dual-energy gamma-ray transmission or laser-induced breakdown spectroscopy to rapidly determine the ash content of the raw material within a minute-scale timescale, outputting the received ash mass percentage data. The particle size analyzer uses laser diffraction or dynamic image analysis to measure the particle size distribution of coal powder and biomass powder, outputting characteristic particle size parameters, including but not limited to median diameter and average diameter. The industrial camera identifies the particle morphology parameters of the raw material through image acquisition and edge detection algorithms, outputting particle aspect ratio and roundness data.

[0061] Step S113: Bind the output data of all online detection devices to the timestamps generated by a unified clock source. The timestamps are accurate to the second, and the feeding point location number generated using the device topology coding rules is marked to ensure that each detection data can be traced back to a specific physical location. Package the volatile matter, moisture, ash, fixed carbon, particle size, particle morphology indicators, timestamps, and feeding point location numbers into a raw material online detection data package. At the same time, perform standard industrial analysis on the biomass and coal raw materials before mixing in the laboratory on a daily or batch-by-batch basis. This includes, but is not limited to, determination of volatile matter on a dry ash-free basis, determination of total moisture, determination of ash on an as-received basis, determination of fixed carbon on a dry basis, and determination of Hardy's grindability index. Align the laboratory analysis results with the online detection data package in terms of time and batch to form a laboratory verification data package. The raw material online detection data package and the laboratory verification data package together constitute the raw material characteristic data source.

[0062] For example, in a 660MW coal-fired power plant biomass co-generation project, the specific data acquisition process in step S11 is as follows: The online moisture meter installed at the No. 1 conveyor belt measuring point collected a raw material detection data at 09:23:47 on March 15, 2024. At that time, the moisture meter output a moisture mass percentage of 22.6%; the near-infrared spectrometer at the same measuring point simultaneously output a volatile matter content of 38.2% (dry ash-free basis), a fixed carbon content of 48.5% (dry ash-free basis), and an ash content of 7.8% (dry ash-free basis); the weighing belt installed at the No. 2 feeding hopper measuring point output an instantaneous mass flow rate of 45.3 tons per hour, and the ash rapid analyzer output an received ash content of 6.1%; the particle size analyzer installed at the pulverizing system measuring point output a median diameter D50 of 85 micrometers and an average diameter of 92 micrometers, and the industrial camera identified an average aspect ratio of 2.3 and a roundness of 0.68 for the particles. All the above data were bound to a unified clock source timestamp of 2024-03-15T09:23:47, and the feeding point location was marked as "BP-01-A3". On the same day, the laboratory conducted standard industrial analysis on this batch of straw biomass samples, and measured the dry ash-free volatile matter to be 37.9%, total moisture to be 23.1%, received ash to be 6.3%, dry fixed carbon to be 49.2%, and Hardgrove grindability index to be 42. A laboratory calibration data package was formed and time-aligned with the online detection data package.

[0063] Step S12: Map the industrial analysis indicators from the raw material characteristic data source to a multi-dimensional characteristic coordinate system, with each dimension corresponding to a key raw material characteristic. Then, divide the intervals on each dimension according to the DL / T 5145-2012 standard or based on the statistical distribution of historical data, forming a regular raw material characteristic phase space grid on this multi-dimensional characteristic coordinate system. Further, step S12 includes the following three sub-steps:

[0064] Step S121: Define a six-dimensional feature coordinate system. The first dimension is the volatile matter dimension, with the coordinate axis ranging from 0% to 100%. The second dimension is the moisture dimension, with the coordinate axis ranging from 0% to 50%. The third dimension is the ash dimension, with the coordinate axis ranging from 0% to 40%. The fourth dimension is the fixed carbon dimension, with the coordinate axis ranging from 0% to 90%. The fifth dimension is the particle size dimension, with the coordinate axis ranging from 0 micrometers to 5000 micrometers. The sixth dimension is the time dimension, with the coordinate axis covering at least one complete quarterly operating cycle in days.

[0065] Step S122: Divide the data into 20 discrete intervals at 5% intervals along the volatile matter dimension; 25 discrete intervals at 2% intervals along the moisture dimension; 14 discrete intervals at 3% intervals along the ash content dimension; 18 discrete intervals at 5% intervals along the fixed carbon dimension; 50 discrete intervals at 100-micrometer intervals along the particle size dimension; and at least 13 discrete intervals at 7-day intervals along the time dimension. The interval division thresholds for each dimension can be dynamically adjusted based on the statistical histogram of the actual raw material distribution. If the number of data points in an interval is less than a preset lower threshold, the interval is merged with adjacent intervals. If the number of data points in an interval exceeds a preset upper threshold, the interval is further subdivided. The preset lower threshold is determined by calculating the 10th percentile of the number of samples in all non-empty grid cells. The preset upper threshold is derived by inversely calculating the upper limit value based on the memory capacity of the online computing platform and the single update time constraint.

[0066] Step S123: Perform Cartesian product operation on the intervals of each dimension to generate a six-dimensional hypercube grid structure. Each cell of this grid structure is called a raw material characteristic phase space grid cell. Each raw material characteristic phase space grid cell is uniquely determined by six interval indices, corresponding to the volatile matter interval index, moisture interval index, ash content interval index, fixed carbon interval index, particle size interval index, and time interval index, respectively. The geometric meaning of the raw material characteristic phase space grid cell is that it defines a closed hyperrectangular region in the six-dimensional feature space. The physical meaning is that it represents a combination of raw material states with similar industrial analysis characteristics. The mathematical attributes include the cell center coordinates formed by the midpoints of the intervals of each dimension, the cell boundary coordinates formed by the upper and lower limits of the intervals of each dimension, and the cell volume calculated by the product of the lengths of the intervals of each dimension. The entire raw material characteristic phase space grid object is composed of all raw material characteristic phase space grid cells, forming a discretized data structure that completely covers the raw material characteristic parameter space.

[0067] Step S13: Maintain statistical information for each feedstock characteristic phase space grid cell in the feedstock characteristic phase space grid to comprehensively characterize the distribution and migration trajectory of biomass and coal feedstock characteristics in phase space over time. Further, step S13 includes the following three-level sub-steps:

[0068] Step S131: Create and maintain a cell statistical record for each raw material characteristic phase space grid cell. The cell statistical record includes the following fields: a sample counter field, which records the total number of raw material samples falling into the cell. The sample counter is an unsigned integer type, with an initial value of 0, and is incremented by 1 for each new sample falling into the cell; a time series index set field, which records the set of timestamps corresponding to all samples falling into the cell. The time series index set is implemented using an ordered linked list or a balanced binary tree data structure, supporting efficient insertion and deduplication operations; and a volatile matter statistics field, including the volatile matter mean and volatile matter variance. The maximum and minimum volatile matter values ​​are maintained using an incremental update algorithm. When a new sample is added, the mean and variance are updated using the Welford algorithm, and the maximum and minimum values ​​are compared and updated. The structure and update method of the moisture, ash, fixed carbon, and particle size statistics fields are the same as those of the volatile matter statistics field. The most recent update timestamp field records the last time this unit was updated. The batch identifier set field records the batch numbers of all raw materials added to this unit. The batch identifier set is implemented using a hash set data structure, which supports fast querying and deduplication.

[0069] Step S132: After the raw material online detection data package or laboratory verification data package is generated, the volatile matter, moisture, ash, fixed carbon, particle size data and timestamp are extracted. The coordinate position of the data point in the six-dimensional feature coordinate system is calculated, and the index of each dimension interval is determined to locate the corresponding raw material characteristic phase space grid cell. Then, the cell statistical record of the cell is atomically updated, that is, the sample counter is updated, the timestamp is inserted into the time series index set, the statistics are updated using new sample data, the most recently updated timestamp is refreshed, and the batch number is added to the batch identifier set. After the update operation is completed, the raw material characteristic phase space grid object reflects the distribution density, distribution range and time evolution characteristics of all historical raw material samples in the feature space up to the current moment.

[0070] In step S133, as new data continues to arrive, cell location and statistical updates are performed directly on the overall object of the raw material characteristic phase space grid, without breaking the data down into independent sample lists for storage. Thus, the overall data structure of the raw material characteristic phase space grid is always used as the sole carrier for describing the state of the raw material. The data structure of the overall object of the raw material characteristic phase space grid adopts a sparse storage mode, allocating storage space only for raw material characteristic phase space grid cells with sample counters greater than 0, and not allocating storage space for cells that have never been accessed, in order to reduce memory usage.

