Adaptive Predictive Control Method and System for Boiler Flue Gas Temperature under Wide Load Operation

Abstract

This invention relates to the field of predictive control technology, specifically to an adaptive predictive control method and system for boiler flue gas temperature under wide load operation. The method includes the following steps: acquiring actual and predicted flue gas temperatures and calculating residual directional quantities; determining load-temperature lag by combining main steam flow, coal feed rate, and flue gas temperature changes; adjusting damper opening boundaries, coal feed frequency boundaries, and prediction step size; constraining and reconstructing the flue gas temperature prediction sequence to obtain a flue gas temperature predictive control command sequence; comparing measured and predicted flue gas temperatures point-by-point and refining residual directional quantities to convert temperature deviations into continuous information, reducing instantaneous disturbances; combining time-shifted comparisons of main steam flow and flue gas temperature changes to identify the lag relationship between load transfer and flue gas temperature response, making the adjustment closer to actual causality; and performing dynamic boundary convergence and relaxation processing on combustion-related setpoints to synchronously correct the control range and response rhythm with changes in operating conditions, improving flue gas temperature tracking accuracy and suppressing overshoot risk.

Description

Technical Field

[0001] This invention relates to the field of predictive control technology, and in particular to a method and system for adaptive predictive control of boiler flue gas temperature for operation under wide load conditions. Background Technology

[0002] Predictive control technology involves control methods that estimate future behavior using dynamic mathematical models of the controlled object and generate control inputs accordingly. Its core aspects include object modeling, state variable selection, predictive time-domain rolling optimization, constraint handling, and online parameter correction. Typically, it describes the system's dynamic characteristics by establishing state-space equations or difference equations, and calculates the control sequence and implements the first control variable in each control cycle using a performance index function, thereby achieving the regulation of continuous industrial process variables. Among these, the traditional boiler flue gas temperature adaptive predictive control method refers to a control approach that adjusts flue gas temperature changes under different load conditions. Addressing the dynamic fluctuations in flue gas temperature during boiler combustion caused by changes in fuel quantity, air supply, and load, traditional methods typically employ a linear model predictive control based on the boiler combustion process. This involves selecting flue gas temperature, fuel feed rate, and air supply as state variables, using a discrete-time model with fixed parameters to predict the temperature several steps ahead, and obtaining the control input by solving a performance index function with a temperature deviation squared term and a control increment constraint term. Simultaneously, proportional-integral regulation is used to correct the prediction error to achieve flue gas temperature regulation.

[0003] In actual operation, existing technologies mostly rely on fixed structural prediction relationships to handle flue gas temperature regulation. The control judgment focuses on the current temperature deviation and changes in conventional control quantities. When faced with thermal inertia, heat transfer lag, load transfer delay, and ventilation coupling disturbances during boiler combustion, it is difficult to distinguish the source of deviation. Temperature anomalies and load lag are often treated as the same thing, which makes the control action prone to responding directly to the apparent deviation, lacking detailed identification of the direction of deviation continuation, the state of deviation accumulation, and the causal sequence. For example, after a sudden increase in unit load, the main steam flow has changed in advance, but the flue gas temperature response is still in the lag stage. If the coal feed or air supply is still adjusted according to the established rhythm, it may be necessary to apply strong correction in advance before the flue gas temperature has fully changed. Subsequently, due to the gradual release of heat, reverse fluctuations occur, resulting in an expansion of the flue gas temperature fluctuation. Another shortcoming lies in the fact that the adjustment boundary and prediction range are mostly set statically. The constraint scale and prediction length are almost identical during the stable operation period and the period of increased disturbance, making it difficult to balance conservatism and flexibility. When the negative pressure fluctuations increase, the induced draft conditions change, or the fuel quality fluctuates, the fixed boundary can easily lead to the control quantity being too tight and insufficient correction, or too wide and causing the actuator to frequently and significantly move, which increases wear and amplifies flue gas temperature fluctuations. For example, under long-term delay conditions, a uniform prediction range may ignore the impact of subsequent heat transfer, while in short-term strong disturbance scenarios, it may divert attention from near-end risks. The result is slow tracking, large overshoot, and insufficient ability to maintain the stable zone. In severe cases, it can also cause high flue gas temperature, increased thermal deviation of the heating surface, and decreased combustion efficiency, thereby compressing the boiler's economic operating space and safety margin. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive predictive control method for boiler flue gas temperature for wide-load operation, comprising the following steps: S1: Collect the flue gas temperature values ​​output by the thermocouples at the boiler furnace temperature measuring point and the tail flue gas temperature measuring point, collect the predicted flue gas temperature values ​​at the corresponding time, calculate the difference between the flue gas temperature values ​​and the predicted flue gas temperature values ​​point by point to obtain the residual sequence, sort them according to the sampling period, accumulate the positive and negative values ​​of the residual sequence respectively, and calculate the difference to obtain the flue gas temperature residual direction quantity. S2: Based on the flue gas temperature residual direction quantity, the main steam flow meter signal of the steam turbine and the flow signal of the coal feeder belt scale are collected and sorted according to the sampling period. The flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point is collected. The main steam flow meter signal is time-shifted and the difference is calculated with the flue gas temperature change sequence. The time shift result is statistically analyzed and the time offset position is filtered. The main steam flow change at the time offset position is called and the sign consistency is judged with the flue gas temperature residual direction quantity to obtain the load flue gas temperature lag judgment result. S3: Based on the load flue gas temperature hysteresis determination result, collect the opening setting value of the secondary air damper actuator and the frequency setting value of the coal feeder frequency converter, and perform boundary interval adjustment on the opening setting value and the frequency setting value to obtain the combustion regulation parameter boundary set; S4: Based on the load flue gas temperature hysteresis determination result, obtain the number of sampling steps for the prediction sequence and collect the pressure signal change status of the furnace negative pressure measuring point, and adjust the number of sampling steps to obtain the prediction time domain step set; S5: Based on the combustion adjustment parameter boundary set and the prediction time step set, call the current cycle flue gas temperature prediction sequence and collect the opening setting value of the induced draft blade adjustment actuator, perform parameter boundary constraints on the flue gas temperature prediction sequence and reconstruct the time step to obtain the flue gas temperature prediction control command sequence.

[0005] As a further aspect of the present invention, the residual direction quantity of flue gas temperature includes residual trend intensity, cumulative deviation amplitude, and offset direction identifier; the load flue gas temperature lag determination result includes lag time offset, response consistency identifier, and load correlation degree; the combustion adjustment parameter boundary set includes upper limit of air volume adjustment, lower limit of fuel input, and parameter change margin; the prediction time domain step size set includes short-term prediction step size, medium-term prediction step size, and dynamic adjustment coefficient; the flue gas temperature prediction control command sequence includes control output gradient, adjustment rhythm parameter, and execution constraint combination.

