Coal-fired unit pulverized coal distribution adjustment control method based on boiler wall temperature distribution field

Abstract

The invention provides a coal-fired unit pulverized coal distribution adjustment control method based on a boiler wall temperature distribution field. According to the method, automatic adjustment of pulverized coal distribution of the coal-fired unit under the full-load working condition can be achieved, the uniformity of combustion intensity distribution is effectively improved, and the safety, the environmental protection property and the economical efficiency of unit operation are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of coal-fired unit combustion control, and in particular to a coal-fired unit pulverized coal distribution regulation control method based on a boiler wall temperature distribution field. BACKGROUND

[0002] As an important part of the power system, coal-fired units are widely used in energy supply and peak shaving operation. With the increasing demand for deep peak shaving, the uniformity of combustion intensity distribution has become a key factor to ensure the safety, environmental protection and economic operation of the boiler. In related technologies, a technical system for coal powder conveying and combustion control is constructed through the coordinated operation of the medium-speed mill direct-fired pulverizing system and the pulverized coal distributor. Specifically, the system covers the whole process from medium-speed mill pulverizing, pulverized coal distribution regulation, burner powder supply to furnace combustion, including key links such as air-powder ratio control, combustion stability regulation, wall temperature monitoring, etc. Among them, the uniformity of pulverized coal distribution directly affects the stability of the furnace combustion field, and is one of the core technologies to achieve efficient combustion and low emission operation.

[0003] However, in the existing pulverized coal distribution regulation method, direct artificial sampling or air-powder online measurement technology is used as the feedback signal, which has significant limitations. Specifically, although artificial sampling has high accuracy, it takes a long time for a single operation, and the data update cycle is counted in hours, which is difficult to meet the real-time control demand; while the air-powder online measurement technology has real-time performance, but its measurement result is generally distorted, leading to misjudgment of the control algorithm, and the system is difficult to converge, so the automatic control basically fails. Based on this, the traditional control means cannot effectively maintain the uniformity of the combustion intensity distribution under the deep peak shaving condition with frequent load fluctuations, thereby causing a chain of problems such as local over-temperature of the water-cooled wall, excessive NOx emission and increased coal consumption, which seriously restricts the safety, environmental protection and economy of the coal-fired unit operation. SUMMARY

[0004] The present application aims to at least partially solve one of the technical problems in the related art.

[0005] To this end, a first object of the present application is to propose a coal-fired unit pulverized coal distribution regulation control method based on a boiler wall temperature distribution field.

[0006] A second object of the present application is to propose a coal-fired unit pulverized coal distribution regulation control device based on a boiler wall temperature distribution field.

[0007] To achieve the above-mentioned objects, a coal-fired unit pulverized coal distribution regulation control method based on a boiler wall temperature distribution field is proposed in the first aspect of the present application, comprising:

[0008] S1, real-time temperature data of each temperature measuring point on the water wall of the boiler furnace is acquired, the temperature measuring points are divided into several groups according to the positions of the measuring points, the temperature deviation rate of each measuring point in each group is calculated, and a temperature deviation matrix T is constructed;

[0009] S2, the running load rate of each medium-speed coal mill is acquired, a diagonal matrix B is formed, the opening matrix A of the adjusting baffle of each coal powder distributor is acquired, and the matrix AB is calculated to reflect the adjusting baffle opening state under the current running condition;

[0010] S3, based on the trained support vector regression model, the temperature deviation matrix T' is predicted according to the matrix AB, and the error function E is calculated by combining the measured temperature deviation matrix T, as an evaluation index of the adjusting control;

[0011] S4, according to the change of the error function E, a double-stage simulated annealing algorithm is used for adjusting decision, wherein the first-stage control judges whether the adjusting baffle needs to be adjusted, the second-stage control iteratively searches the global optimal solution of the adjusting baffle opening when the adjusting baffle needs to be adjusted, and the adjusting instruction is output through the DCS control system to realize automatic control.

[0012] In an embodiment of the present application, the S1 comprises:

[0013] S11, the temperature measuring points are divided into symmetrically combustion intensity related region groups according to the combustion region distribution of the boiler furnace, each group contains a measuring point, so as to enhance the local sensitivity to the combustion unevenness;

[0014] S12, when the temperature deviation rate of each measuring point in each group is calculated, the sliding window average method is used to calculate the temperature average value in the group, so as to reduce the influence of the instantaneous temperature fluctuation on the deviation rate calculation.

[0015] In an embodiment of the present application, the S2 comprises:

[0016] S21, the calculation of the running load rate λj uses the ratio of the current coal mill output to the maximum output, and the historical data of the DCS system is used for dynamic correction, so as to eliminate the interference of the coal quality fluctuation on the load rate calculation;

[0017] S22, the opening of the adjusting baffle corresponding to the non-running medium-speed coal mill is set to 0 in the matrix AB by multiplying the matrix, so that the opening of the adjusting baffle corresponding to the non-running medium-speed coal mill has no influence on the overall control strategy.

[0018] In an embodiment of the present application, the S3 comprises:

[0019] S31, the support vector regression model uses the radial basis function RBF as the kernel function in the training process, and the optimal penalty coefficient C and kernel width γ are selected through cross-validation;

[0020] S32, the calculation of the error function E adopts a Frobenius norm to quantify the overall deviation between the predicted value and the measured value.

[0021] In one embodiment of the present application, further comprising:

[0022] S5, when the system detects that the coal quality of the coal-fired boiler changes significantly, adjusting the input feature weight of the support vector regression model according to the coal quality parameter, and re-performing local model updating to improve the prediction accuracy of the model under the coal quality fluctuation working condition.

[0023] To achieve the above purpose, the second aspect of the present application provides a coal pulverizer distribution adjustment control device based on the boiler wall temperature distribution field, comprising: a temperature data acquisition and grouping module, used for acquiring real-time temperature data of each temperature measuring point on the boiler furnace water wall, and dividing the temperature measuring points into several groups according to the positions of the measuring points, calculating the temperature deviation rate of each measuring point in each group, and constructing a temperature deviation matrix T;

[0024] An operating parameter acquisition and matrix construction module is used for acquiring the operating load rate of each medium-speed coal mill, forming a diagonal matrix B, and acquiring an opening matrix A of each coal pulverizer adjustment baffle, and calculating the matrix AB to reflect the adjustment baffle opening state under the current operating condition.