[0071] Specifically, step S1 addresses the technical challenge of online comparison and tracking of biomass and coal feedstock characteristics from different batches and sources within the same framework by constructing a raw material characteristic phase space grid. Traditional methods typically store detection data for various raw materials in independent time series or two-dimensional table formats, leading to the loss of topological correlation information between high-dimensional characteristics and making it difficult to intuitively reflect the migration path of raw material states in the feature space. The raw material characteristic phase space grid maps discrete and noisy industrial analysis results into a regular six-dimensional hypercube structure. Each raw material characteristic phase space grid cell simultaneously carries the feature value range and time series distribution information, preserving the topological relationships between adjacent characteristic states while avoiding the masking of extreme operating conditions by simple moving averages. This structure provides a spatially structured information carrier for subsequent construction of dynamic response models, enabling algorithms to identify the clustering regions, migration directions, and abrupt change locations of raw material characteristics in phase space—something that simple time series or two-dimensional tables cannot easily replace. Through sparse storage and incremental update algorithms, the raw material characteristic phase space grid can continuously evolve online, adapting to the characteristics of streaming data arrival in industrial sites and laying the data structure foundation for real-time state identification and prediction.

[0072] Step S2: Integrate online monitoring data of gasification products and boiler combustion, and construct a Markov chain of operating condition response along the temporal evolution path of the feedstock characteristic phase space grid. Further, step S2 includes the following sub-steps:

[0073] Step S21: Simultaneously collect gasification and combustion process data at key locations in the gasifier outlet and furnace, aligning them with the raw material collection timestamps to form a process monitoring data stream. Further, step S21 includes the following three sub-steps:

[0074] Step S211: Install a gas composition analyzer and a tar content detector in the gasifier outlet flue, install a calorific value meter in the gasifier outlet, install a temperature sensor array and an oxygen analyzer in the boiler furnace, and install a fly ash carbon content detector and an exhaust temperature sensor in the boiler tail flue, forming a monitoring network covering the entire process from gasification to combustion.

[0075] In step S212, the gas composition analyzer uses non-dispersive infrared absorption or tunable semiconductor laser absorption spectroscopy to measure the volume concentrations of carbon monoxide, hydrogen, methane, and carbon dioxide in the combustible gas in real time, and outputs the dry basis volume percentage data for each component; the tar content detector uses flame ionization detection or ultraviolet fluorescence to measure the tar mass concentration in the outlet gas, and outputs the tar content data in milligrams per cubic meter; the calorific value analyzer uses a water flow calorimeter or a continuous combustion calorific value analyzer to measure the lower heating value of the combustible gas. The system outputs calorific value data in megajoules per cubic meter; a temperature sensor array, using thermocouples or infrared thermometers, is arranged on the front wall, rear wall, side walls, and superheater area of ​​the furnace, outputting furnace temperature field distribution data in degrees Celsius; an oxygen analyzer uses a zirconia probe to measure the oxygen content in flue gas, outputting oxygen data in volume percentage; a fly ash carbon content detector uses microwave absorption or loss on ignition to measure the percentage of unburned carbon by mass, outputting fly ash carbon content data; and an exhaust gas temperature sensor outputs exhaust gas temperature data in degrees Celsius.

[0076] Step S213: All process monitoring data are accompanied by timestamps generated by the same unified clock source as the raw material online detection data packets, ensuring precise temporal correspondence with the raw material characteristic data source. For process monitoring data with transmission delays, a buffer delay alignment algorithm is used for timestamp compensation. The compensation value is pre-calibrated based on the sensor response time and signal transmission path length, ensuring that the raw material characteristic data and process response data are consistent in causal timing. The aligned gas composition, tar content, calorific value, furnace temperature, oxygen content, fly ash carbon content, and flue gas temperature data are packaged to form a process monitoring data stream. Each data record in this data stream includes a timestamp field, a measurement point location field, a physical quantity type field, a measured value field, and a data quality label field. The data quality label field is used to mark whether the data is valid, whether it has undergone interpolation compensation, and whether it exceeds the measurement range.

[0077] Step S22: Based on the feedstock characteristic phase space grid cell into which the feedstock falls at each moment, the gasification and combustion responses at that moment are statistically included in that cell, thus upgrading the feedstock characteristic phase space grid from a pure feedstock characteristic distribution carrier to a feedstock-response joint distribution carrier. Further, step S22 includes the following three sub-steps:

[0078] Step S221: For each data record in the process monitoring data stream, extract its timestamp field, find the time interval index corresponding to the timestamp in the overall object of the raw material characteristic phase space grid, and at the same time extract the volatile matter, moisture, ash, fixed carbon and particle size data in the raw material online detection data package that has been updated at that time, calculate its coordinate position in the six-dimensional characteristic coordinate system, and determine the volatile matter interval index, moisture interval index, ash interval index, fixed carbon interval index and particle size interval index, so as to completely locate the unique raw material characteristic phase space grid cell.

[0079] Step S222: Expand the cell statistical records of the phase space grid cell for the raw material characteristics by adding a response statistical field group. The response statistical field group includes gas calorific value statistics, tar content statistics, carbon monoxide concentration statistics, hydrogen concentration statistics, furnace temperature statistics, flue gas oxygen content statistics, and fly ash carbon content statistics. Each statistical field includes four subfields: mean, variance, maximum value, and minimum value. The same incremental update algorithm as the raw material characteristic statistics is used for maintenance.

[0080] Step S223 involves updating the calorific value, tar content, gas composition, temperature, oxygen content, and fly ash carbon content data at the corresponding time moment in the process monitoring data stream to the response statistics field group of the raw material characteristic phase space grid cell. As a result, the data structure of the raw material characteristic phase space grid cell evolves into a joint distribution cell that simultaneously contains raw material characteristic statistics and corresponding operating condition response statistics. The data semantics of the overall raw material characteristic phase space grid object also changes from simply describing the distribution of raw material state to describing the distribution of the mapping relationship from raw material state to operating condition response. For the same raw material characteristic phase space grid cell, if it is accessed multiple times at different times, its response statistics field group will accumulate the statistical characteristics of multiple operating condition responses, thereby reflecting the average performance and fluctuation range of the gasification and combustion process under the raw material state.

[0081] Step S23: Track the movement trajectory of raw material properties on the raw material property phase space grid in chronological order. Treat each visited raw material property phase space grid cell as a state. Record the transitions between raw material property phase space grid cells falling into adjacent time points as state transitions. Statistically obtain the transition frequency and condition response changes between each state pair to construct a working condition response Markov chain. Further, step S23 includes the following three-level sub-steps:

[0082] Step S231: Traverse all raw material characteristic phase space grid cells in the overall raw material characteristic phase space grid object, filter out cells with a sample counter greater than 0, and define each cell that meets the condition as a working condition response Markov chain state node. Each state node has a unique state number, which is generated by combining the six-dimensional interval index of the raw material characteristic phase space grid cell to ensure that the state number corresponds one-to-one with the raw material characteristic phase space grid cell. Create a state attribute record for each state node. The state attribute record includes a raw material characteristic core field, which is composed of the mean of each raw material characteristic statistic in the cell, representing the typical raw material characteristics of the state; a response characteristic core field, which is composed of the mean of each response statistic field in the cell, representing the typical working condition response characteristics of the state; a dwell frequency field, which records the total number of times the state has been accessed in the historical time series, and its value is equal to the sample counter; and a dwell time distribution field, which records the duration of each access to the state. The data is stored using a histogram data structure, and the histogram of dwell time distribution is divided into 12 equal-width intervals, with an interval width of 30 minutes. For example, all material property phase space grid cells in the overall material property phase space grid object are traversed, and 8532 cells with a sample counter greater than 0 are selected. Taking cell (14,11,3,3,8,10) as an example, it is defined as a working condition response Markov chain state node. The state number is generated by combining six-dimensional interval indexes. The encoding rule is to concatenate the indices of each dimension by a fixed number of bits. The state number is 14-11-03-03-08-10 (abbreviated as S1411030308010). Create a state attribute record for this state node: The raw material characteristic core field is composed of the mean of the statistical values ​​of each raw material characteristic within this unit. The raw material characteristic core field is (volatile matter 72.7%, moisture 22.3%, ash 9.2%, fixed carbon 18.4%, particle size 845 micrometers); the response characteristic core field is composed of the mean of the statistical values ​​of each response within this unit. The response characteristic core field is (calorific value 5.72 MJ / m³, tar 85 mg / m³, furnace temperature 1058 degrees Celsius, oxygen content 3.9%); the residence frequency field is 156 times; in the histogram of the residence time distribution field, the frequency of the interval [0 minutes, 30 minutes) is 45, the frequency of the interval [30 minutes, 60 minutes) is 38, the frequency of the interval [60 minutes, 90 minutes) is 28, and the remaining intervals are recorded sequentially.