[0006] As a further aspect of the present invention, in the calculation of the smoke temperature residual direction quantity, the cumulative results of positive values ​​and the cumulative results of negative values ​​of the residual sequence are accumulated by a sliding window with a length of 5 sampling periods, and the difference is calculated at the end of each sliding window to obtain the local smoke temperature residual direction quantity. During the time shifting process, the main steam flow meter signal is shifted sequentially in steps of two sampling periods, and the time offset position corresponding to the smallest absolute value of the difference between the shifted signal and the flue gas temperature change sequence is used as the filtering time offset position.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Collect the flue gas temperature values ​​output by the thermocouples at the boiler furnace temperature measurement point and the flue gas temperature measurement point at the tail flue, synchronously collect the predicted flue gas temperature values ​​at the corresponding time, organize the node collection sequence, and obtain the multi-node flue gas temperature collection data. S102: Call the thermocouple output value and the predicted smoke temperature value within the multi-node smoke temperature acquisition data, calculate the point-by-point difference between the thermocouple output value and the predicted smoke temperature value, arrange the difference sequence according to the sampling period, and generate the sorting period residual value. S103: For the residual values ​​of the sorting period, extract the positive and negative residual values, perform cumulative calculations on the positive and negative residual values ​​respectively, calculate the difference between the positive and negative cumulative terms, and obtain the smoke temperature residual direction quantity.

[0008] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the flue gas temperature residual direction quantity, collect the main steam flow meter signal of the steam turbine and the flow signal of the coal feeder belt scale, arrange the signal sequence according to the sampling period, call the flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point at the corresponding time, and establish a flow temperature synchronization sequence. S202: For the flow and temperature synchronization sequence, perform time shift processing on the main steam flow signal, calculate the point-by-point difference between the shifted main steam flow signal and the flue gas temperature change, statistically analyze the position of the minimum difference, and obtain the offset time positioning amount. S203: The offset time positioning quantity is called to extract the corresponding main steam flow change quantity. The maximum value of the main steam flow change quantity and the peak value of the flue gas temperature residual direction are introduced. The consistency operation is performed by combining the main steam flow change quantity and the flue gas temperature residual direction quantity to obtain the hysteresis consistency coefficient. The mapping is performed by combining the state judgment interval to obtain the load flue gas temperature hysteresis judgment result.

[0009] As a further aspect of the present invention, the specific calculation formula for performing consistency calculation by combining the main steam flow rate change and the flue gas temperature residual direction is as follows: ; The lag consistency coefficient is calculated and mapped in combination with the state determination interval to obtain the load smoke temperature lag determination result; in, Represents the lag consistency coefficient. Represents the change in main steam flow rate and subscript Characterizing the main steam, Represents the directional quantity of the residual smoke temperature and its subscript Characterize residuals, This represents the maximum change in main steam flow rate. This represents the peak value in the direction of the residual smoke temperature.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the load flue gas temperature hysteresis determination result, obtain the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter, merge the corresponding equipment setting record sequence, and establish a combustion setting data group; S302: For the combustion setting data set, extract the opening setting extreme value and the frequency setting extreme value, calculate the deviation of the setting extreme value from the current setting value and perform extreme value range normalization processing to generate the boundary offset reference value; S303: Call the boundary offset reference value, extract the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter, combine the upper limit of the opening operation, the reference frequency of the frequency converter and the safety margin deviation limit to perform boundary correction calculation, calculate the adjustment boundary correction coefficient, and perform mapping processing with the original opening setting boundary and the original frequency setting boundary respectively to obtain the combustion adjustment parameter boundary set.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the load and temperature hysteresis determination result, obtain the current prediction sequence sampling step size, extract the corresponding sampling period and the previous prediction step size sequence, and establish a prediction step size basic data group. S402: Collect the pressure signal from the negative pressure measuring point in the furnace, extract the peak-valley difference of the pressure signal fluctuation per unit time, and generate the negative pressure change state quantity by combining it with the prediction step size basic data group. S403: Call the negative pressure change state quantity and the current prediction sequence sampling step size, calculate the ratio of the two to the pressure change set extreme value and the step size set extreme value respectively, obtain the step size adjustment coefficient based on the ratio, map the step size adjustment coefficient to the current prediction sequence sampling step size, and obtain the prediction time domain step size set.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the combustion adjustment parameter boundary set, call the current cycle flue gas temperature prediction sequence, collect the opening set value of the induced draft fan blade adjustment actuator, compare the predicted sequence temperature value with the boundary set extreme value parameter, limit the data amplitude of the over-limit node, and generate the boundary constraint prediction sequence; S502: For the boundary constraint prediction sequence, call the prediction time domain step size set to extract the corresponding step size time interval, and perform time axis repositioning and amplitude allocation operations on the sequence nodes according to the step size time interval to establish a step size reconstruction prediction sequence; S503: The step-size reconstruction prediction sequence is called to extract the current node temperature value of the sequence. Combined with the opening setting value of the induced draft fan blade adjustment actuator, the previous command change of the associated control node is extracted. The extreme value of the sequence set temperature, the limit value of the opening adjustment and the basic command reference value are introduced to perform parameter constraint adjustment calculation. The command reconstruction adjustment coefficient is calculated. The command reconstruction adjustment coefficient and the current node temperature value of the sequence are mapped to obtain the smoke temperature prediction control command sequence.