[0025] A model prediction and error evaluation module is used for predicting the temperature deviation matrix T' based on the trained support vector regression model according to the matrix AB, and calculating an error function E in combination with the measured temperature deviation matrix T as an evaluation index of adjustment control.

[0026] An adjustment decision and control output module is used for making adjustment decisions by using a two-stage simulated annealing algorithm according to the change of the error function E, wherein the first-stage control judges whether the adjustment baffle needs to be adjusted, the second-stage control iteratively searches for a global optimal solution of the adjustment baffle opening when adjustment is needed, and outputs an adjustment instruction through a DCS control system to realize automatic control.

[0027] The method and device of the embodiment of the present application realize automatic adjustment of coal pulverizer distribution under full load working condition, improve the uniformity of combustion intensity distribution, and enhance the safety, environmental protection and economy of the unit operation.

[0028] Additional aspects and advantages of the present application will be made apparent from the following description of embodiments, which proceeds with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0029] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description, taken in conjunction with the accompanying drawings, in which:

[0030] Figure 1 is a flow chart of a coal pulverized coal distribution adjustment control method based on a boiler wall temperature distribution field of an embodiment of the present application;

[0031] Figure 2 is a logic diagram of a coal pulverized coal distribution adjustment control system based on a boiler wall temperature distribution field of an embodiment of the present application;

[0032] Figure 3 is a structural schematic diagram of a coal pulverized coal distribution adjustment control device based on a boiler wall temperature distribution field of an embodiment of the present application. DETAILED DESCRIPTION

[0033] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0034] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0035] A coal pulverized coal distribution adjustment control method and device based on a boiler wall temperature distribution field will be described below with reference to the accompanying drawings according to an embodiment of the present application.

[0036] Embodiment 1

[0037] Figure 1 is a flow chart of a coal pulverized coal distribution adjustment control method based on a boiler wall temperature distribution field of an embodiment of the present application. As shown in Figure 1 , the method comprises the following steps:

[0038] S1, real-time temperature data of each temperature measuring point on the boiler furnace water cooling wall is obtained, and the temperature measuring points are divided into several groups according to the positions of the measuring points, the temperature deviation rate of each measuring point in each group is calculated, and a temperature deviation matrix T is constructed.

[0039] Specifically, this step involves real-time data collection of boiler furnace water wall temperature measurement points and construction of temperature deviation matrix T, which is the key data input link for realizing automatic control of coal powder distribution in coal-fired units. In some implementations, the system obtains real-time temperature data of each measurement point at a fixed sampling frequency (e.g., 1-5 seconds) through multiple groups of wall temperature sensors (such as K-type thermocouples or optical fiber temperature measurement probes) arranged on the boiler furnace water wall. These measurement points are usually spatially divided according to the geometric structure of the furnace and the distribution of burners to ensure that they can fully reflect the spatial distribution characteristics of the combustion intensity in the furnace.

[0040] Specifically, the system divides the temperature measurement points into several groups, each containing a measurement points, a total of b groups, forming an a x b measurement point layout. At the parameter index level, the calculation formula for the temperature deviation rate in each group is: (T_i-T_avg) / T_avg, where T_i is the real-time temperature of the i-th measurement point, and T_avg is the average temperature of all measurement points in the group. This deviation rate reflects the deviation of each measurement point relative to the average temperature in the group, and its value range is usually between ±0.1 and ±0.3. The smaller the deviation rate, the more uniform the temperature distribution in the group.

[0041] Further, the system integrates the temperature deviation rates of all measurement points by group to construct an a-row and b-column temperature deviation matrix T. This matrix not only reflects the temperature deviation of each measurement point, but also facilitates subsequent mathematical operations and model training with the adjustment damper opening matrix A and the operating load matrix B through the matrix form. At the application scenario level, this step is applicable to the operation adjustment of coal-fired units under different load conditions, especially under deep peak shaving conditions. Through the construction of real-time temperature deviation matrix T, accurate feedback signals can be provided for the simulated annealing algorithm, thereby optimizing the coal powder distribution strategy.

[0042] The technical value of this step lies in replacing the instability of traditional wind powder online measurement with mature and high-precision wall temperature measurement means, providing reliable and real-time combustion state feedback for the automatic control system, thereby improving the convergence and robustness of the system control and ensuring the safety and economy of the boiler operation.

[0043] Further, S1 comprises:

[0044] S11, the temperature measurement points are divided into symmetrically related to the combustion intensity region groups according to the combustion region distribution of the boiler furnace, each group containing a measurement points, to enhance the local sensitivity to combustion unevenness.

[0045] Specifically, one of the key steps in constructing the wall temperature distribution deviation matrix T in the present application is to divide the temperature measurement points according to the combustion area distribution of the boiler furnace into symmetrically related combustion intensity area groups. The technical implementation principle is based on the thermodynamic characteristics of the boiler combustion area and the spatial correlation of the wall temperature response. In some implementations, this division method divides the temperature measurement points on the water wall into several symmetrically related area groups according to the arrangement of the furnace burners, the jet angle of the primary air powder flow, the heat load distribution of the combustion area, etc. Each group contains a measurement point (a is a positive integer, usually the value range is 3-10, and the specific value is determined according to the furnace size and the density of the measurement points), so as to enhance the local sensitivity to combustion unevenness.

[0046] This step first needs to model the combustion area of the boiler furnace in three dimensions, determine the heat load center of each area in combination with the arrangement position of the burners (such as the front wall, rear wall, and side wall arrangement) and the flow trajectory of the primary air powder flow. Then, according to the installation position of the wall temperature sensor, it is mapped to the corresponding combustion area and grouped according to the symmetry principle (such as left-right symmetry, up-down symmetry). Each group of measurement points should cover the area affected by the same burner or adjacent burners as much as possible to ensure that the measurement points in the group have similar thermal response characteristics.

[0047] The number of divided area groups b is usually consistent with the number of burners or the number of coal powder distributor outlets, and is generally 4-8 groups. The number of measurement points in each group a should meet the statistical significance requirement, and is usually not less than 3 measurement points to ensure the representativeness of the average temperature in the group. In addition, the calculation formula of the temperature deviation rate is: (Ti-T_avg) / T_avg, where Ti is the temperature of a single measurement point, and T_avg is the average temperature in the group. This index is used to quantify the local combustion unevenness, and the larger the absolute value, the more intense the combustion intensity fluctuation in the area.