[0083] Step S232: Based on the timestamp order, pair the raw material characteristic phase space grid cells that fall into two adjacent time points. If the cell at the previous time point is the cell corresponding to state node i, and the cell at the next time point is the cell corresponding to state node j, then record a transition from state node i to state node j in the working condition response Markov chain. Scan all historical data, count the frequency of all directed transitions, and construct a state transition frequency matrix. This matrix is ​​a square matrix, and its number of rows and columns are equal to the total number of state nodes. The matrix element F(i,j) represents the cumulative number of transitions from state node i to state node j. For example, pairing the raw material characteristic phase space grid cells that fall into two adjacent time points is based on the timestamp order. Let state node i (state number S1411030308010) be the state node corresponding to cell (14,11,3,3,8,10) at 14:32:45 on March 15, 2024; state node j (state number S1412030308010, moisture interval index changes from 11 to 12) be the state node corresponding to cell (14,11,3,3,8,10) at 14:32:55 on March 15, 2024; and state node j (state number S1412030308010, moisture interval index changes from 11 to 12) be the state node corresponding to cell (14,12,3,3,8,10) at 14:33:05 on March 15, 2024. Then, record one transition from state node i to state node j. After scanning the historical data for one quarter, construct a state transition frequency matrix. Assuming the total number of state nodes is 8532, the state transition frequency matrix is ​​an 8532×8532 square matrix. Taking the row containing state node i (S1411030308010) as an example, the matrix element F(i,i)=2345 represents the cumulative number of transitions from state i to state i, F(i,j)=89 represents the cumulative number of transitions from state i to state j (S1412030308010), and F(i,k)=56 represents the cumulative number of transitions from state i to state k (S1410030308010). The other elements in this row are recorded sequentially.

[0084] Step S233: The state transition frequency matrix is ​​normalized row-wise to obtain the state transition probability matrix. The normalization method is to divide each row element by its row sum so that the sum of each row element equals 1. Matrix element P(i,j) represents the probability of transitioning to state node j at the next moment, given the current state node i. For each pair of transitioning state nodes (i,j), the conditional response change vector is calculated. The conditional response change vector is calculated by subtracting the response characteristic tacticity tacticity of state node i from that of state node j. The result includes changes in calorific value, tar content, furnace temperature, and oxygen content, used to quantify the drift direction and amplitude of the operating condition response during the state transition process. The set of state nodes, the state transition probability matrix, the set of conditional response change vectors, and the set of state attribute records are encapsulated to form a complete Markov chain object for the operating condition response. For example, the state transition frequency matrix is ​​normalized row-wise. Taking the row containing state node i (S1411030308010) as an example, the sum of the elements in this row is 2345 + 89 + 56 + the cumulative number of state transitions of other elements = 2650. After normalization, we get the state transition probability matrix elements P(i,i) = 2345 / 2650 = 0.885, P(i,j) = 89 / 2650 = 0.034, P(i,k) = 56 / 2650 = 0.021. The calculation of other state transition probability matrix elements in this row is similar, and the sum of all elements in this row is equal to 1. For the state node pair (i,j) that undergoes a transition, its conditional response change vector is calculated. The tith of the response characteristics of state node j is (calorific value 5.65 MJ / m³, tar 92 mg / m³, furnace temperature 1045°C, oxygen content 4.1%), and the tith of the response characteristics of state node i is (calorific value 5.72 MJ / m³, tar 85 mg / m³, furnace temperature 1058°C, oxygen content 3.9%). The conditional response change vector is (calorific value change -0.07 MJ / m³, tar content change +7 mg / m³, furnace temperature change -13°C, oxygen content change +0.2%). The set of state nodes (8532), the state transition probability matrix (8532×8532), the set of conditional response change vectors, and the set of state attribute records are encapsulated to form a complete Markov chain object for the operating condition response.

[0085] See Figure 2 This is a schematic diagram of a Markov chain state transition for a working condition response provided in an embodiment of this application. Figure 2As shown, cells with a sample counter greater than 0 in the raw material characteristic phase space grid are mapped to visualized state nodes. Each node is color-coded to distinguish the risk level of the operating condition (blue represents the initial state of high volatile matter and high moisture, green represents the transition state of medium volatile matter, and dark green with a thick border indicates the target state of stable combustion). The nodes display state attribute records such as the core properties of raw materials (volatile matter, moisture, ash content), the core properties of response characteristics (calorific value of fuel gas), and the frequency of residence. The black arrows between states represent the natural transition probability P obtained based on the statistical normalization of historical data (e.g., P=0.35), while the green thick arrows specifically indicate the active guidance path after modulation by the enhancement coefficient under the action of feedforward control of the proportioning (e.g., P=0.42), intuitively reflecting the directional intervention capability of the control variables on the state evolution. Although tables are not directly drawn in the figure, each transition arrow is bound to a conditional response change vector {Δcalorific value, Δtar, Δfurnace temperature, Δoxygen content} in the engineering implementation, transforming the discrete phase space transition into a continuous response prediction. At the biomass co-firing site, when the online moisture meter detects a sudden increase in moisture content from 18% to 28%, the status recognition can switch from "status node 3" to "status node 1" within 10 seconds, and the control system will then query the status. Figure 2 The topology discovery showed that the enhanced transition probability to "stable combustion state 5" was as high as 0.55. Based on this, the feedforward function was triggered to adjust the blending ratio and coal powder fineness, which advanced the intervention time by 60-90 seconds compared with the traditional PID feedback. This prevented the furnace temperature from falling below the critical value of 850℃ from the source and significantly reduced the risk of slagging and tar enrichment caused by high alkali metal biomass.

[0086] Specifically, step S2 addresses the technical challenge of dynamically mapping changes in feedstock properties to gasification products and combustion conditions by constructing a condition-response Markov chain. Traditional methods typically employ static regression models or empirical curves to describe the relationship between feedstock properties and condition response, which struggles to capture time-varying characteristics and nonlinear coupling. The condition-response Markov chain utilizes the topological structure of the feedstock property phase space grid as the state space foundation, jointly encoding the temporal evolution of the feedstock state and its response characteristics into a stochastic process model. This transforms the high-dimensional non-convex mapping relationship into a computable state-transition structure. This structure is naturally suitable for predicting potential future states and their impact on the furnace, as the Markov chain's transition probability matrix directly provides the probability distribution of state evolution, and the condition response change vector gives the expected consequences of each evolution. This modeling approach avoids the difficulty of explicitly fitting high-dimensional nonlinear functions required by traditional methods, instead implicitly learning the input-output relationship through statistical state transition frequencies, exhibiting stronger robustness to noise and anomalous samples. Meanwhile, the working condition response Markov chain solidifies the joint distribution of raw material characteristics and working condition response within the state nodes and the dynamic evolution law within the transition edges, achieving a high degree of integration between the model structure and parameters, and providing an efficient computational framework for rapid querying and simulation of subsequent feedforward control.

[0087] Step S3: Based on the Markov chain of operating condition response, perform forward simulation of the state transition under different blending assumptions to generate a set of feedforward control functions that provide fuel ratios and auxiliary control quantities for each state. Further, step S3 includes the following sub-steps:

[0088] Step S31: Treat the biomass blending ratio and pulverized coal fineness parameters as control variables that drive the shift of the raw material state on the raw material characteristic phase space grid, and set a set of candidate control combinations for each state node in the working condition response Markov chain. Further, step S31 includes the following three-level sub-steps:

[0089] Step S311: Define a control variable vector. The control variable vector contains two main components. The first component is the biomass blending ratio, which is defined as the ratio of biomass feed mass flow rate to total fuel mass flow rate. The value ranges from 0% to 100%, and it is discretized in 5% increments to form 21 discrete values. The second component is the coal powder fineness, which is defined as the percentage of coal powder passing through a 200-mesh standard sieve by mass. The value ranges from 60% to 95%, and it is discretized in 5% increments to form 8 discrete values.