[0013] An adaptive predictive control system for boiler flue gas temperature designed for wide-load operation includes: The flue gas temperature residual extraction module acquires the flue gas temperature values ​​output by the thermocouples at the boiler furnace flue gas temperature measuring point and the tail flue gas temperature measuring point, and simultaneously collects the predicted flue gas temperature values ​​at the corresponding time. It calls the actual flue gas temperature values ​​and the predicted flue gas temperature values ​​to perform difference calculations point by point, and arranges the residual sequence according to the sampling period. It accumulates the positive and negative values ​​in the residual sequence and performs difference calculations to obtain the flue gas temperature residual direction quantity. The load lag determination module acquires the turbine main steam flow meter signal and the coal feeder belt scale flow signal based on the flue gas temperature residual direction quantity and arranges them according to the sampling period. At the same time, it calls the flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point at the corresponding time, performs time shift processing on the main steam flow signal and performs difference calculation with the flue gas temperature change, performs difference statistics on the time shift result and filters the corresponding time offset position, calls the main steam flow change corresponding to the time offset and performs sign consistency judgment with the flue gas temperature residual direction quantity to obtain the load flue gas temperature lag determination result. The combustion boundary generation module obtains the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter based on the load flue gas temperature hysteresis determination result, performs interval adjustment processing on the boundary corresponding to the opening setting value and the frequency setting value, and obtains the combustion adjustment parameter boundary set. The prediction step size construction module calls the current prediction sequence sampling step size number based on the load flue gas temperature hysteresis determination result, and calls the pressure signal change status of the furnace negative pressure measuring point to perform adjustment processing on the sampling step size, so as to obtain the prediction time domain step size set. The control sequence output module calls the current cycle flue gas temperature prediction sequence based on the combustion adjustment parameter boundary set and the prediction time step set, and simultaneously calls the opening setting value of the induced draft fan blade adjustment actuator. It performs parameter boundary constraint processing on the prediction sequence and reconstructs the time step to obtain the flue gas temperature prediction control command sequence.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by comparing the measured and predicted flue gas temperatures point by point and extracting the residual direction quantity, the temperature deviation is transformed into continuous information, reducing instantaneous disturbance interference. By combining the time-shift comparison of the main steam flow and flue gas temperature changes, the hysteresis relationship between load transmission and flue gas temperature response is identified, making the adjustment closer to the actual cause and effect. Dynamic boundary convergence and relaxation processing are performed on combustion-related setpoints, so that the control range and response rhythm are synchronously corrected with changes in operating conditions, improving the flue gas temperature tracking accuracy and suppressing the risk of overshoot. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides an adaptive predictive control method for boiler flue gas temperature for wide-load operation, comprising the following steps: S1: Obtain the flue gas temperature values ​​output by the thermocouples at the boiler furnace temperature measuring point and the tail flue gas temperature measuring point, and simultaneously collect the predicted flue gas temperature values ​​at the corresponding time. Call the actual flue gas temperature values ​​and the predicted flue gas temperature values ​​point by point to perform difference calculation and arrange the residual sequence according to the sampling period. Accumulate the positive and negative values ​​in the residual sequence and perform difference calculation to obtain the flue gas temperature residual direction quantity. S2: Based on the flue gas temperature residual direction quantity, the main steam flow meter signal of the steam turbine and the flow signal of the coal feeder belt scale are obtained and arranged according to the sampling period. At the same time, the flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point is called. The main steam flow signal is processed by time shift and the difference is calculated with the flue gas temperature change. The difference is statistically analyzed on the time shift result and the corresponding time offset position is selected. The main steam flow change corresponding to the time offset is called and the sign consistency judgment is performed with the flue gas temperature residual direction quantity to obtain the load flue gas temperature lag judgment result. S3: Based on the load flue gas temperature hysteresis determination result, obtain the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter, perform interval adjustment processing on the corresponding boundary of the opening setting value and the frequency setting value, and obtain the combustion regulation parameter boundary set; S4: Based on the load flue gas temperature lag determination result, call the current prediction sequence sampling step size and call the pressure signal change status of the furnace negative pressure measuring point to perform adjustment processing on the sampling step size and obtain the prediction time domain step size set; S5: Based on the set of combustion regulation parameters boundary and the set of predicted time step, call the current cycle flue gas temperature prediction sequence, and at the same time call the opening set value of the induced draft fan blade adjustment actuator. Perform parameter boundary constraint processing on the prediction sequence and reconstruct the time step to obtain the flue gas temperature prediction control command sequence.

[0020] The residual directional quantities of flue gas temperature include residual trend intensity, cumulative deviation magnitude, and offset direction identifier; the load flue gas temperature lag judgment results include lag time offset, response consistency identifier, and load correlation degree; the combustion regulation parameter boundary set includes upper limit of air volume regulation, lower limit of fuel supply, and parameter change margin; the prediction time domain step set includes short-term prediction step, medium-term prediction step, and dynamic adjustment coefficient; the flue gas temperature prediction control command sequence includes control output gradient, regulation rhythm parameter, and execution constraint combination.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Collect the flue gas temperature values ​​output by the thermocouples at the boiler furnace temperature measurement point and the flue gas temperature measurement point at the tail flue, synchronously collect the predicted flue gas temperature values ​​at the corresponding time, organize the node collection sequence, and obtain the multi-node flue gas temperature collection data. A matrix of K-type high-temperature thermocouple temperature sensors, positioned at the center section of the boiler furnace and the tail flue outlet, continuously collects multiple measured temperature signals at a sampling frequency of 200 milliseconds. The sensor operating range is set from 0 to 1500℃. After signal acquisition, the signals immediately enter the programmable logic controller (PLC) in the field control cabinet. The PLC performs median filtering and low-pass filtering on the input signals to remove high-frequency noise caused by electromagnetic interference and abnormal pulse data with amplitude jumps greater than 50℃. After filtering, the real-time effective flue gas temperature sequence for the furnace and tail flue regions at the corresponding time is obtained. Simultaneously, a synchronization request is sent to the advanced process control server via the Ethernet communication interface to retrieve the corresponding flue gas temperature values ​​predicted and output based on a long short-term memory network model at the same timestamp. The Long Short-Term Memory (LSTM) network model was previously trained using 6000 hours of historical operating data from the unit. The input layer included 24 feature variables, such as fuel quantity, primary and secondary air volume, and feedwater flow rate. Two hidden layers were set with 128 and 64 nodes respectively, both using ReLU activation functions. The output layer outputs predicted flue gas temperatures for multiple future nodes. To ensure absolute alignment between measured and predicted data on the timeline, the system reads the network time protocol timestamps carried in each data packet. For samples with transmission delays causing time deviations of no more than 50 milliseconds, cubic spline interpolation was used to calculate the resampled fitted values. The measured thermocouple output sequence and the model-predicted flue gas temperature sequence, after timeline alignment, were merged. Invalid nodes with missing values ​​at either end were removed. Dual-source temperature data from 300 sampling nodes over a continuous 300-second period were stored in the time-series database at a uniform 1-second interval, thus constructing a multi-node flue gas temperature acquisition data structure containing a time-series index and multi-dimensional temperature attributes.

[0022] S102: Call the thermocouple output value and the predicted smoke temperature value within the multi-node smoke temperature acquisition, calculate the point-by-point difference between the thermocouple output value and the predicted smoke temperature value, arrange the difference sequence according to the sampling period, and generate the sorted period residual value. For the multi-node smoke temperature acquisition data structure already stored in the time-series database, the measured thermocouple output value vector and the smoke temperature value vector predicted by the Long Short-Term Memory (LSTM) network model, which are time-series identical, are extracted separately. For this discrete data vector of length 300, a vectorized element-wise subtraction operation is performed. The operation rule is to subtract the predicted temperature value of the corresponding node from the measured thermocouple temperature value of the current node, resulting in an initial difference vector containing 300 positive and negative real number nodes. To reflect the fluctuation trend characteristics within a specific sampling period, this initial difference vector is divided into groups according to a 5-second sliding window, resulting in 60 observation periods. Within each observation period, these 5 consecutive difference data are fed into the quicksort algorithm module and sorted in ascending order of value. Quicksort selects the value at the middle position as the pivot element, placing smaller values ​​on the left and larger values ​​on the right, and iterates to complete the local sorting within the period. After locally sorting all 60 observation periods, these rearranged difference segments are spliced ​​together end-to-end according to the original time period sequence, forming a new, periodically ordered difference data queue with overall temporal progression characteristics. The temperature difference data within the same period in this queue exhibits a stepped distribution pattern from the lowest negative deviation to the highest positive deviation. The system stores this discrete difference sequence, which has undergone periodic ordering and filtered out high-frequency jitter information, in a memory buffer, forming the sorted period residual values.