[0048] This step is suitable for the coal powder distribution automatic adjustment system of the coal-fired unit under different load conditions. By dividing the wall temperature measurement points into symmetrically related area groups, the system can more accurately identify the asymmetric changes of the combustion intensity in the furnace, thereby providing more refined feedback signals for the subsequent optimization of the regulation baffle opening. Especially in the deep peak shaving condition, the coal powder distribution deviation between the burners is easy to cause local thermal stress concentration, and this division method helps to timely find and correct such deviation, thereby improving the combustion stability and safety.

[0049] Through the structured area division, the local perception ability of the system to combustion unevenness is enhanced, the representation accuracy of the temperature deviation matrix T is improved, high-quality input data is provided for the subsequent coal powder distribution optimization based on the support vector regression (SVR) model, thereby improving the response speed and regulation accuracy of the automatic control system, and realizing the dynamic and refined control of the combustion state of the furnace.

[0050] S12, when calculating the temperature deviation rate of each measuring point in each group, the sliding window average method is used to calculate the temperature average value in the group to reduce the influence of instantaneous temperature fluctuation on the deviation rate calculation.

[0051] Specifically, in the automatic control method of the present application, when calculating the temperature deviation rate of each measuring point in each group, the sliding window average method is used to preprocess the temperature in the group to reduce the influence of instantaneous temperature fluctuation on the deviation rate calculation. The technical implementation of this step is based on the smoothing principle of time series data. By setting a reasonable sliding window length, the temperature measuring points in each group on the water-cooled wall of the boiler furnace are dynamically averaged, thereby improving the stability and representativeness of the temperature deviation rate calculation.

[0052] Further, the length w of the sliding window needs to be set according to the dynamic response characteristics of the boiler operating condition. Generally, the length should be greater than the minimum response time of the boiler wall temperature change (for example, 30 seconds to 2 minutes) to filter out the instantaneous noise caused by combustion disturbance, local thermal load fluctuation, etc. At the same time, in order to ensure the real-time nature of the data, the window should not be too long, and generally controlled within 5-10 sampling periods, with a sampling frequency of 1-2 seconds / time, which meets the industrial temperature measurement standards such as IEC 60601 or GB / T 30470.

[0053] In practical application, this step realizes dynamic updating of the sliding window through online processing of the raw data collected by the wall temperature sensor by the DCS control system, combined with the time stamp information. In the process of pulverized coal distribution adjustment, the temperature deviation rate, as a key input feature, is used to construct the temperature deviation matrix T, and then participates in the training and prediction of the SVR model, improving the robustness and convergence of the control strategy.

[0054] Through the sliding window average method, the interference of instantaneous temperature fluctuation on the deviation rate calculation is effectively suppressed, the credibility and stability of the deviation rate are improved, and more accurate input basis is provided for the subsequent optimization decision of adjusting the baffle opening, thereby enhancing the response accuracy and adjustment effect of the entire automatic control system.

[0055] S2, obtain the running load rate of each medium-speed coal mill, form a diagonal matrix B, and obtain the opening matrix A of each pulverized coal distributor adjusting baffle, calculate the matrix AB to reflect the adjusting baffle opening state under the current operating condition.

[0056] Specifically, this step involves matrix modeling and calculation of the operating load rate of the medium-speed coal mill and the state of the regulating baffle opening in the coal pulverizing distribution regulation automatic control system of the coal-fired unit, and is a basic link for building the system control model. In some implementations, the system first acquires the operating load rate λj(j = 1, 2, …, n) of each medium-speed coal mill in real time through the DCS control system, where λj is defined as the ratio of the current operating output to the maximum operating output, and the value range is [0, 1]. If a coal mill is out of operation, λj = 0, indicating that the coal mill does not participate in the coal pulverizing distribution under the current operating condition. These λj values are constructed into an n-order diagonal matrix B = diag(λ1, λ2, …, λn), which is used to represent the operating state of each coal mill under the current load and coal quality conditions, and has good physical significance and mathematical expression ability.

[0057] At the same time, the system acquires the opening value φij(i = 1, 2, …, m; j = 1, 2, …, n) of the regulating baffle of each coal pulverizing distributor, where m is the number of regulating baffles of each coal pulverizing distributor, and n is the total number of coal mills. φij is constructed into an m × n opening matrix A, where each column corresponds to the regulating baffle opening configuration of a coal mill. Further, AB is calculated by matrix multiplication to obtain an m × n regulating baffle opening state matrix, which can exclude the influence of the out-of-operation coal mill and only retain the regulating baffle opening information corresponding to the current operating coal mill, so as to more accurately reflect the coal pulverizing distribution state under the actual operating condition.

[0058] This step plays a key role in the system, and the calculation result AB will be used as the input feature of the subsequent support vector regression (SVR) model for predicting the boiler furnace wall temperature distribution deviation matrix T. By coupling modeling of the operating load rate and the regulating baffle opening, the system can realize dynamic identification and optimization control of the coal pulverizing distribution state under different load and coal quality conditions, provide reliable initial state input for the subsequent simulated annealing algorithm optimization, and thus improve the real-time performance and robustness of the control system.

[0059] Further, S2 includes:

[0060] S21, the calculation of the operating load rate λj uses the ratio of the current coal mill output to the maximum output, and is dynamically corrected through the historical data of the DCS system to eliminate the interference of coal quality fluctuations on the calculation of the load rate.

[0061] Specifically, in the present application, the calculation of the operation load rate λj adopts the ratio of the current coal mill output to the maximum output, and is dynamically corrected through the historical data of the DCS system to eliminate the interference of coal quality fluctuations on the load rate calculation. This step is one of the basic parameters for building the automatic control system of the pulverized coal distribution regulation, and its technical implementation principle is based on the quantitative description of the running state of the coal mill, and dynamically modified by combining historical operation data for data-driven, thereby improving the robustness and control accuracy of the system under different coal quality conditions.

[0062] Optionally, the dynamic correction process can be weightedly corrected based on coal quality parameters (such as received net low calorific value Qnet,ar, Hardgrove grindability index HGI, and coal fineness R90, etc.). For example, the system can set a coal quality fluctuation threshold ΔQnet,ar (such as ±500kJ / kg), and when the current coal quality deviates from the historical reference value by more than the threshold, the calculation coefficient of λj is automatically adjusted to reflect the change of the actual coal conveying capacity. Further, the correction process can combine time series analysis methods (such as moving average, exponential smoothing, etc.) to ensure the continuity and stability of the load rate calculation.