[0090] Step S312: For each working condition response Markov chain state node, based on the volatile matter, moisture, ash, fixed carbon, and particle size data in its raw material characteristic core field, calculate the expected raw material characteristic offset when using different control variable vectors in that state. The calculation logic for the expected raw material characteristic offset is as follows: the change in biomass blending ratio will linearly change the weighted average of volatile matter, moisture, ash, and fixed carbon in the mixed fuel, with the weight determined by the blending ratio; the change in coal powder fineness will affect the actual combustion characteristics of coal powder, and the fixed carbon reactivity will be modulated by the fineness correction coefficient, which adopts a power function form; traverse all discretized biomass blending ratio values ​​and coal powder fineness values ​​to form 168 candidate control combinations, i.e., 21 multiplied by 8 equals 168 combinations.

[0091] Step S313: For each candidate control combination, calculate the expected raw material characteristic offset caused by it under the current state node, add the expected raw material characteristic offset to the raw material characteristic centroid of the current state node to obtain the virtual target raw material characteristic coordinates; find the raw material characteristic phase space grid cell that is closest to the virtual target raw material characteristic coordinates in the overall raw material characteristic phase space grid object. The closest criterion is the smallest Euclidean distance. The state node of the working condition response Markov chain corresponding to this cell is the candidate target state that may be transferred to under the action of this control combination; record each candidate target state and its corresponding control variable vector and expected raw material characteristic offset in the candidate transfer record table. The candidate transfer record table maintains a list containing 168 records for each state node. Each record contains the candidate target state number, biomass blending ratio value, coal powder fineness value, virtual target raw material characteristic coordinates, and expected residence probability; the expected residence probability is initialized to a uniform distribution, that is, each candidate target state obtains an equal probability, and subsequent iterative optimization and adjustment are performed through forward simulation results.

[0092] Step S32: For each candidate control combination, using the transition probabilities of the Markov chain of the operating condition response, perform a multi-step forward simulation on the chain to predict the possible state sequence visited within a finite number of steps and the corresponding fluctuations in gas calorific value, tar enrichment trends, and boiler load changes. Further, step S32 includes the following three-level sub-steps:

[0093] Step S321: Define the upper limit of the forward simulation steps as 10 steps. This value is determined based on the inertial time constant of the boiler combustion system to ensure that the simulation duration covers the main dynamic processes from fuel injection to furnace stabilization. For each state node in the Markov chain of the operating condition response, extract its candidate transition record table. For each candidate control combination in the table, start an independent forward simulation process. During simulation initialization, set the current state node as the starting point, set the current gas calorific value as the average calorific value in the response characteristic center of this state node, set the current tar content as the average tar content in the response characteristic center of this state node, and set the current boiler load as the load percentage obtained by linear transformation of the average furnace temperature in the response characteristic center of this state node. For example, define the upper limit of the forward simulation steps as 10 steps. For the candidate control combination (biomass blending ratio 25%, coal powder fineness 80%) of state node i (S1411030308010), start an independent forward simulation process. Simulation initialization: Set the current state node to state i, the current gas calorific value to the average calorific value of the response characteristic center of this state node, which is 5.72 MJ / m³, the current tar content to 85 mg / m³, and the current boiler load to the load percentage obtained by linear transformation of the average furnace temperature of 1058 degrees Celsius, which is (1058-800) / (1200-800)×100%=64.5%.

[0094] Step S322: The simulation process follows the random walk rule of a Markov chain. In each step, the next state node is determined by sampling the state transition probability matrix of the current state node. The sampling method uses the roulette wheel algorithm, which generates a random number between 0 and 1 and selects the target state based on the position of the random number in the cumulative probability distribution. If there is a direct transition record between the current state node and the candidate target state, the transition probability of the candidate target state is multiplied by an enhancement coefficient. The enhancement coefficient is greater than 1 but less than the row sum and reciprocal of the transition probability matrix to reflect the guiding role of active control on natural transition. After completing one state transition, the current gas calorific value, tar content, and boiler load are updated using the conditional response change vector. The update method is to add the corresponding component in the conditional response change vector to the current value. The above transition and update process is repeated until the upper limit of the forward simulation steps is reached or the absorption state is entered. The absorption state is defined as a state in which the volatile matter is lower than the preset lower threshold. In this state, the system cannot maintain stable combustion. The state access sequence, gas calorific value sequence, tar content sequence, and boiler load sequence of each simulation are recorded to form a single simulation trajectory. The preset lower threshold is calculated using an adiabatic flame temperature calculation model to determine the minimum volatile volume fraction required to maintain a furnace temperature of no less than 850℃. For example, in the first simulation step, the row vector of the state transition probability matrix for the current state node i is [P(i,i)=0.885, P(i,j)=0.034, P(i,k)=0.021, P(i,m)=0.015, ...]. Since the candidate target state m (S1511020308010) has a direct transition record with the current state i, P(i,m) is multiplied by an enhancement coefficient of 1.8 (greater than 1 but less than the inverse of the row sum of the transition probability matrix, 1 / 0.015=66.7), resulting in the modulated probability P'(i,m)=0.015×1.8=0.027. After renormalization, the cumulative probability distribution is updated. A roulette wheel algorithm is used to generate a random number 0.923, and the target state j is selected based on the cumulative probability distribution. After the state transition is complete, the current values ​​are updated using the conditional response change vector (calorific value -0.07, tar +7, furnace temperature -13, oxygen +0.2): the gas calorific value is updated from 5.72 to 5.65 MJ / m³, the tar content is updated from 85 to 92 mg / m³, and the boiler load is updated from 64.5% to 61.2%. The above process is repeated until step 10 or the absorption state is entered. Using the adiabatic flame temperature calculation model, the minimum volatile matter volume fraction required to maintain a furnace temperature of no less than 850°C is determined to be 15%. A state with a volatile matter content below 15% is defined as the absorption state.Record the state access sequence [i,j,j,k,k,k,l,l,m,m], gas calorific value sequence [5.72,5.65,5.63,5.58,5.55,5.52,5.60,5.62,5.68,5.70], tar content sequence [85,92,95,98,102,105,100,96,90,88], and boiler load sequence [64.5,61.2,60.8,59.5,58.2,57.0,59.5,60.2,62.0,63.5] for this simulation.

[0095] Step S323: For each candidate control combination at each state node, repeat the Monte Carlo simulation 100 times to obtain the statistically significant average performance; aggregate the results of the 100 simulations to calculate the state access frequency distribution, i.e., the average number of times each state is accessed; calculate the gas calorific value fluctuation statistics, including the standard deviation and range of the calorific value sequence, to quantify calorific value stability; calculate the tar enrichment trend index, defined as the linear regression slope of the tar content sequence, with a positive slope indicating a continuous tar accumulation trend; calculate the boiler load change statistics, including the mean and variance of the load sequence, to assess load stability; store the above simulation results in a forward simulation result database. The result database maintains one record for each candidate control combination at each state node, and the record contains six fields: state access frequency distribution vector, calorific value fluctuation standard deviation, calorific value range, tar enrichment trend slope, load mean, and load variance. See also Figure 3 This is a schematic diagram of forward simulation and state trajectory provided in an embodiment of this application. For example, 100 Monte Carlo simulations are repeatedly performed on the candidate control combination (25% biomass blending ratio, 80% coal powder fineness) for state node i. The aggregated results of the 100 simulations specifically include: the state access frequency distribution is as follows: state i is accessed an average of 3.2 times, state j an average of 2.8 times, state k an average of 1.5 times, state m an average of 1.2 times, and other states a total of 1.3 times; the average standard deviation of the gas calorific value sequence is 0.12 MJ / m³, and the average range is 0.45 MJ / m³; the average linear regression slope of the tar content sequence is +1.8 mg / m³ / step, a positive value indicating a continuous tar accumulation trend; the mean of the boiler load sequence is 60.2%, and the variance is 12.5. The above results are stored in the forward simulation results database. The record of this candidate control combination includes: state access frequency distribution vector [3.2,2.8,1.5,1.2,...], calorific value fluctuation standard deviation 0.12, calorific value range 0.45, tar enrichment trend slope +1.8, load mean 60.2%, and load variance 12.5.