[0023] S103: For the residual values ​​of the sorting period, extract the positive and negative residual values, perform cumulative calculations on the positive and negative residual values ​​respectively, calculate the difference between the positive and negative cumulative terms, and obtain the smoke temperature residual direction quantity; A traversal scan is performed on the sorted periodic residual value queue in the memory buffer, with a judgment benchmark of 0°C. When a node's residual value is detected to be strictly greater than 0°C, it is assigned to the positive polarity residual sequence; when a node's residual value is detected to be strictly less than 0°C, it is assigned to the negative polarity residual sequence; nodes equal to 0°C are ignored. After separation, a trapezoidal integral approximation algorithm is used to discretely accumulate the integral effects of the positive and negative polarity residual sequences over time. Specifically, the absolute values ​​of all elements in the positive polarity residual sequence are summed to calculate the sum of the positive deviation prediction model, denoted as the total positive area; the absolute values ​​of all elements in the negative polarity residual sequence are summed to calculate the sum of the negative deviation prediction model, denoted as the total negative area. After obtaining these two scalars representing the total degree of deviation, a subtraction operation is performed, that is, the scalar value of the total positive area is subtracted from the scalar value of the total negative area. The result of this difference directly reflects the tendency of the actual flue gas temperature to cumulatively deviate from the predicted flue gas temperature over the entire 300-second sampling time window. If the difference is positive, it indicates that the actual furnace combustion heat release level is generally in an overheating trend; if the difference is negative, it indicates that the actual flue gas temperature generally has an underheating trend. The system normalizes and encodes this final scalar difference value, which reflects the deviation characteristics of the macroscopic combustion state, and saves it to obtain the flue gas temperature residual direction quantity that characterizes the main trend of the deviation.

[0024] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the flue gas temperature residual direction quantity, the main steam flow meter signal of the steam turbine and the flow signal of the coal feeder belt scale are collected, the signal sequence is arranged according to the sampling period, the flue gas temperature change sequence of the corresponding time of the thermocouple at the boiler furnace flue gas temperature measuring point is called, and a flow temperature synchronization sequence is established. The analog input card of the distributed control system retrieves the 4-20 mA current signal from the orifice plate flow transmitter installed on the main steam pipeline before the inlet of the turbine high-pressure cylinder, and the frequency pulse signal from the weighing belt scale installed on the belt drive shaft of each coal feeder. After processing by an analog-to-digital converter, the analog signals are converted into engineering unit values ​​in tons per hour according to the preset instrument range. Both flow signals are synchronously and equidistantly resampled at 1-second sampling steps to establish a signal recording queue with 600 time nodes. Simultaneously, the forward differential algorithm is used to differentiate the obtained boiler furnace thermocouple flue gas temperature measurement sequence. Specifically, by calculating the algebraic difference between flue gas temperature values ​​at adjacent time nodes, the rate of change of flue gas temperature per second within the same time window is obtained, generating a flue gas temperature change sequence. The cleaned and unit-unified main steam flow signal vector, the total coal feeder feed signal vector, and the previously calculated flue gas temperature change sequence are strictly paired according to the network time protocol timestamp information of each data point. Abnormal record points with a time difference exceeding 100 milliseconds during the pairing process are removed, and missing points are filled in using linear interpolation. Finally, the parameters of these three dimensions are merged into a single two-dimensional data matrix. The rows of the matrix represent the time axis, and the columns correspond to the three state variables, thereby completing the construction and encapsulation of the flow and temperature synchronization sequence in memory.

[0025] S202: For the flow and temperature synchronization sequence, perform time shift processing on the main steam flow signal, calculate the point-by-point difference between the shifted main steam flow signal and the flue gas temperature change, statistically analyze the position of the minimum difference, and obtain the offset time positioning amount. The main steam flow rate signal and flue gas temperature change signal are extracted from the two-dimensional data matrix of the flow and temperature synchronization sequence. The maximum time shift window is set to 180 seconds, and the time shift step is 1 second. In the loop structure, the main steam flow rate signal sequence is shifted sequentially by 1 to 180 time nodes relative to the flue gas temperature change sequence. At each specific shift step, the data segment where the two coincide on the time axis is extracted, and the Pearson correlation coefficient and Euclidean distance of the corresponding node values ​​of the shifted main steam flow rate signal and the flue gas temperature change sequence at the same position are calculated. The Pearson correlation coefficient is used to represent the linear correlation by calculating the quotient of the product of the covariance and standard deviation of the two variables. To enhance the sensitivity of the search for the minimum value, a joint evaluation index is constructed. This index is the Euclidean distance divided by the absolute value of the correlation coefficient. The joint difference value of each node is calculated to generate the error evaluation scalar under this shift amount. After traversing all 180 possible shift steps, the system obtains a one-dimensional array containing 180 error evaluation scalars. The array is smoothed by applying a quadratic polynomial to fit the curve, and Newton's iteration method is used to find the coordinates of the position where the first derivative of the polynomial curve is 0 and the second derivative is greater than 0. The time shift step value corresponding to this global minimum error is found, the specific number of time offset nodes is recorded, and the shift amount at this time is established as the offset time positioning amount between combustion and steam production in the entire system, and stored in a global variable for subsequent logic calls.

[0026] S203: Call the offset time positioning quantity to extract the corresponding main steam flow change quantity, introduce the maximum value of the main steam flow change and the peak value of the flue gas temperature residual direction, combine the main steam flow change quantity and the flue gas temperature residual direction quantity to perform consistency calculation to obtain the lag consistency coefficient, combine the state judgment interval to perform mapping to obtain the load flue gas temperature lag judgment result, and convert the load flue gas temperature lag judgment result into the coal feeder coal feed quantity compensation control command and output it to the boiler combustion control system. The stored global variable offset time positioning value is retrieved. In the flow and temperature synchronization sequence, a local data segment corresponding to the current calculation cycle is extracted from the main steam flow sequence according to the offset time. The difference between the maximum and minimum values ​​within this data segment is calculated to extract the main steam flow change. The maximum value parameter of the main steam flow change defined under the rated operating conditions and full load operation of the unit is read. At the same time, the extreme peak value data of the flue gas temperature residual direction quantity clearly defined in the historical operation data statistics are retrieved. The obtained main steam flow change quantity (value 35), flue gas temperature residual direction quantity (value 120), maximum main steam flow change quantity (value 150), and flue gas temperature residual direction peak value (value 300) are substituted into the calculation formula of the aforementioned lag consistency operation, and the final result is 1.54. This result indicates that the system dynamic response characteristic parameter is currently in the medium lag range. The pre-defined state judgment intervals are as follows: when the calculated result is less than 0.8, it is in the extremely low lag interval; between 0.8 and 1.5, it is in the mild lag interval; between 1.5 and 2.5, it is in the moderate lag interval; and greater than 2.5, it is in the severe lag interval. By comparing the result 1.54 with the above four numerical intervals, the system automatically identifies that it belongs to the moderate lag interval. Based on the preset compensation coefficient of 1.2 for the moderate lag interval, a load flue gas temperature lag judgment result status word is generated. Using a digital-to-analog conversion interface, this status word with the intention of proportional amplification adjustment is converted into a standard control voltage command for the coal feeder frequency converter control, and output to the bottom-level actuator of the boiler combustion control system in the distributed control system. The advantage of the formula is that by introducing the absolute ratio of the maximum value of the main steam flow change to the peak value of the flue gas temperature residual direction, it effectively suppresses the command oscillation caused by small deviations under extreme operating conditions. Experimental data show that when the maximum change in main steam flow is set to 150, the overshoot rate of the control system following the command is reduced by 15%. Compared with the traditional method without peak normalization, the command jitter frequency is reduced by 28%, which significantly enhances the regulation stability.