[0063] This step plays a key role in the entire control system, and its calculation result is used as the diagonal element of matrix B, which together with the regulation damper opening matrix A forms the input feature for the training and prediction of the support vector regression (SVR) model. Through dynamic correction, the system can more accurately identify the actual running state of the coal mill under frequent coal quality changes, thereby improving the adaptability and control effect of the pulverized coal distribution regulation strategy, and ensuring the uniformity and safety of the boiler combustion.

[0064] S22, the regulation damper opening corresponding to the non-operating medium-speed coal mill in the matrix AB is obtained by setting λj to 0 and multiplying the matrix, so that the regulation damper opening corresponding to the coal mill has no effect on the overall control strategy.

[0065] Specifically, in the automatic control method of the present application, the step "the regulation damper opening corresponding to the non-operating medium-speed coal mill in the matrix AB is obtained by setting λj to 0 and multiplying the matrix, so that the regulation damper opening corresponding to the coal mill has no effect on the overall control strategy", the technical implementation principle is based on the coupling modeling of matrix operation and running state, aiming to realize the automatic shielding of the regulation damper opening corresponding to the stopped coal mill, thereby improving the robustness and accuracy of the control system.

[0066] In some implementations, the matrix A is an m-row n-column adjustment damper opening matrix, where m is the number of adjustment dampers of a single pulverized coal distributor, and n is the total number of medium-speed coal mills. The matrix B is an n-order diagonal matrix, and the diagonal element λj represents the operation load rate of the jth medium-speed coal mill, with a value range of [0, 1]. When a medium-speed coal mill is shut down, the corresponding λj is set to 0, and the diagonal element at this position in the matrix B is 0, and the rest are 1 or operation load rate values between 0 and 1. By matrix multiplication AB, the adjustment damper opening φij corresponding to the shut-down coal mill is multiplied by 0, so as to be excluded in the overall control strategy, avoiding its interference with the subsequent control model.

[0067] This step is suitable for coal-fired units under deep peak regulation and frequent load fluctuations, especially when some coal mills are shut down or switched, which can effectively isolate invalid control variables and improve the identification accuracy of effective adjustment parameters. Through the matrix shielding mechanism, the control system can focus on the adjustment strategy of the current running coal mill, avoiding the introduction of noise or misleading data by the shut-down equipment. This step realizes the automatic elimination of the adjustment parameters of the shut-down coal mill, improves the quality of the input data of the control model, and thus enhances the prediction accuracy of the SVR model and the convergence of the control strategy. In addition, this method does not require additional sensors or complex logical judgments, and can realize state isolation through matrix operation, with good engineering practicability and computational efficiency.

[0068] S3, based on the trained support vector regression model, predicts the temperature deviation matrix T' according to the matrix AB, and calculates the error function E in combination with the measured temperature deviation matrix T as an evaluation index of adjustment control.

[0069] Specifically, this step is based on a trained support vector regression (SVR) model, which predicts the temperature deviation matrix T' of the boiler water wall by inputting the current adjustment damper opening matrix A and the operation load rate matrix B, and compares it with the actually measured temperature deviation matrix T to calculate the error function E as an evaluation index of adjustment control. This step is the core feedback mechanism of the entire automatic control system, and its technical implementation principle is based on the method of combining machine learning and process control, aiming to evaluate the influence of the current pulverized coal distribution strategy on the uniformity of the furnace combustion through the comparison between real-time data and model prediction.

[0070] Firstly, the system collects the current opening values of each regulating damper through the DCS control system, constructs matrix A, which has the dimension of m rows and n columns, where m is the number of regulating dampers of a single coal distributor, and n is the total number of running medium-speed coal mills. At the same time, the system obtains the running load rate of each medium-speed coal mill, and constructs diagonal matrix B = diag(λ1,..., λn), where λj∈[0, 1]. Multiply matrix A and B to get matrix AB, which excludes the influence of the shutdown coal mill on the state of the regulating damper, thus more accurately reflecting the coal distribution state under the current running condition.

[0071] Subsequently, the system inputs the AB matrix into the trained support vector regression model to predict the temperature deviation matrix T' of the boiler water-cooled wall. The model is trained based on the historical data set [A, B, T], where T is the temperature deviation rate matrix calculated after dividing the multiple temperature measuring points by groups, and its element is defined as (measuring point temperature / group average temperature-1), with a row and b column structure. The prediction result T' and the real-time collected T are calculated by Frobenius norm to get the error function, which quantifies the deviation between the model prediction and the actual temperature distribution, and is the key basis for determining whether to adjust the damper opening in the adjustment control strategy.

[0072] This step is mainly used to evaluate the influence of coal distribution adjustment on the uniformity of the furnace combustion in real time. When the system detects that the running load rate changes more than the set threshold, it will trigger the control process to determine whether to optimize and adjust the regulating damper according to the size of the error function E. This mechanism is particularly important in deep peak regulation conditions, which can quickly respond to load changes and avoid problems such as water-cooled wall thermal stress concentration or coking caused by uneven combustion. By introducing the SVR model and error function E, this step realizes the nonlinear mapping from the state of the regulating damper to the temperature distribution of the furnace, providing a scientific and quantifiable evaluation basis for automatic control. Compared with traditional methods that rely on artificial experience or immature online wind and coal measurement technology, this method has higher prediction accuracy and control stability, which helps to improve the safety, economy and environmental performance of coal-fired units under complex operating conditions.

[0073] Further, S3 comprises:

[0074] S31, the support vector regression model uses radial basis function RBF as the kernel function in the training process, and selects the optimal penalty coefficient C and kernel width γ through cross-validation.

[0075] Specifically, in the present application, the Support Vector Regression (SVR) model adopts Radial Basis Function (RBF) as the kernel function in the training process, and selects the optimal penalty coefficient C and kernel width γ through cross-validation. This step is the core link of building the coal distribution adjustment control model, and its technical implementation is based on the structural risk minimization principle in the statistical learning theory, aiming to establish the nonlinear mapping relationship of the boiler furnace wall temperature distribution deviation matrix T under the joint action of the medium-speed coal mill operating state matrix B and the adjustment damper opening matrix A.

[0076] The SVR model maps the input feature space to a high-dimensional feature space by introducing the RBF kernel function, thereby improving the model's fitting ability for complex nonlinear relationships.