[0096] like Figure 3As shown, the raw material characteristic phase space grid constructed in step S1 serves as the base map (gray grid lines). The red semi-transparent area on the left is marked as the absorption state (unable to maintain stable combustion), and the red dashed line indicates the safety constraint boundary formed by the preset lower threshold. The current state node (t=0) is located in the middle of the feature space and is marked with a black square. The figure shows the state evolution paths driven by three typical control strategies: the blue solid line (preferred strategy path) extends to the upper right, marked with a blue square at each step, representing that when using a 20% biomass blending ratio and 80% coal powder fineness, the state migrates to the high volatile matter and low moisture region, the system calorific value increases and moves away from the absorption state; the orange dashed line (poor environmental performance path) extends downward, marked with an orange square, indicating that although a certain candidate strategy can maintain the load in the short term, it causes the tar enrichment trend slope to be positive, and long-term operation will block the tail flue; the red dotted line (safety violation path) moves to the left to touch the safety boundary, marked with a red square, predicting that under this strategy the system will fall into the absorption state in step 3, triggering combustion instability. This visualization method transforms the abstract statistics of Monte Carlo simulation in step S323 into interpretable topological trajectories, enabling control decisions to no longer rely on single-point evaluations but to make predictions based on multi-step evolutionary consequences. This effectively identifies "hidden risk strategies" that are difficult to detect using traditional empirical methods, and selects globally optimal control combinations without secondary risks for the matching feedforward control function set.

[0097] Step S33: The forward simulation results are comprehensively and quantitatively evaluated against safety constraints, efficiency targets, and environmental indicators. A set of control strategies with the best overall performance in the simulation time domain is selected for each state node, constructing a set of proportionate feedforward control functions. Further, step S33 includes the following three sub-steps:

[0098] Step S331: Define a multi-objective evaluation function. The evaluation function contains four sub-items, each of which is normalized to eliminate the influence of dimensions. The first sub-item is the safety sub-item, which is calculated as follows: when the standard deviation of calorific value fluctuation exceeds a preset safety threshold, the safety sub-item value linearly decreases to 0; when the calorific value range exceeds the preset safety threshold, the safety sub-item value is multiplied by a penalty coefficient less than 1. When the absorption state is accessed during the simulation, the safety sub-item is directly set to 0. The preset safety threshold is based on boiler tube fatigue life experimental data, extracting the critical fluctuation standard deviation corresponding to the inflection point of calorific value fluctuation amplitude and creep damage rate as the physical basis for setting the safety threshold. The second sub-item is the economic sub-item, which is calculated as follows: The calculation method is as follows: the square of the ratio of the average load to the rated load, minus the load variance multiplied by the stability penalty weight. The stability penalty weight is determined through regression analysis of historical operating data, with a value ranging from 0.1 to 0.8. The third sub-item is the environmental protection sub-item, which is calculated by applying an exponential decay mapping based on the slope of the tar enrichment trend. The closer the slope of the tar enrichment trend is to 0, the closer the environmental protection sub-item is to 1; the larger the slope of the tar enrichment trend, the faster the environmental protection sub-item approaches 0. The fourth sub-item is the state accessibility sub-item, which is calculated based on the sum of the probabilities of the top three high-frequency states in the state access frequency distribution vector. The larger the sum of the high-frequency states, the more the system tends to concentrate in the stable region under the control strategy, and the closer the accessibility sub-item is to 1. For example, for the candidate control combination of state node i (biomass blending ratio 25%, coal powder fineness 80%), the four sub-items of the multi-objective evaluation function are calculated. Based on boiler tube fatigue life test data, the critical standard deviation of the fluctuation amplitude corresponding to the inflection point of creep damage rate is extracted as 0.3 MJ / m³. A preset safety threshold of 0.3 MJ / m³ is set, and the safety threshold for the calorific value range is 1.0 MJ / m³. For the first sub-item, safety: the standard deviation of calorific value fluctuation (0.12) is less than the preset safety threshold of 0.3, so the safety sub-item value = 1 - (0.12 / 0.3) = 0.6; the calorific value range (0.45) is less than the preset safety threshold of 1.0, so no penalty coefficient is applied; the absorption state was not visited during the simulation, so the final value of the safety sub-item is 0.6. For the second sub-item, economy: the square of the ratio of the average load (60.2%) to the rated load (100%) is 0.362. The load variance (12.5) multiplied by the stability penalty weight (0.4, determined through regression analysis of historical operating data) equals 5.0, so the economy sub-item value = 0.362 - 0.05 = 0.312 (after normalization). The third sub-item, environmental friendliness: The slope of the tar enrichment trend is +1.8. Using an exponential decay mapping, the environmental friendliness sub-item = exp(-1.8 / 5) = 0.70. The fourth sub-item, state accessibility: The sum of the probabilities of the top three high-frequency states (i,j,k) in the state access frequency distribution vector is (3.2+2.8+1.5) / 10 = 0.75, therefore the accessibility sub-item is 0.75.

[0099] Step S332: The four sub-items are weighted and summed. The weight vector is determined using the Analytic Hierarchy Process (AHP). The judgment matrix of the AHP is constructed by the operation experts based on actual production preferences, ultimately obtaining the comprehensive score of each candidate control combination. For the 168 candidate control combinations of each state node, they are sorted in descending order of comprehensive score, and the top 3 candidate control combinations with the highest scores are selected as the preferred control strategy set for that state node. For example, the weight vector is determined using the AHP, and the judgment matrix is ​​constructed by the operation experts based on actual production preferences, ultimately obtaining the weight vector as [Safety 0.4, Economy 0.3, Environmental Protection 0.2, Accessibility 0.1]. The comprehensive score of the candidate control combination is calculated by weighting and summing the four sub-items = 0.4 × 0.6 + 0.3 × 0.312 + 0.2 × 0.70 + 0.1 × 0.75 = 0.24 + 0.094 + 0.14 + 0.075 = 0.549. The 168 candidate control combinations for state node i are sorted in descending order of comprehensive score. Assuming that the combination with a biomass blending ratio of 20% and a coal powder fineness of 80% has a comprehensive score of 0.682 and ranks first, the combination with a biomass blending ratio of 15% and a coal powder fineness of 85% has a comprehensive score of 0.665 and ranks second, and the combination with a biomass blending ratio of 25% and a coal powder fineness of 75% has a comprehensive score of 0.658 and ranks third, these top three candidate control combinations are selected as the preferred control strategy set for state node i.

[0100] Step S333: For each state node, construct a feedforward control function for the optimal control strategy set. This function takes the raw material characteristic centroid, response characteristic centroid, and forward simulation steps of the current state node as input. Internally, the function first selects the strategy with the highest matching degree from the optimal control strategy set based on the input parameters. The matching degree is calculated as the reciprocal of the Euclidean distance between the input centroid and the virtual target raw material characteristic coordinates corresponding to the optimal strategy. The strategy with the highest matching degree is selected. Then, the function outputs the biomass blending ratio, coal powder fineness, and expected value corresponding to the strategy. The feed rate correction coefficient, primary air distribution ratio, and secondary air distribution ratio are defined. The expected feed rate correction coefficient is calculated based on the deviation between the average load and the target load. The primary air distribution ratio is dynamically adjusted based on the volatile matter content, and the secondary air distribution ratio is dynamically adjusted based on the oxygen content deviation. The proportion feedforward control functions of all state nodes in the Markov chain of the operating condition response are aggregated to form a set of proportion feedforward control functions. This set of functions is implemented using a hash mapping data structure with state number as the key and proportion feedforward control function as the value, supporting function retrieval with constant time complexity. For example, for the optimal control strategy set of state node i, a proportion feedforward control function is constructed. When the function input parameters are: current state raw material characteristic centroid (72.7%, 22.3%, 9.2%, 18.4%, 845 μm), response characteristic centroid (5.72, 85, 1058, 3.9), forward simulation steps 10, target load setpoint 70%, and current measured oxygen content 4.0%, the matching degree of each optimal strategy is calculated based on the function's internal calculation. Strategy 1 (20%, 80%) has virtual target raw material characteristic coordinates of (74.1%, 22.6%, 8.9%, 18.5%, 852 μm), with an Euclidean distance of 1.85 from the input centroid, resulting in a matching degree of 1 / 1.85 = 0.54; Strategy 2 (15%, 85%) has an Euclidean distance of 2.12, resulting in a matching degree of 0.47; and Strategy 3 (25%, 75%) has an Euclidean distance of 2.35, resulting in a matching degree of 0.43. Strategy 1, with the highest matching degree, is selected. The function output is as follows: biomass blending ratio is 20%, coal powder fineness is 80%, expected feed rate correction coefficient = 1 + (70% - 60.2%) / 100 = 1.098, primary air distribution ratio (piecewise linear function output corresponding to volatile matter 72.7%) = 0.38, secondary air distribution ratio (oxygen deviation 4.0% - 3.5% = 0.5% calculated by PI controller) = 0.42. The proportion feedforward control functions of all 8532 state nodes in the working condition response Markov chain are summarized to form a total object of proportion feedforward control function set, which is implemented by hash mapping with state number as key (e.g., S1411030308010) and proportion feedforward control function as value.