[0027] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the load flue gas temperature hysteresis determination result, obtain the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter, merge the corresponding equipment setting record sequence, and establish a combustion setting data group; By analyzing the control compensation flag carried in the load flue gas temperature hysteresis determination result, a data read request is sent to the underlying device registers via the programmable logic controller's master-slave communication bus. The valve position opening percentage setting value command sent to all secondary damper actuators within the current working cycle is extracted from the servo amplifier module of the regulating damper; simultaneously, the current operating frequency (Hertz) setting value command sent to the lower-level frequency converter is read from the communication interface board of the coal feeder drive frequency converter. To ensure data continuity and traceability, the system continuously collects setting value change records over the past 10 minutes, with a sampling interval of 2 seconds, acquiring a total of 300 sets of two-dimensional data pairs. Each set of records, containing a timestamp, damper opening command value, and frequency converter frequency command value, is concatenated and encapsulated according to chronological order. A moving median filtering algorithm is used to smooth out abrupt noise in the sequence (i.e., data where the difference between two adjacent points exceeds 5% of the range), ensuring the continuity of data transition. Subsequently, the cleaned and time-sequentially sorted opening instruction array and frequency instruction array are structurally integrated and stored in a temporary data table in a relational database, thus completing the establishment of the combustion setting data group.

[0028] Table 1 Initial Equipment Setup Record of Monitoring Points As shown in Table 1, a structured record sequence containing time, opening degree and frequency was continuously collected and constructed in the database.

[0029] S302: For the combustion setting data set, extract the extreme values ​​of the opening setting and the frequency setting, calculate the deviation between the setting extreme value and the current setting value, perform extreme value range normalization processing, and generate boundary offset reference value; Access the combustion setting data group stored in the temporary data table, and perform extreme value scanning operations on the damper opening command array and the inverter frequency command array, which contain 300 sampling nodes. Using an extreme value search algorithm, find the maximum and minimum set percentages of damper opening and the maximum and minimum set Hertz of inverter frequency during this historical period, establishing their respective opening and frequency setting extreme values. Read the latest damper opening and inverter frequency setting values ​​issued at the current moment. For damper opening, calculate the absolute difference between the current opening setting value and the average of historical opening settings; for inverter frequency, calculate the absolute difference between the current frequency setting value and the average of historical frequency settings. Then, divide these two absolute differences by the difference between the maximum and minimum setting extreme values ​​in the corresponding dimension. Through this division operation based on the maximum and minimum limit range, the absolute deviation is converted into a dimensionless relative deviation rate value between 0 and 1. The relative deviation rates of these two dimensions are weighted and summed, with the weight of the damper opening deviation rate set to 0.4 and the weight of the coal feeder frequency deviation rate set to 0.6. This weighted summation directly generates a comprehensive evaluation scalar between 0 and 1. This scalar is retained and temporarily stored in the cache stack as a boundary offset reference for subsequent calls.

[0030] S303: Call the boundary offset reference value, extract the opening set value of the secondary damper actuator and the frequency set value of the coal feeder frequency converter, combine the upper limit of the opening operation, the reference frequency of the frequency converter and the safety margin deviation limit to perform boundary correction calculation, calculate the adjustment boundary correction coefficient, and perform mapping processing with the original opening set boundary and the original frequency set boundary respectively to obtain the combustion adjustment parameter boundary set. The calculated boundary offset reference value is extracted from the cache stack. Simultaneously, the upper limit of the absolute opening of the secondary damper actuator under mechanical limits (e.g., 90% full opening), the reference operating frequency of the coal feeder frequency converter under the corresponding load (e.g., 35 Hz), and the safety margin deviation limit (e.g., the 5% safety limit reserved to prevent damper jamming) are read from the equipment operating parameter library. A linear boundary correction function is constructed. This function uses the product of the reference frequency and the upper limit of the absolute opening as the reference scaling factor. The extracted boundary offset reference value is subtracted from the safety margin deviation limit and then multiplied by this reference scaling factor. Subsequently, an adjustment boundary correction factor for dynamically scaling the equipment control limits is calculated. The factory default original opening setting boundaries (minimum 10%, maximum 85%) and original frequency setting boundaries (minimum 20 Hz, maximum 45 Hz) of the control system are read. Using the calculated adjustment boundary correction factor as a multiplier, scalar multiplication amplification or reduction mapping processing is performed on the upper and lower limits of the original opening setting boundaries and the upper and lower limits of the original frequency setting boundaries, respectively. After processing, new upper and lower limit values ​​are generated, and a set of combustion regulation parameter boundaries containing dynamically adjustable upper and lower limit constraints is constructed in the form of a hash mapping table. This set is then written into the system's memory-safe sandbox.

[0031] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the load and flue temperature hysteresis determination results, obtain the current prediction sequence sampling step size, extract the corresponding sampling period and previous prediction step size sequence, and establish a prediction step size basic data group. The load and smoke temperature hysteresis judgment results transmitted from the underlying control system are analyzed, and the dynamic response delay time tags contained therein are extracted. A query command is sent to the predictive control core module to obtain the number of sampling steps of the Long Short-Term Memory (LSTM) network smoke temperature prediction algorithm currently being executed. For example, the predictive controller is currently set to extrapolate 50 steps in the future time domain. Historical operation logs are retrieved to extract the absolute clock cycles corresponding to these 50 prediction steps. Simultaneously, tracing back, the previous prediction step sequence used by the predictive core module in the previous operation cycle and its corresponding time span record are extracted. The current number of sampling steps, the set of timestamps spanned by the current cycle prediction, and the number and timestamp sequence of previously executed prediction steps are concatenated according to the requirement of time continuity. For prediction segments with time overlap, a weighted average method is used for data fusion to generate a one-dimensional array structure covering both past and future time spans with a clear step structure. This structure is stored in a temporary file on the local solid-state drive and established as the basic data group for prediction steps.