[0077] This step is used to build a prediction model for coal distribution adjustment, providing a mathematical basis for the subsequent simulated annealing control strategy. The SVR model obtained through offline training can quickly predict the wall temperature deviation matrix T under different B value conditions in online control, thereby evaluating the control effect of the adjustment damper opening A and achieving closed-loop feedback control. The SVR model using the RBF kernel function can effectively capture the nonlinear coupling relationship between coal distribution and wall temperature distribution, while cross-validation ensures the robustness and generalization ability of the model under different operating conditions. This step provides high-precision prediction capability for the system and is the key support for achieving automatic adjustment of coal distribution and improving the uniformity and operating economy of the boiler combustion.

[0078] S32, the error function E is calculated using the Frobenius norm to quantify the overall deviation between the predicted value and the measured value.

[0079] The Frobenius norm is a measurement method suitable for matrix space, defined as the square root of the sum of squares of all elements of the matrix. Using the Frobenius norm as the error function can effectively capture the overall trend of temperature deviation in spatial distribution, avoiding distortion of the control strategy caused by local outliers. At the same time, this method is highly matched with the output form of the SVR model, facilitating the construction of a unified optimization objective function. By converting the multi-dimensional temperature deviation into a scalar error, the system can achieve global optimization of the adjustment damper opening under complex conditions, thereby improving the uniformity of coal distribution, improving the boiler combustion efficiency and operating stability, and having significant engineering application value.

[0080] S4, according to the change of the error function E, a two-stage simulated annealing algorithm is used for adjustment decision, where the first-stage control judges whether the adjustment damper needs to be adjusted, and the second-stage control iteratively searches for the global optimal solution of the adjustment damper opening when adjustment is needed, and outputs the adjustment command through the DCS control system to realize automatic control.

[0081] Specifically, this step is based on real-time monitoring data of the boiler wall temperature distribution field, and adopts a two-stage simulated annealing algorithm for adjustment decision-making, aiming to realize automatic control of the pulverized coal distribution adjustment of the coal-fired unit. The technical implementation principle is based on comparison of the deviation between the actual wall temperature predicted by a support vector regression (SVR) model, and the error function is constructed to quantify the difference between the current state and the target state of the system. The first-stage control is used to determine whether the adjustment baffle needs to be adjusted, and the second-stage control is used to determine the global optimal solution of the adjustment baffle opening degree through iterative search when adjustment is needed.

[0082] In specific operation, the first-stage control takes a sampling interval t1 as the period, calculates the Frobenius norm difference △B of the medium-speed coal mill operation load rate matrix B of the current and the last period. When △B exceeds the set upper threshold △Bu, the system reverses the suggested value of the adjustment baffle opening degree matrix A through the SVR model for manual decision-making. When △B is in the middle interval (△Bd<△B<△Bu), the system enters the second-stage control process to perform iterative optimization with a finer sampling interval t2. In the second-stage control, the system randomly generates a new solution A1 in the neighborhood of the current solution A0, and predicts the corresponding wall temperature deviation matrix T”1 through the SVR model, and calculates the error E1. If E1 is less than E0, the new solution is directly accepted; if E1 is greater than E0, whether the new solution is retained is determined according to the probability acceptance criterion of the simulated annealing algorithm, so as to avoid falling into a local optimum.

[0083] In this step, the temperature parameter T, the cooling rate a, and the neighborhood disturbance range Δφ of the simulated annealing algorithm all need to be reasonably set according to the dynamic response characteristics of the system, usually T is initially set to 100-200, a is taken as 0.95-0.98, and Δφ is controlled between ±5% and ±10% to balance the search efficiency and stability. Finally, the DCS control system outputs the optimized adjustment baffle opening degree command to the remote execution mechanism to realize closed-loop automatic control of the pulverized coal distribution.

[0084] This step plays a key decision-making and optimization role in the whole system, especially when the load fluctuates or the coal quality changes, it can quickly respond and adjust the pulverized coal distribution strategy, thereby effectively improving the uniformity of the combustion in the furnace, reducing the local thermal stress, and improving the safety and economy of the boiler operation. Its technical value lies in combining the mature wall temperature monitoring method with the robust simulated annealing algorithm, and solving the problems of hysteresis and instability in the automatic control of the traditional pulverized coal distribution system.

[0085] The pulverized coal distribution adjustment control method for a coal-fired unit based on a boiler wall temperature distribution field can effectively improve the uniformity of the wind-pulverized coal distribution under different load and coal quality conditions by monitoring the boiler wall temperature distribution in real time and optimizing the pulverized coal distribution combined with the simulated annealing algorithm, and can ensure the safety, economy and environmental protection of the unit operation.

[0086] Further comprising: S5, when the system detects that the coal quality changes significantly, adjusting the input feature weight of the support vector regression model according to the coal quality parameter, and re-performing local model updating to improve the prediction accuracy of the model under the coal quality fluctuation condition.

[0087] Specifically, when the system detects that the coal quality changes significantly, the input feature weight of the support vector regression (SVR) model needs to be adjusted according to the coal quality parameter, and local model updating is re-performed to improve the prediction accuracy of the model under the coal quality fluctuation condition. This step is based on the nonlinear mapping relationship between the boiler wall temperature distribution field and the coal powder distribution adjustment, and by dynamically optimizing the input feature weight of the SVR model, the sensitivity and adaptability of the model to coal quality changes are enhanced.

[0088] In terms of technical implementation, the system first collects the coal quality parameters (such as volatile matter, ash content, calorific value, moisture content, etc.) under the current working condition through the DCS control system, and fuses them with the existing input features (such as the adjustment damper opening matrix A and the coal mill load rate matrix B). Considering the nonlinear and lagging effect of coal quality changes on combustion characteristics, the system adopts a weighted feature input method, introducing the coal quality parameters as additional features into the input space of the SVR model. Specifically, the system dynamically adjusts the weight coefficient of the coal quality parameter in the feature vector according to the change amplitude (such as a change in calorific value of ±5% or more, a change in ash content of ±2% or more) to reflect its significant impact on the wall temperature distribution field.

[0089] The system sets the judgment threshold of coal quality change, such as calorific value change rate ΔQ≥5%, ash content change rate ΔA≥2%, volatile matter change rate ΔV≥3%, etc., as the basis for triggering model weight adjustment. At the same time, the kernel function type (such as RBF kernel) of the SVR model, the penalty coefficient C, the kernel width γ, etc. Hyperparameters need to be cross-validated and optimized according to the historical data set [A, B, T] to ensure the generalization ability of the model under coal quality fluctuation. Local model updating adopts an incremental learning strategy, which only fine-tunes the affected feature weights, rather than re-trains the entire model, thereby improving computational efficiency and maintaining model stability.