[0101] Specifically, step S3, through forward simulation and multi-objective optimization, solves the technical problem of how to predetermine the optimal blending ratio based on possible future state evolution before the raw materials enter the furnace. Traditional blending strategies rely on experience or single-point calorific values ​​for static decision-making, which cannot predict the cumulative effects of control actions over multiple time steps, easily leading to local optima and global suboptimal results. The blending feedforward control function set utilizes the overall structure of the operating condition response Markov chain, embedding the local transition information of each state node and multi-step response predictions into the function, making the control decision inherently forward-looking and robust. By evaluating the long-term performance of different control combinations through Monte Carlo simulation, the algorithm can identify control strategies that seem effective in the short term but will cause tar accumulation or load oscillations in the long term, and eliminate them in the evaluation function through environmental and economic sub-items. This simulation-based optimization method avoids the necessity of dangerous trial and error on real units, transforming the optimization problem of high-dimensional non-convex response surfaces into a discrete candidate strategy ranking problem, realizing a computable and callable feedforward decision mapping at the control level. The proportioning feedforward control function set encapsulates the control logic using a functional programming paradigm, so that the control system does not need to care about the internal optimization details. It can obtain the optimal control command simply by using the state index, thereby decoupling the decision-making layer from the execution layer. This facilitates online updates and maintenance, and also provides a standardized interface for rapid strategy switching when raw materials change.

[0102] Step S4: Embed the biomass and coal ratio feedforward control function set into the fuel feeding and boiler automatic control system, and call the corresponding function according to the real-time identified status to dynamically adjust the biomass and coal ratio and combustion conditions. Further, step S4 includes the following sub-steps:

[0103] Step S41: During the operation phase, continue to collect online characteristic data of biomass and coal in real time, and use the same feature coordinate system as when constructing the phase space grid of raw material characteristics to locate the current raw material characteristics, identifying the corresponding state number of the current operating condition in the Markov chain of the operating condition response. Further, step S41 includes the following three-level sub-steps:

[0104] Step S411: Maintain the continuous operation of the online detection device set up in step S11, and collect the online detection data packet of the raw material at a cycle of 10 seconds. The data packet format is completely consistent with that used when constructing the phase space grid of the raw material characteristics in the early stage, and includes volatile matter, moisture, ash, fixed carbon, particle size data, as well as timestamp and feeding point location number.

[0105] Step S412: The collected real-time raw material characteristic data is input into the state recognition module. The execution flow of the state recognition module is as follows: First, five feature values—volatile matter, moisture, ash, fixed carbon, and particle size—are extracted from the real-time data to form a real-time feature vector. Then, a nearest neighbor search is performed on the entire raw material characteristic phase space grid object. The search algorithm uses a KD-tree acceleration structure. The KD-tree is pre-built and continuously updated after step S13. The nearest neighbor search returns the raw material characteristic phase space grid cell with the smallest Euclidean distance to the real-time feature vector. Next, the six-dimensional interval index combination of this cell is extracted, and a candidate state number is generated according to the encoding rules defined in step S23. Finally, it is verified whether the sample counter of the candidate state number is greater than 0. If it is greater than 0, the state is confirmed as a valid state and output as the current state number. For example, real-time data is input into the state recognition module. The real-time feature vector (71.8, 23.5, 9.5, 17.9, 862) is extracted. A nearest neighbor search is performed within the overall material characteristic phase space grid object. The KD tree returns the cell with the smallest Euclidean distance to this vector (14, 11, 3, 3, 8, 10), which is 1.23. The six-dimensional interval index combination (14, 11, 3, 3, 8, 10) of this cell is extracted, and a candidate state number S1411030308010 is generated according to the encoding rules. The sample counter for this state number is verified to be 156, which is greater than 0, confirming it as a valid state. The state recognition module outputs: current state number S1411030308010, the raw material characteristic centroid corresponding to the current state (72.7%, 22.3%, 9.2%, 18.4%, 845 micrometers), the response characteristic centroid corresponding to the current state (5.72, 85, 1058, 3.9), and the out-edge transition probability vector of the current state in the working condition response Markov chain [P(i,i)=0.885, P(i,j)=0.034, P(i,k)=0.021,...].

[0106] Step S413: If the sample counter of the candidate state number is equal to 0, it indicates that the raw material characteristic region has never been visited in history and belongs to an unknown state. At this time, the state rollback strategy is executed. The state rollback strategy is to expand the search range of each dimension interval step by step. Each time, the index of each dimension interval is expanded one step to the adjacent interval, and the nearest neighbor cell with the sample is searched again until a valid state is found. The output of the state recognition module includes the current state number, the raw material characteristic centroid corresponding to the current state, the response characteristic centroid corresponding to the current state, and the out-edge transition probability vector of the current state in the working condition response Markov chain. The state recognition process is executed once every 10 seconds to ensure that the control system can capture subtle changes in the raw material characteristics in a timely manner. When a sudden change occurs in the raw material batch, the real-time feature vector will quickly jump to a different region of the raw material characteristic phase space grid. The state recognition module can immediately detect this jump in the next acquisition cycle and output a new state number to achieve a millisecond-level response to the state switch.

[0107] Step S42: The control system selects a matching biomass blending feedforward control function from the set of blending feedforward control functions based on the current state number. Using the state and its local transition characteristics as input, it calculates the biomass blending ratio to be used at the current moment, the corresponding pulverized coal feed rate and biomass feed rate setpoints, as well as the recommended blending ratios for primary air, secondary air, and air distribution zones. Further, step S42 includes the following three sub-steps:

[0108] Step S421: Use the current state number output in step S41 as the key value to perform a hash lookup operation in the overall object of the ratio feedforward control function set. The hash lookup uses open addressing to handle collisions and ensures that the corresponding ratio feedforward control function is located in constant time.

[0109] Step S422: Call the feedforward control function for the mix proportion. The input parameters of the function include the raw material characteristic centroid of the current state, the response characteristic centroid of the current state, the upper limit of the forward simulation step count (10), the target load setpoint, and the current measured oxygen content. The feedforward control function first selects the strategy with the highest matching degree from the set of preferred control strategies for the current state node based on the input parameters. In addition to considering the Euclidean distance, the matching degree calculation also introduces a load deviation penalty term. That is, if the average load of a certain preferred strategy deviates significantly from the target load setpoint, its matching degree is multiplied by a penalty factor less than 1. The penalty factor is linearly determined based on the absolute value of the load deviation.

[0110] Step S423: After selecting the strategy with the highest matching degree, the function outputs the biomass blending ratio value B and the pulverized coal fineness value F corresponding to this strategy; calculate the biomass feed rate setpoint based on the biomass blending ratio value B, the calculation formula being the total fuel demand multiplied by B, the total fuel demand being calculated by the boiler main control loop according to the load command; calculate the pulverized coal feed rate setpoint as the total fuel demand multiplied by (1-B); adjust the separator speed or air-coal ratio of the pulverizing system according to the pulverized coal fineness value F, so that the actual pulverized coal fineness approaches F; calculate the primary air distribution ratio based on the volatile matter content in the current state of the raw material properties, the higher the volatile matter content, the larger the primary air ratio, to provide sufficient oxygen to support the volatile matter release and combustion, the primary air distribution ratio is determined using a piecewise linear function, the function... The nodes are pre-calibrated through combustion tests. The secondary air distribution ratio is calculated based on the deviation between the current measured oxygen level and the target oxygen level. Secondary air is used to supplement the oxygen required for combustion and control the furnace temperature distribution. The secondary air distribution ratio is calculated using a PI controller, but the controller parameters are adaptively adjusted based on the oxygen fluctuation variance in the response characteristics of the current state. The larger the oxygen fluctuation variance, the smaller the controller gain, to enhance stability. The calculated biomass blending ratio, biomass feed rate setpoint, pulverized coal feed rate setpoint, primary air distribution ratio, and secondary air distribution ratio are encapsulated into a control instruction package. The format of the control instruction package conforms to the boiler distributed control system communication protocol and includes an instruction timestamp, instruction source identifier, instruction priority, and setpoint fields for each control loop.