[0032] S402: Collect pressure signals from furnace negative pressure measuring points, extract the peak-to-valley difference of pressure signal fluctuation per unit time, and generate negative pressure change state quantities by combining the prediction step size base data group. Micro differential pressure transmitters, installed on-site at the top and side water-cooled wall areas of the boiler furnace, are activated to continuously acquire analog signals of furnace negative pressure at an extremely high sampling rate of 100 milliseconds. The acquired negative pressure time-domain signal is converted to the frequency domain using a Fast Fourier Transform (FFT), filtering out frequencies higher than 5 Hz caused by mechanical vibrations from the induced draft fan blades. The signal is then restored to the time domain using an Inverse Fourier Transform (IFT). Within a 1-second time window, the highest and lowest pressure extremes of the filtered negative pressure signal are searched, and subtraction is performed to obtain the peak-to-valley difference. The prediction step size base data set is retrieved from a temporary file on the solid-state drive, and the average time interval of the previous step size is read. The obtained negative pressure peak-to-valley difference is divided by this average time interval to calculate the coefficient of negative pressure drastic change per unit time. This value, representing the negative pressure fluctuation gradient, is compressed to a standard range of 0 to 1 using a nonlinear Sigmoid function. The resulting dimensionless continuously changing value is directly defined as the current negative pressure change state quantity on the furnace side and updated in shared memory.

[0033] S403: Call the negative pressure change state quantity and the current prediction sequence sampling step size, calculate the ratio of the two to the pressure change set extreme value and the step size set extreme value respectively, calculate the step size adjustment coefficient based on the ratio, map the step size adjustment coefficient to the current prediction sequence sampling step size, and obtain the prediction time domain step size set. The system retrieves the negative pressure change state quantity and the current prediction sequence sampling step size from shared memory. Simultaneously, it reads the pre-set extreme values ​​for pressure change settings (e.g., a maximum allowable peak-to-valley difference of 500 Pa) and step size settings (e.g., a maximum allowed prediction step size of 100 steps) from the system configuration table. Division operations are performed to calculate the ratio of the negative pressure change state quantity to the extreme value for pressure change, and the ratio of the current prediction sequence sampling step size to the extreme value for step size. Using these two ratios, a dynamic adjustment function model is established. The product of these two ratios is added to a base adjustment bias of 0.5 to calculate the step size adjustment coefficient used to scale the prediction time scale. This coefficient is then used as a scaling factor, directly multiplied by the current prediction sequence sampling step size, and the product is rounded down to obtain a new deviation step size value. Based on this new step size, the node intervals in the future prediction time domain are proportionally re-divided, constructing a structure array containing the millisecond time offset of each specific time node relative to zero, ultimately obtaining the prediction time domain step size set.

[0034] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the boundary set of combustion regulation parameters, call the current cycle flue gas temperature prediction sequence, collect the opening set value of the induced draft fan blade adjustment actuator, compare the predicted sequence temperature value with the boundary set extreme value parameter, limit the data amplitude of the over-limit node, and generate the boundary constraint prediction sequence; The system accesses the boundary set of combustion regulation parameters within the memory-secure sandbox via the internal data bus and extracts the dynamic upper and lower limit values ​​of the damper opening and inverter frequency defined therein. Subsequently, it calls the smoke temperature prediction sequence for the next 60 nodes generated by the prediction algorithm within the current execution cycle. The system synchronously acquires the current opening setpoint of the induced draft fan blade adjustment actuator through the distributed control system data interface. A loop structure iterates through the temperature value of each node in the smoke temperature prediction sequence. During this iteration, the predicted temperature value of the current node is input into the preset temperature-opening conversion logic to calculate the theoretically required induced draft fan blade opening value to maintain the predicted temperature. This theoretical opening requirement value is compared one by one with the dynamic upper and lower limit extreme values ​​extracted from the boundary set. When the theoretical opening requirement value exceeds the upper limit of the boundary set, the theoretical opening requirement value of that node is forcibly truncated to the upper limit extreme value; conversely, if it is below the lower limit, it is truncated to the lower limit extreme value. After the restriction processing is completed, the corresponding node smoke temperature value after restriction is calculated in reverse based on the truncated opening requirement value, and the data at that position in the original prediction sequence is replaced. Finally, a boundary constraint prediction sequence whose variation range is strictly limited within the safe boundary of equipment operation is generated. Table 2 Constraint Data Table for Embodiments As shown in Table 2, this embodiment performs limitation and truncation operations on the data amplitude that exceeds the limit.

[0035] S502: For the boundary-constrained prediction sequence, call the prediction time-domain step size set to extract the corresponding step size time interval, perform time axis repositioning and amplitude allocation operations on the sequence nodes according to the step size time interval, and establish a step size reconstruction prediction sequence; For the boundary constraint prediction sequence that has already undergone amplitude limiting, the system reads the millisecond time offsets corresponding to each repartition node recorded within the prediction time-domain step set and extracts the new step time interval values ​​between adjacent nodes. Because the new step time interval is inconsistent with the fixed equidistant time intervals of the original prediction sequence, a non-uniform time resampling operation is performed. The system first constructs a new time axis vector, whose node positions are strictly generated by sequentially accumulating the extracted new step time intervals. Subsequently, the original absolute time coordinates corresponding to each data point in the boundary constraint prediction sequence are mapped onto the newly constructed time axis. In the blank intervals where the two sets of time coordinates do not overlap, the cubic Hermite interpolation polynomial algorithm is called to calculate and allocate the temperature amplitude at the new node positions based on the amplitudes of adjacent nodes in the original sequence and their first derivatives. After this operation is completed, the temperature amplitudes of all sequence nodes have been redistributed on the new time axis with dynamic interval characteristics. The system stores this one-dimensional sequence, which integrates dynamic time scales and amplitude-limited temperature values, as the step-reconstruction prediction sequence.

[0036] S503: Call the step size reconstruction prediction sequence to extract the current node temperature value of the sequence, combine it with the opening set value of the induced draft fan blade adjustment actuator, extract the previous command change amount of the associated control node, introduce the sequence set temperature extreme value, opening adjustment limit value and basic command reference amount to perform parameter constraint adjustment calculation, calculate the command reconstruction adjustment coefficient, perform mapping calculation with the command reconstruction adjustment coefficient and the current node temperature value of the sequence to obtain the smoke temperature prediction control command sequence; The temperature value of the first node at the current discrete position is extracted from the step-size reconstruction prediction sequence and used as the baseline feedforward variable. Simultaneously, using the historical record buffer of the distributed control system, the opening command setting sequence of the induced draft fan's moving blade adjustment actuator over the past five control cycles is retrieved. The command difference between adjacent cycles is calculated, and the previous command change amount (value 3) of the associated control node is extracted. Then, the allowable extreme flue gas temperature alarm setting value of the unit is read from the system engineer station configuration library as the sequence set temperature extreme value (value 900), and the maximum effective opening change rate of the induced draft fan mechanical limit calibration is read as the opening adjustment limit value (value 15). The system's built-in basic command baseline quantity (value 45) is read. Parameter constraint adjustment calculations are performed, specifically by calculating the ratio of the previous command change amount to the opening adjustment limit value, adding the logarithm of the ratio of the sequence set temperature extreme value to the basic command baseline quantity, and multiplying by a damping coefficient of 0.8 to calculate the command reconstruction adjustment coefficient (value 1.25) used to correct the final output. The adjustment coefficient is directly used as a multiplier and a scalar multiplication mapping is performed with the current node temperature value (value 800) of the sequence. The resulting product is the converted feedforward control demand, which is encapsulated into a digital command with a standard 4 to 20 mA signal specification and sent to the induced draft fan frequency converter control cabinet. This ultimately establishes a flue gas temperature prediction control command sequence to smooth flue gas temperature fluctuations. By introducing the adjustment coefficient, the drastic adjustment of the induced draft fan caused by extreme predicted temperatures is effectively constrained. Experimental data shows that when the basic command baseline is set to 45, the fluctuation amplitude of the induced draft fan motor current is reduced from 12% to 4%, the overall negative pressure stability of the unit is improved by 30% under dynamic load change conditions, and the system regulation accuracy is significantly improved.