[0090] This step is suitable for automatic control and adjustment of coal-fired units when coal quality fluctuates frequently (such as switching of coal sources and blending of different coal types). By monitoring the changes in coal quality parameters in real time, the system can quickly respond and optimize the coal powder distribution strategy to ensure uniform distribution of combustion intensity in the furnace, avoid local overheating or coking, and improve the safety and economy of boiler operation. This step effectively improves the prediction accuracy and robustness of the SVR model under coal quality fluctuation conditions, providing more reliable prediction basis for the subsequent simulated annealing algorithm, thereby enhancing the adaptability and adjustment efficiency of the entire automatic control system.

[0091] The coal powder distribution regulation and control method for coal-fired power units based on boiler wall temperature distribution field in this invention improves the adaptability and prediction accuracy of the coal powder distribution optimization strategy by dynamically adjusting the input feature weights of the support vector regression model and updating the local model under coal quality fluctuations. This enhances the system's adaptive regulation capability under complex coal quality conditions and ensures the stability and efficiency of the combustion process.

[0092] Example 2

[0093] To achieve the above embodiments, Figure 2 This is a schematic diagram of the structure of a coal-fired power unit pulverized coal distribution and regulation control system based on the boiler wall temperature distribution field, according to an embodiment of the present invention. Figure 2 As shown, the process includes: coal is ground into pulverized coal by a medium-speed coal mill (1), and the pulverized coal airflow is formed by the hot primary air and enters the pulverized coal distributor (2). Under the guidance of the regulating baffle inside the pulverized coal distributor (2), the pulverized coal airflow is evenly divided into 4 to 8 streams (depending on the number of outlets of the pulverized coal distributor (2)), and enters the downstream burner (3) through the corresponding pulverized coal feeding pipes, and finally completes combustion in the furnace of the boiler (4). Several wall temperature sensors (5) are evenly arranged on the water-cooled wall of the boiler (4) furnace to monitor the water-cooled wall temperature. The DCS control system (6) is connected to the wall temperature sensor (5) through a signal transmission cable to realize the input of the water-cooled wall temperature signal, and connected to the remote actuator of the regulating baffle of the pulverized coal distributor (2) to realize the output and feedback measurement of the regulating baffle opening signal. Its automatic control scheme is as follows:

[0094] In one embodiment of the present invention, S101 Basic parameters and model establishment:

[0095] The DCS control system (6) receives the water-cooled wall temperature signal from the wall temperature sensor (5) via a transmission cable. Based on the different locations of the wall temperature sensor (5) in the furnace of the boiler (4), the temperature measuring points are divided into several groups (assuming each group has a temperature measuring points, for a total of b groups). The temperature values ​​of each measuring point are converted into the temperature deviation rate within the group (defined as: measuring point temperature / average temperature within the group of the measuring point - 1), and then integrated to construct a temperature deviation matrix T with a row and b column, which is used to reflect the wall temperature distribution deviation of the furnace of the boiler (4).

[0096] S1012) Let n be the total number of medium-speed coal mills (1) in the boiler (4), and let each coal powder distributor (2) correspond one-to-one with a medium-speed coal mill (1); each coal powder distributor (2) has m adjusting baffles. Therefore, a matrix A with m rows and n columns is defined as (φ ij ), where φ ijφij represents the opening degree of the i-th regulating damper of the j-th pulverized coal distributor (2) (i = 1, 2, …, m, j = 1, 2, …, n). Therefore, A can reflect the opening degrees of the regulating dampers of all the pulverized coal distributors (2) of the unit.

[0097] S1013) Define the n-order diagonal matrix B = diag (λ1, …, λn), where λ1, …, λn represent the running load rates of the j-th medium-speed coal mill (1) (defined as the current running output / maximum running output; the value range is 0 ~ 1, 0 represents that the coal mill is shut down, and 1 represents that the coal mill is running at the maximum output). When the coal quality is maintained unchanged, the change of the boiler load is directly caused by the change of the running load rate of each medium-speed coal mill (1); and when the boiler load is maintained unchanged, the change of the coal quality will also cause the change of the running load rate of each medium-speed coal mill (1). Therefore, B can comprehensively reflect the running conditions of all the medium-speed coal mills (1) of the unit under various different boiler loads and coal qualities. n j

[0098] S1014) Since the pulverized coal distributor (2) is one-to-one corresponding to the medium-speed coal mill (1), multiplying the matrix AB can represent the opening degree of the regulating damper in the pulverized coal distributor (2) of the running medium-speed coal mill (1) under various different boiler loads and coal qualities (for example, the j-th mill is not running, and λj of the non-running mill is 0). j

[0099] S1015) Under various different B value conditions, the opening degree of the regulating damper of each pulverized coal distributor (2) is repeatedly manually adjusted to change the A value, so as to determine the relatively optimal T value under the corresponding condition, and finally the matrix data set [A, B, T] with a sufficient sample number is formed; the data set is subjected to necessary data cleaning and preprocessing pre-operations, and then the support vector regression method is used to train the SVR model, so as to establish the mapping relationship from the matrix set AB to the matrix set T.

[0100] In an embodiment of the present application, S102 sets the control parameters:

[0101] S1021) Let t1 and t2 be the first and second sampling intervals respectively. By reasonably selecting t1 and t2, the changes of the running conditions of each medium-speed coal mill (1) of the unit can be tracked in time, and the optimization of the pulverized coal distribution and regulation results based on the wall temperature distribution field of the boiler (4) can be easily converged, so as to finally realize stable automatic control of the system.

[0102] S1022) Define △B = ||B1-B0|| F ​​​where B0 is the B value at the beginning of the sampling, B1 is the B value after one t1, ||B1-B0|| is the Frobenius norm of the matrix difference (B1-B0) (the square root of the sum of squares of all elements of the matrix). F △B can be used to represent the overall change of the operation load rate of each medium-speed coal mill (1) in the unit.

[0103] S1023) Define△T=T'-T", where T' is the measured furnace water wall temperature deviation matrix of the boiler (4), and T" is the furnace water wall temperature deviation matrix of the boiler (4) predicted by the trained SVR model according to the A and B values; define the error function E=||△T||. F as the objective function of automatic control, where ||△T|| F is the Frobenius norm of the matrix△T(the square root of the sum of squares of all elements of the matrix).

[0104] S1024) Others: related control parameters of the first and second simulated annealing algorithms, etc.