[0111] Step S43: The control quantity given by the feedforward control function set is used as the feedforward setting input to the boiler combustion control loop, and coordinated with the furnace temperature and oxygen feedback signals to achieve coordinated regulation of feedforward and feedback. When a sudden change occurs in the raw material batch, the state recognition immediately switches to the new state, and the corresponding feedforward control function also switches accordingly. The control system can execute the adapted new proportioning strategy without remodeling. Further, step S43 includes the following three sub-steps:

[0112] Step S431: The boiler combustion control loop includes a fuel main control loop, an forced draft control loop, an induced draft control loop, and a furnace pressure control loop. The setpoint of the fuel main control loop receives the biomass feed rate setpoint and the pulverized coal feed rate setpoint from the proportioning feedforward control function setpoint, and simultaneously receives the total fuel quantity correction signal from the boiler main control loop. The fuel main control loop adopts a cascade control structure, with the main regulator being the total fuel quantity regulator, and the secondary regulators being divided into a biomass feed rate regulator and a pulverized coal feed rate regulator. The secondary regulator receives the feedforward setpoint as the main setpoint and receives the output of the total fuel quantity regulator as the secondary setpoint, implementing feedforward-feedback composite control. Precise feeding is achieved. The setpoint of the air supply control loop receives the primary air distribution ratio and secondary air distribution ratio from the proportioning feedforward control function set. The air supply control loop adopts cross-limit control logic to ensure that the total air supply volume is not less than the theoretical air volume required for complete fuel combustion, while preventing the furnace temperature from becoming too high. The cross-limit value is dynamically adjusted according to the maximum temperature value in the response characteristics of the current state. The induced draft control loop maintains a slightly negative furnace pressure. Its setpoint is proportionally feedforward compensated according to the air supply volume. The compensation coefficient is determined by collecting experimental data on the correlation between air supply volume, air leakage, and furnace pressure under different operating conditions, and then verifying it through data fitting analysis and actual trial operation.

[0113] In step S432, the furnace temperature and oxygen feedback signals come from the temperature sensor array and oxygen analyzer arranged in step S21, respectively. The feedback signals enter the feedback compensation module, which calculates the feedback correction amount. The correction amount calculation adopts the model predictive control concept, that is, a small state-space model is established to describe the dynamic relationship between the control quantity and the controlled quantity. The model is used to predict the temperature and oxygen change trends at several future sampling times. The deviation between the predicted value and the set value is solved by quadratic programming to obtain the optimal feedback correction amount. The correction amount is added to the feedforward set value in the form of an increment.

[0114] Step S433: When a sudden change in the raw material batch causes a change in the state number, the entire object of the proportioning feedforward control function set immediately outputs the control strategy corresponding to the new state. The setpoints of the fuel main control loop and the air supply control loop are switched within the next control cycle. The switching process uses a setpoint ramp transition, and the ramp rate is set according to the mechanical constraints of the equipment to prevent violent movement of the actuator. At the same time, the state space model parameters in the feedback compensation module are reconstructed online according to the response characteristics of the new state. The model parameters are quickly extracted from historical data using a subspace identification algorithm to ensure that the feedback control loop is quickly adapted to the characteristics of the new raw material. In the entire control architecture, the proportioning feedforward control function set acts as the decision-making brain, and the boiler combustion control loop acts as the execution limb. The feedforward path achieves rapid coarse adjustment, and the feedback path achieves precise fine adjustment. The two work together to suppress the disturbance of raw material fluctuations to the furnace operating conditions. See also Figure 4 This is a schematic diagram of a feedforward-feedback collaborative control provided in an embodiment of this application. Figure 4The upper-middle layer blue solid arrows form the feedforward path: online raw material detection data (volatile matter, moisture, particle size, etc.) are input into the status recognition module, which generates the current status number through six-dimensional phase space grid matching. This number serves as a key-value query for the feedforward control function set (including strategy switching mechanism), and instantly outputs feedforward setpoints such as biomass ratio, pulverized coal fineness, and primary / secondary air distribution ratio. The middle layer red dashed arrows form the feedback path: the boiler furnace sensor array collects furnace temperature and oxygen levels in real time and sends them to the feedback compensation module (built-in model predictive control MPC). Based on the state space model, it predicts future time-series deviations and calculates feedback corrections. At the superposition point (the "+" node in the diagram), it is vector-superimposed with the feedforward setpoints in an incremental manner. The lower black arrows represent the execution and object link: the superimposed composite command is sent to the boiler combustion control loop (fuel main control / air supply main control), driving the feeder, dampers, and other actuators to operate, ultimately acting on the controlled objects in the boiler furnace.

[0115] Specifically, step S4 solves the technical problem of reliably implementing complex feedforward proportioning decisions to actual units by embedding the proportioning feedforward control function set into the actual control system. Traditional methods typically hard-code proportioning strategies into the control logic using empirical tables or fixed curves. When raw material characteristics change significantly, manual parameter retuning is required, which is time-consuming and risky. The proportioning feedforward control function set encapsulates the decision logic with a functional interface. The control system only needs to use state numbers to switch strategies without needing to know the internal optimization details, achieving loose coupling between the decision-making and execution layers. This design makes online updates and iterations of the proportioning strategy possible. After accumulating sufficient new data, steps S1 to S3 can be re-executed in the background to update the proportioning feedforward control function set, while the front-end control system can hot-update the function mapping relationship without shutting down the system, greatly improving the maintainability and scalability of the system. The coordinated adjustment mechanism of feedforward and feedback fully utilizes the prior predictive capability of the feedforward control function set and the real-time error correction capability of the feedback loop. The feedforward path, based on Markov chain simulation, compensates for the main impacts of raw material changes in advance, while the feedback path, based on model predictive control, eliminates residual deviations and unknown disturbances. The combination of the two ensures both rapid response and control accuracy. When raw material batches change abruptly, the millisecond-level response capability of state recognition and function switching enables the control system to complete the ratio adjustment in a very short time before or immediately after the fuel enters the furnace, suppressing boiler load fluctuations and combustion instability from the source. Compared with traditional adjustment methods that rely on lagging feedback such as furnace temperature and oxygen content, this advances the control intervention time by tens of seconds to several minutes, significantly reducing the risk of combustion runaway. At the same time, since the fineness of pulverized coal is included as a control variable in the optimization, the control system can dynamically adjust the fineness of pulverized coal according to the biomass blending ratio, achieving coordinated optimization of combustion characteristics and solving the problem of mismatch between pulverized coal and biomass combustion under traditional fixed fineness.

[0116] For example, a typical workflow of this method is as follows: During the unit startup phase, the system first executes step S1, using historical data from the previous cycle to construct an initial feedstock characteristic phase space grid object. This grid covers a typical operating condition area with 18% to 25% volatile matter in pulverized coal and 15% to 30% moisture in biomass during winter. Then, step S2 is executed, mapping data such as the calorific value of the gasified gas outlet and the furnace temperature to the grid cells to construct an initial operating condition response Markov chain object. The probability of transitioning from a high volatile matter, high moisture state to a low calorific value state is identified as 0.3, and the conditional response change vector is a decrease in calorific value of 0.5 MJ / m³. Next, step S3 is executed, performing a 10-step forward simulation on 168 control combinations for each state node in the computer cluster. It is found that when the biomass blending ratio is 20% and the pulverized coal fineness is 80%, the comprehensive score is the highest. Therefore, this strategy is encapsulated as a proportion feedforward control function and added to the proportion feedforward control function set object. During the unit's load operation phase, step S41 identifies the current raw material status in real time at 10-second intervals. On a certain day, due to a change in the biomass supplier, the moisture content jumped from 18% to 28%. The status identification module detected the status change within 10 seconds and switched from the original status number to the new status number. Step S42 immediately retrieves the optimization strategy from the feedforward control function corresponding to the new status, adjusting the biomass blending ratio from 25% to 18%, the coal powder fineness from 75% to 85%, increasing the primary air ratio by 3%, and adjusting the secondary air ratio to maintain oxygen at 3.5%. Step S43 sends the above control commands to the distributed control system. The fuel main control loop completes the feed rate switching within 30 seconds, and the air supply control loop synchronously adjusts the air volume. The furnace temperature fluctuation is controlled within 15 degrees Celsius, without triggering significant load fluctuations. The entire process requires no manual intervention. The feedforward control function set continuously receives new data in the background, automatically triggering an incremental update every 24 hours to ensure that the decision model remains consistent with the latest raw material characteristics.