[0037] An adaptive predictive control system for boiler flue gas temperature designed for wide-load operation includes: The flue gas temperature residual extraction module acquires the flue gas temperature values ​​output by the thermocouples at the boiler furnace flue gas temperature measuring point and the tail flue gas temperature measuring point, and simultaneously collects the predicted flue gas temperature values ​​at the corresponding time. It calls the actual flue gas temperature values ​​and the predicted flue gas temperature values ​​to perform difference calculations point by point, and arranges the residual sequence according to the sampling period. It accumulates the positive and negative values ​​in the residual sequence and performs difference calculations to obtain the flue gas temperature residual direction quantity. The load lag determination module acquires the main steam flow meter signal of the steam turbine and the flow signal of the coal feeder belt scale based on the flue gas temperature residual direction quantity and arranges them according to the sampling period. At the same time, it calls the flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point at the corresponding time, performs time shift processing on the main steam flow signal and performs difference calculation with the flue gas temperature change, performs difference statistics on the time shift result and filters the corresponding time offset position, calls the main steam flow change corresponding to the time offset and performs sign consistency judgment with the flue gas temperature residual direction quantity to obtain the load flue gas temperature lag determination result. The combustion boundary generation module obtains the opening set value of the secondary damper actuator and the frequency set value of the coal feeder frequency converter based on the load flue gas temperature hysteresis determination result, performs interval adjustment processing on the boundary corresponding to the opening set value and the frequency set value, and obtains the combustion regulation parameter boundary set. The prediction step size construction module calls the number of sampling steps of the current prediction sequence based on the load flue gas temperature hysteresis determination result, and calls the pressure signal change status of the furnace negative pressure measuring point to perform adjustment processing on the sampling step size, so as to obtain the prediction time domain step size set; The control sequence output module calls the current cycle flue gas temperature prediction sequence based on the combustion adjustment parameter boundary set and the prediction time step set, and simultaneously calls the opening set value of the induced draft fan blade adjustment actuator. It performs parameter boundary constraint processing on the prediction sequence and reconstructs the time step to obtain the flue gas temperature prediction control command sequence.

[0038] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A boiler flue gas temperature adaptive predictive control method for wide-load operation, characterized in that, Includes the following steps: S1: Collect the flue gas temperature values ​​output by the thermocouples at the boiler furnace temperature measuring point and the tail flue gas temperature measuring point, collect the predicted flue gas temperature values ​​at the corresponding time, calculate the difference between the flue gas temperature values ​​and the predicted flue gas temperature values ​​point by point to obtain the residual sequence, sort them according to the sampling period, accumulate the positive and negative values ​​of the residual sequence respectively, and calculate the difference to obtain the flue gas temperature residual direction quantity. S2: Based on the flue gas temperature residual direction quantity, the main steam flow meter signal of the steam turbine and the flow signal of the coal feeder belt scale are collected and sorted according to the sampling period. The flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point is collected. The main steam flow meter signal is time-shifted and the difference is calculated with the flue gas temperature change sequence. The time shift result is statistically analyzed and the time offset position is filtered. The main steam flow change at the time offset position is called and the sign consistency is judged with the flue gas temperature residual direction quantity to obtain the load flue gas temperature lag judgment result. S3: Based on the load flue gas temperature hysteresis determination result, collect the opening setting value of the secondary air damper actuator and the frequency setting value of the coal feeder frequency converter, and perform boundary interval adjustment on the opening setting value and the frequency setting value to obtain the combustion regulation parameter boundary set; S4: Based on the load flue gas temperature hysteresis determination result, obtain the number of sampling steps for the prediction sequence and collect the pressure signal change status of the furnace negative pressure measuring point, and adjust the number of sampling steps to obtain the prediction time domain step set; S5: Based on the combustion adjustment parameter boundary set and the prediction time step set, call the current cycle flue gas temperature prediction sequence and collect the opening setting value of the induced draft blade adjustment actuator, perform parameter boundary constraints on the flue gas temperature prediction sequence and reconstruct the time step to obtain the flue gas temperature prediction control command sequence.

2. The adaptive predictive control method for boiler flue gas temperature for wide-load operation according to claim 1, characterized in that, The residual direction quantity of flue gas temperature includes residual trend intensity, cumulative deviation amplitude, and offset direction identifier; the load flue gas temperature lag determination result includes lag time offset, response consistency identifier, and load correlation degree; the combustion adjustment parameter boundary set includes upper limit of air volume adjustment, lower limit of fuel input, and parameter change margin; the prediction time domain step size set includes short-term prediction step size, medium-term prediction step size, and dynamic adjustment coefficient; the flue gas temperature prediction control command sequence includes control output gradient, adjustment rhythm parameter, and execution constraint combination.

3. The adaptive predictive control method for boiler flue gas temperature for wide-load operation according to claim 1, characterized in that: In the calculation of the smoke temperature residual direction quantity, the cumulative positive value result and the cumulative negative value result of the residual sequence are accumulated by a sliding window with a length of 5 sampling periods, and the difference is calculated at the end of each sliding window to obtain the local smoke temperature residual direction quantity. During the time shifting process, the main steam flow meter signal is shifted sequentially in steps of two sampling periods, and the time offset position corresponding to the smallest absolute value of the difference between the shifted signal and the flue gas temperature change sequence is used as the filtering time offset position.

4. The adaptive predictive control method for boiler flue gas temperature for wide-load operation according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collect the flue gas temperature values ​​output by the thermocouples at the boiler furnace temperature measurement point and the flue gas temperature measurement point at the tail flue, synchronously collect the predicted flue gas temperature values ​​at the corresponding time, organize the node collection sequence, and obtain the multi-node flue gas temperature collection data. S102: Call the thermocouple output value and the predicted smoke temperature value within the multi-node smoke temperature acquisition data, calculate the point-by-point difference between the thermocouple output value and the predicted smoke temperature value, arrange the difference sequence according to the sampling period, and generate the sorting period residual value. S103: For the residual values ​​of the sorting period, extract the positive and negative residual values, perform cumulative calculations on the positive and negative residual values ​​respectively, calculate the difference between the positive and negative cumulative terms, and obtain the smoke temperature residual direction quantity.