[0105] In an embodiment of the present application, the S103 control strategy is:

[0106] With t1 as the period, the△B is calculated by cyclic sampling (i.e. the value at the end of the sampling is taken as the value at the beginning of the next sampling), and the following control process is executed:

[0107] S1031 When△B≥△B u (the upper threshold), the system inversely predicts the recommended value of A from T'1, B1 (the T' and B values at the end of the first sampling) according to the trained SVR model, and manually decides whether the adjusting baffle of the pulverized coal distributor (2) is adjusted.

[0108] S1032 When△B d (the lower threshold) <△B <△B u (the upper threshold), the system executes the following control process.

[0109] S10321 First-level control process:

[0110] S103211) The system predicts T"0 from A0, B0 (the A and B values at the beginning of the first sampling) according to the trained SVR model, and calculates the error value E0 by combining the measured T'0 at the same time.

[0111] S103212) The system predicts T"1 from A0, B1 (the B value at the end of the first sampling) according to the trained SVR model, and calculates the error value E1 by combining the measured T'1 at the same time.

[0112] S103213) If E1≤E0, maintain A unchanged; if E1>E0, decide whether to accept maintaining A unchanged by the first-level simulated annealing algorithm;

[0113] S10322 If the system accepts maintaining A unchanged, end the current control; if the system does not accept maintaining A unchanged, start the secondary control process with t2 as the period, and iteratively execute the secondary control process in a loop until the stable global optimal solution of A under B1 is obtained, and then end the current control.

[0114] S10323 Secondary control process:

[0115] S103231) The system predicts T”0 according to A0 and B1 (the values of A and B at the start of secondary sampling) based on the trained SVR model, and then calculates the error value E0 by combining the measured T’0 at the same time;

[0116] S103232) A new solution A1 is randomly generated in the neighborhood of A0. The system predicts T”1 according to A1 and B1 based on the trained SVR model, and then calculates the error value E1 by combining T’0 (T’ basically does not change in a very short time);

[0117] S103233) If E1<E0, directly accept the new solution A1; if E1≥E0, decide whether to accept the new solution A1 by the secondary simulated annealing algorithm;

[0118] S103234) If the system does not accept the new solution A1, end the current control; if the system accepts the new solution A1, the DCS control system (6) issues an instruction to adjust the opening degree of the regulating baffle of each pulverized coal distributor (2), and the remote actuator completes the adjustment. The change of the opening degree of the regulating baffle of the pulverized coal distributor (2) changes the distribution of the amount of pulverized coal and the amount of air in the outlet pipe of the pulverized coal distributor (2), affects the distribution of the combustion intensity in the furnace of the boiler (4), and ultimately changes the wall temperature distribution in the furnace, resulting in a new T’ value, and ending the current control.

[0119] S1033 When △B≤△B d (lower threshold), the value of A is maintained unchanged, and the regulating baffle of the pulverized coal distributor (2) is not adjusted.

[0120] In summary, compared to the distribution of pulverized coal and air volume in each pulverized coal pipeline, the boiler furnace wall temperature field directly reflects the combustion intensity distribution within the furnace. Online pulverized coal and air volume measurement technology is immature, resulting in generally distorted test results and hindering the effective construction of feedback-based automatic regulation. In contrast, mature wall temperature measurement technology provides rapid and accurate temperature measurement, enabling real-time and accurate construction of the temperature field to promptly evaluate the final effect of pulverized coal distribution regulation. Historical operating data of the DCS system is traceable, facilitating periodic evaluation of control effectiveness and allowing for timely updates to the dataset and SVR model, thus improving system control performance. Simulated annealing algorithms are suitable for global optimization of complex problems, offering computational simplicity and robustness.

[0121] Example 3

[0122] To achieve the above embodiments, such as Figure 3 As shown, this embodiment also provides a pulverized coal distribution regulation and control device 10 for coal-fired units based on the boiler wall temperature distribution field, including:

[0123] The temperature data acquisition and grouping module 100 is used to acquire real-time temperature data of each temperature measuring point on the water-cooled wall of the boiler furnace, divide the temperature measuring points into several groups according to the location of the measuring points, calculate the temperature deviation rate of each measuring point in each group, and construct the temperature deviation matrix T.

[0124] The operating parameter acquisition and matrix construction module 200 is used to acquire the operating load rate of each medium-speed coal mill, form a diagonal matrix B, and acquire the opening degree matrix A of each coal powder distributor regulating baffle. The matrix AB is calculated to reflect the opening degree state of the regulating baffle under the current operating conditions.

[0125] The model prediction and error evaluation module 300 is used to predict the temperature deviation matrix T' based on the trained support vector regression model and matrix AB, and calculate the error function E in combination with the measured temperature deviation matrix T, which serves as the evaluation index for regulation and control.

[0126] The adjustment decision and control output module 400 is used to make adjustment decisions based on the changes in the error function E using a two-stage simulated annealing algorithm. The first-stage control determines whether the adjustment baffle needs to be adjusted, and the second-stage control iteratively searches for the global optimal solution of the adjustment baffle opening when adjustment is needed, and outputs adjustment commands through the DCS control system to achieve automatic control.

[0127] Furthermore, the temperature data acquisition and grouping module is also used for:

[0128] The temperature measuring points are divided into symmetrical combustion intensity related area groups according to the combustion area distribution in the boiler furnace. Each group contains a measuring points to enhance the local sensitivity to combustion non-uniformity.

[0129] When calculating the temperature deviation rate of each measuring point in each group, the sliding window average method is used to calculate the temperature average value in the group to reduce the influence of instantaneous temperature fluctuation on the deviation rate calculation.

[0130] Further, the operation parameter acquisition and matrix construction module is further used for:

[0131] The calculation of the operation load rate λj adopts the ratio of the current coal mill output to the maximum output, and is dynamically corrected through the historical data of the DCS system to eliminate the interference of coal quality fluctuation on the load rate calculation.

[0132] The adjustment baffle opening corresponding to the non-operating medium-speed coal mill is set to 0 in the matrix AB and multiplied by the matrix, so that the adjustment baffle opening corresponding to the coal mill has no influence on the overall control strategy.

[0133] Further, the model prediction and error evaluation module is further used for:

[0134] The support vector regression model adopts radial basis function RBF as the kernel function in the training process, and selects the optimal penalty coefficient C and kernel width γ through cross-validation.

[0135] The calculation of the error function E adopts the Frobenius norm to quantify the overall deviation between the predicted value and the measured value.