[0117] Example 2:

[0118] This embodiment, based on Embodiment 1, provides a biomass gasification coupled with coal-fired power generation ratio system adapted to the characteristics of raw materials, such as... Figure 5 As shown, it includes:

[0119] Raw material property data acquisition and phase space grid construction module: used to collect online and laboratory analysis data of biomass and coal raw materials, construct a unified feature coordinate system and discretize it to generate raw material property phase space grid;

[0120] Operating condition response Markov chain construction module: used to integrate gasification products and boiler combustion online monitoring data to construct an operating condition response Markov chain along the temporal evolution path of the feedstock characteristic phase space grid;

[0121] Proportion feedforward control function set generation module: used for forward simulation and multi-objective optimization based on the Markov chain of operating condition response, to generate proportion feedforward control function set;

[0122] Control execution and coordinated adjustment module: It is used to embed the set of feedforward control functions for biomass ratio into the automatic control system, call the corresponding function according to the real-time status, and combine feedback signals to realize the dynamic adjustment of biomass and coal ratio and combustion conditions.

Claims

1. A method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation, characterized in that, The method includes: S1: Collect online and laboratory analysis data of biomass and coal feedstocks, construct a unified feature coordinate system and discretize it to generate a phase space grid for characterizing the characteristics of feedstocks with batch and time variations; S2: Integrate online monitoring data of gasification products and boiler combustion, and construct a Markov chain of operating condition response along the temporal evolution path of the feedstock characteristic phase space grid; S3: Based on the Markov chain of operating condition response, forward simulation of state transition is performed under different mixing assumptions to generate a set of feedforward control functions that provide fuel ratio and auxiliary control quantities for each state; S4: Embed the biomass and coal ratio feedforward control function set into the fuel feeding and boiler automatic control system, and call the corresponding function according to the real-time identified status to perform dynamic adjustment of the biomass and coal ratio and combustion conditions; the step of constructing the condition response Markov chain includes: Data on the gasification and combustion processes are simultaneously collected at the gasifier outlet and key locations in the furnace, and aligned with the raw material collection timestamps to form a process monitoring data stream. Based on the raw material characteristic phase space grid cell into which the raw material falls at each moment, the gasification and combustion responses at that moment are statistically recorded in the corresponding cell, upgrading the raw material characteristic phase space grid to a raw material-response joint distribution carrier. The movement trajectory of raw material characteristics on the raw material characteristic phase space grid is tracked chronologically, each visited raw material characteristic phase space grid cell is considered a state, and the transitions between adjacent raw material characteristic phase space grid cells are recorded as state transitions. The transition frequency and conditional response changes between each state pair are statistically obtained, constructing a condition-response Markov chain. The steps for dynamically adjusting the biomass-coal ratio and combustion conditions include: During the operation phase, online characteristic data of biomass and coal are continuously collected in real time to locate the current raw material characteristics and identify the corresponding state number in the working condition Markov chain of the working condition response. The control system selects a matching function from the set of proportioning feedforward control functions based on the current state number, and calculates the biomass blending ratio, feed rate setpoint, and air distribution ratio to be used at the current moment. The control quantity is used as a feedforward setting input to the boiler combustion control loop, and works in conjunction with the furnace temperature and oxygen feedback signals to achieve coordinated regulation of feedforward and feedback, and the ratio strategy is switched synchronously when the state changes.

2. The method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation according to claim 1, characterized in that, The steps for generating the phase space mesh for raw material properties include: Online detection devices are installed at the measuring points of the conveyor belt, feeding hopper and pulverizing system to obtain industrial analysis indicators of biomass and coal, and record the corresponding timestamps and feeding point location information. The industrial analysis indicators in the raw material characteristic data source are mapped to a multi-dimensional feature coordinate system, with each dimension corresponding to the key raw material characteristics. Then, intervals are divided on each dimension according to standards or based on the statistical distribution of historical data, forming a raw material characteristic phase space grid on the multi-dimensional feature coordinate system. Statistical information for each material property phase space grid cell is maintained in the material property phase space grid.

3. The method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation according to claim 2, characterized in that, The steps for obtaining industrial analytical parameters for biomass and coal include: Online moisture meters and near-infrared spectrometers were installed at the measuring points on the conveyor belts; weighing belts and rapid ash analyzers were installed at the measuring points on the feeding hoppers; and particle size analyzers and industrial cameras were installed at the measuring points on the powder making system. Data on raw material industrial analysis indicators are collected in real time through various online detection devices; The output data of all online detection devices are bound to the timestamps generated by a unified clock source, and the material loading point location number is marked to form a raw material online detection data package; at the same time, standard industrial analysis is performed on the biomass and coal raw materials before mixing in the laboratory on a daily or batch-by-batch basis to form a laboratory calibration data package; the raw material online detection data package and the laboratory calibration data package together constitute the raw material characteristic data source.

4. The method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation according to claim 1, characterized in that, The specific steps for constructing a condition-response Markov chain include: Traverse all elements in the phase space grid of raw material properties, filter out elements with sample counters greater than 0, define each element that meets the conditions as a working condition response Markov chain state node, and create a state attribute record for each state node. Based on the timestamp order, the cells that fall into two adjacent time points are paired, the state transitions are recorded, the frequency of all directed transitions is counted, and a state transition frequency matrix is ​​constructed. The state transition frequency matrix is ​​normalized row by row to obtain the state transition probability matrix. The condition response change vector is calculated. The set of state nodes, the state transition probability matrix, the set of condition response change vectors, and the set of state attribute records are encapsulated to form the overall object of the condition response Markov chain.

5. The method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation according to claim 1, characterized in that, The steps for generating the set of proportion feedforward control functions include: The biomass blending ratio and coal powder fineness parameters are regarded as control variables, and a set of candidate control combinations are set for each state node in the working condition response Markov chain. For each candidate control combination, the transition probability of the Markov chain of the operating condition response is used to perform multi-step forward simulation on the chain to predict the state sequence visited within a finite number of steps and the corresponding fluctuations in gas calorific value, tar enrichment trend and boiler load change. The forward simulation results are comprehensively and quantitatively evaluated with safety constraints, efficiency targets and environmental protection indicators. A set of optimal control strategies is selected for each state node, and a set of proportional feedforward control functions is constructed.

6. The method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation according to claim 5, characterized in that, The steps involved in performing a multi-step forward simulation include: Define an upper limit for the number of forward simulation steps. For each candidate control combination at each state node, start an independent forward simulation process and initialize the current state, gas calorific value, tar content and boiler load. The simulation process follows the random walk rule of Markov chains. In each step, the next state node is sampled and determined. The relevant parameters are updated using the conditional response change vector. The transition and update process is repeated until the upper limit of the number of steps is reached or the state of absorption is entered. The simulation trajectory is recorded. For each candidate control combination, Monte Carlo simulations are repeated multiple times. The simulation results are aggregated, relevant statistics are calculated, and stored in the forward simulation results database.

7. The method for adapting raw material characteristics to biomass gasification coupled with coal-fired power generation according to claim 6, characterized in that, The steps to achieve coordinated regulation of feedforward and feedback include: The fuel main control loop adopts a cascade control structure to match the biomass and pulverized coal feed rates; the air supply control loop adopts cross-limit control logic to ensure that the total air supply volume matches the fuel quantity; the induced draft control loop maintains a slight negative pressure in the furnace and performs proportional feedforward compensation based on the air supply volume. The feedback correction is calculated based on the furnace temperature and oxygen content feedback signals and then added to the feedforward setpoint. When a sudden change in the raw material batch leads to a state switch, the control system quickly switches the control strategy, adopts a setpoint ramp transition, and reconstructs the model parameters of the feedback compensation module online to achieve feedforward and feedback synergy to suppress disturbances.

8. A biomass gasification coupled with coal-fired power generation system adapted to the characteristics of raw materials, used to implement the method of any one of claims 1-7, characterized in that, The system includes: Raw material property data acquisition and phase space grid construction module: used to collect online and laboratory analysis data of biomass and coal raw materials, construct a unified feature coordinate system and discretize it to generate raw material property phase space grid; Operating condition response Markov chain construction module: used to integrate gasification products and boiler combustion online monitoring data to construct an operating condition response Markov chain along the temporal evolution path of the feedstock characteristic phase space grid; Proportion feedforward control function set generation module: used for forward simulation and multi-objective optimization based on the Markov chain of operating condition response, to generate proportion feedforward control function set; Control execution and coordinated adjustment module: It is used to embed the set of feedforward control functions for biomass ratio into the automatic control system, call the corresponding function according to the real-time status, and combine feedback signals to realize the dynamic adjustment of biomass and coal ratio and combustion conditions.

Images