5. The adaptive predictive control method for boiler flue gas temperature for wide-load operation according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the flue gas temperature residual direction quantity, collect the main steam flow meter signal of the steam turbine and the flow signal of the coal feeder belt scale, arrange the signal sequence according to the sampling period, call the flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point at the corresponding time, and establish a flow temperature synchronization sequence. S202: For the flow and temperature synchronization sequence, perform time shift processing on the main steam flow signal, calculate the point-by-point difference between the shifted main steam flow signal and the flue gas temperature change, statistically analyze the position of the minimum difference, and obtain the offset time positioning amount. S203: The offset time positioning quantity is called to extract the corresponding main steam flow change quantity. The maximum value of the main steam flow change quantity and the peak value of the flue gas temperature residual direction are introduced. The consistency operation is performed by combining the main steam flow change quantity and the flue gas temperature residual direction quantity to obtain the hysteresis consistency coefficient. The mapping is performed by combining the state judgment interval to obtain the load flue gas temperature hysteresis judgment result.

6. The boiler flue gas temperature adaptive predictive control method for wide-load operation according to claim 5, characterized in that: The specific calculation formula for performing consistency calculation by combining the main steam flow rate change and the flue gas temperature residual direction is as follows: ； The lag consistency coefficient is calculated and mapped in combination with the state determination interval to obtain the load flue gas temperature lag determination result; in, Represents the lag consistency coefficient. Represents the change in main steam flow rate and its subscript Characterizing the main steam, Represents the directional quantity of the residual smoke temperature and its subscript Characterize residuals, This represents the maximum change in main steam flow rate. This represents the peak value in the direction of the residual smoke temperature.

7. The adaptive predictive control method for boiler flue gas temperature for wide-load operation according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the load flue gas temperature hysteresis determination result, obtain the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter, merge the corresponding equipment setting record sequence, and establish a combustion setting data group; S302: For the combustion setting data set, extract the opening setting extreme value and the frequency setting extreme value, calculate the deviation of the setting extreme value from the current setting value and perform extreme value range normalization processing to generate the boundary offset reference value; S303: Call the boundary offset reference value, extract the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter, combine the upper limit of the opening operation, the reference frequency of the frequency converter and the safety margin deviation limit to perform boundary correction calculation, calculate the adjustment boundary correction coefficient, and perform mapping processing with the original opening setting boundary and the original frequency setting boundary respectively to obtain the combustion adjustment parameter boundary set.

8. The adaptive predictive control method for boiler flue gas temperature for wide-load operation according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the load and temperature hysteresis determination result, obtain the current prediction sequence sampling step size, extract the corresponding sampling period and the previous prediction step size sequence, and establish a prediction step size basic data group. S402: Collect the pressure signal from the negative pressure measuring point in the furnace, extract the peak-valley difference of the pressure signal fluctuation per unit time, and generate the negative pressure change state quantity by combining it with the prediction step size basic data group. S403: Call the negative pressure change state quantity and the current prediction sequence sampling step size, calculate the ratio of the two to the pressure change set extreme value and the step size set extreme value respectively, obtain the step size adjustment coefficient based on the ratio, map the step size adjustment coefficient to the current prediction sequence sampling step size, and obtain the prediction time domain step size set.

9. The adaptive predictive control method for boiler flue gas temperature for wide-load operation according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the combustion adjustment parameter boundary set, call the current cycle flue gas temperature prediction sequence, collect the opening set value of the induced draft fan blade adjustment actuator, compare the predicted sequence temperature value with the boundary set extreme value parameter, limit the data amplitude of the over-limit node, and generate the boundary constraint prediction sequence; S502: For the boundary constraint prediction sequence, call the prediction time domain step size set to extract the corresponding step size time interval, and perform time axis repositioning and amplitude allocation operations on the sequence nodes according to the step size time interval to establish a step size reconstruction prediction sequence; S503: The step-size reconstruction prediction sequence is called to extract the current node temperature value of the sequence. Combined with the opening setting value of the induced draft fan blade adjustment actuator, the previous command change of the associated control node is extracted. The extreme value of the sequence set temperature, the limit value of the opening adjustment and the basic command reference value are introduced to perform parameter constraint adjustment calculation. The command reconstruction adjustment coefficient is calculated. The command reconstruction adjustment coefficient and the current node temperature value of the sequence are mapped to obtain the smoke temperature prediction control command sequence.

10. A boiler flue gas temperature adaptive predictive control system for wide-load operation, characterized in that, The system is used to implement the boiler flue gas temperature adaptive predictive control method for wide-load operation as described in any one of claims 1-9, and the system includes: The flue gas temperature residual extraction module acquires the flue gas temperature values ​​output by the thermocouples at the boiler furnace flue gas temperature measuring point and the tail flue gas temperature measuring point, and simultaneously collects the predicted flue gas temperature values ​​at the corresponding time. It calls the actual flue gas temperature values ​​and the predicted flue gas temperature values ​​to perform difference calculations point by point, and arranges the residual sequence according to the sampling period. It accumulates the positive and negative values ​​in the residual sequence and performs difference calculations to obtain the flue gas temperature residual direction quantity. The load lag determination module acquires the turbine main steam flow meter signal and the coal feeder belt scale flow signal based on the flue gas temperature residual direction quantity and arranges them according to the sampling period. At the same time, it calls the flue gas temperature change sequence of the thermocouple at the boiler furnace flue gas temperature measuring point at the corresponding time, performs time shift processing on the main steam flow signal and performs difference calculation with the flue gas temperature change, performs difference statistics on the time shift result and filters the corresponding time offset position, calls the main steam flow change corresponding to the time offset and performs sign consistency judgment with the flue gas temperature residual direction quantity to obtain the load flue gas temperature lag determination result. The combustion boundary generation module obtains the opening setting value of the secondary damper actuator and the frequency setting value of the coal feeder frequency converter based on the load flue gas temperature hysteresis determination result, performs interval adjustment processing on the boundary corresponding to the opening setting value and the frequency setting value, and obtains the combustion adjustment parameter boundary set. The prediction step size construction module calls the current prediction sequence sampling step size number based on the load flue gas temperature hysteresis determination result, and calls the pressure signal change status of the furnace negative pressure measuring point to perform adjustment processing on the sampling step size, so as to obtain the prediction time domain step size set. The control sequence output module calls the current cycle flue gas temperature prediction sequence based on the combustion adjustment parameter boundary set and the prediction time step set, and simultaneously calls the opening setting value of the induced draft fan blade adjustment actuator. It performs parameter boundary constraint processing on the prediction sequence and reconstructs the time step to obtain the flue gas temperature prediction control command sequence.

Images