[0136] Further, it further comprises:

[0137] The coal quality parameter detection and model updating module is used for adjusting the input feature weight of the support vector regression model according to the coal quality parameter when the system detects that the coal quality changes significantly, and re-performs local model updating to improve the prediction accuracy of the model under the coal quality fluctuation working condition.

[0138] The coal pulverizing distribution regulation and control device based on the boiler wall temperature distribution field in the embodiment of the application, under the condition of coal quality fluctuation, dynamically adjusts the input feature weight of the support vector regression model and performs local model updating, further improves the adaptability and prediction accuracy of the coal pulverizing distribution optimization strategy, thereby enhancing the self-adaptive regulation ability of the system under the complex coal quality working condition, and ensuring the stability and efficiency of the combustion process.

[0139] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the description of the specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples, without contradiction.

[0140] In addition, the terms "first", "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise explicitly specified.

Claims

1. A coal pulverizing distribution adjustment control method based on a boiler wall temperature distribution field, characterized in that, Comprise: S1, obtain the real-time temperature data of each temperature measuring point on the boiler furnace water wall, and divide the temperature measuring points into several groups according to the position of the measuring points, calculate the temperature deviation rate of each measuring point in each group, and construct the temperature deviation matrix T; S2, obtain the running load rate of each medium-speed coal mill, form a diagonal matrix B, and obtain the opening matrix A of the adjusting baffle of each coal powder distributor, calculate the matrix AB to reflect the adjusting baffle opening state under the current running condition; S3, based on the trained support vector regression model, the temperature deviation matrix T' is predicted according to the matrix AB, and the error function E is calculated by combining the measured temperature deviation matrix T as the evaluation index of adjusting control; S4, according to the change of error function E, the two-stage simulated annealing algorithm is used for adjusting decision, wherein the first stage control judges whether the adjusting baffle needs to be adjusted, the second stage control iteratively searches the global optimal solution of the adjusting baffle opening when the adjusting baffle needs to be adjusted, and the adjusting instruction is output through the DCS control system to realize automatic control.

2. The method of claim 1, wherein, Said S1 comprises: S11, the temperature measuring points are divided into symmetrically related regions of combustion intensity according to the distribution of the combustion region of the boiler furnace, each group contains a measuring point, so as to enhance the local sensitivity to the unevenness of combustion; S12, when calculating the temperature deviation rate of each measuring point in each group, the sliding window average method is used to calculate the average temperature in the group, so as to reduce the influence of instantaneous temperature fluctuation on the calculation of deviation rate.

3. The method of claim 1, wherein, Said S2 comprises: S21, the calculation of the running load rate λj adopts the ratio of the current coal mill output to the maximum output, and the historical data of the DCS system is used for dynamic correction, so as to eliminate the interference of coal quality fluctuation on the calculation of load rate; S22, the opening of the adjusting baffle corresponding to the non-running medium-speed coal mill is set to 0 in the matrix AB by matrix multiplication, so that the opening of the adjusting baffle corresponding to the non-running medium-speed coal mill has no influence on the overall control strategy.

4. The method of claim 1, wherein, Said S3 comprises: S31, the support vector regression model adopts radial basis function RBF as kernel function in the training process, and the optimal penalty coefficient C and kernel width γ are selected by cross-validation; S32, the calculation of the error function E adopts Frobenius norm to quantify the overall deviation between the predicted value and the measured value.

5. The method of claim 1, wherein, Further comprising: S5, when the system detects that the coal quality of the coal fired changes significantly, the input feature weight of the support vector regression model is adjusted according to the coal quality parameter, and the local model is updated again, so as to improve the prediction accuracy of the model under the fluctuating coal quality condition.

6. A coal pulverizing distribution adjustment control device for a coal-fired unit based on a boiler wall temperature distribution field, characterized by, Comprise: A temperature data acquisition and grouping module is used for obtaining the real-time temperature data of each temperature measuring point on the boiler furnace water wall, and dividing the temperature measuring points into several groups according to the position of the measuring points, calculating the temperature deviation rate of each measuring point in each group, and constructing the temperature deviation matrix T; A running parameter acquisition and matrix construction module is used for obtaining the running load rate of each medium-speed coal mill, forming a diagonal matrix B, and obtaining the opening matrix A of the adjusting baffle of each coal powder distributor, calculating the matrix AB to reflect the adjusting baffle opening state under the current running condition; A model prediction and error evaluation module is configured to predict a temperature deviation matrix T' based on the trained support vector regression model according to the matrix AB, and to calculate an error function E by combining the measured temperature deviation matrix T, as an evaluation index for the adjustment control. An adjustment decision and control output module is configured to make adjustment decisions by using a two-stage simulated annealing algorithm according to the change of the error function E, wherein the first-stage control judges whether the adjustment damper needs to be adjusted, the second-stage control iteratively searches for a global optimal solution of the adjustment damper opening degree when adjustment is needed, and outputs an adjustment instruction through a DCS control system to realize automatic control.

7. The apparatus of claim 6, wherein, The temperature data acquisition and grouping module is further configured to: divide the temperature measuring points into symmetrically distributed groups according to the combustion region of the boiler furnace, each group containing a measuring point, to enhance the local sensitivity to uneven combustion; when calculating the temperature deviation rate of each measuring point in each group, a sliding window average method is used to calculate the average temperature in the group, to reduce the influence of instantaneous temperature fluctuations on the deviation rate calculation.

8. The apparatus of claim 6, wherein, The operating parameter acquisition and matrix construction module is further configured to: the calculation of the operating load rate λj uses the ratio of the current coal mill output to the maximum output, and is dynamically corrected through historical data of the DCS system, to eliminate the interference of coal quality fluctuations on the load rate calculation; the adjustment damper opening degree corresponding to the non-operating medium-speed coal mill is set to 0 in the matrix AB by matrix multiplication, so that the adjustment damper opening degree corresponding to the coal mill has no effect on the overall control strategy.

9. The apparatus of claim 6, wherein, The model prediction and error evaluation module is further configured to: the support vector regression model uses a radial basis function RBF as a kernel function during the training process, and selects the optimal penalty coefficient C and kernel width γ through cross-validation; the calculation of the error function E uses the Frobenius norm to quantify the overall deviation between the predicted value and the measured value.

10. The apparatus of claim 6, wherein, Further comprising: a coal quality parameter detection and model updating module is configured to adjust the input feature weight of the support vector regression model according to the coal quality parameters when the system detects that the coal quality has changed significantly, and to re-perform local model updating, to improve the prediction accuracy of the model under fluctuating coal quality conditions.

Images