Method and system for coordinated regulation of petroleum coke calcination furnace temperature

Abstract

The application relates to the technical field of furnace temperature regulation, and discloses a petroleum coke calcination furnace temperature cooperative regulation method and system, the method comprising the following steps: acquiring a temperature distribution, a fuel addition amount, a material movement speed and a convection radiation value; obtaining a predicted temperature sequence according to the temperature distribution and the fuel addition amount; obtaining an adjusted addition amount according to the predicted temperature sequence and a preset target temperature distribution; obtaining a heat transfer correction coefficient according to the adjusted addition amount; generating an optimized distribution scheme according to the heat transfer correction coefficient; determining a heat absorption rate change according to the optimized distribution scheme; generating a compensation input sequence according to the heat absorption rate change and the convection radiation value; and outputting a furnace temperature stable state according to the compensation input sequence. The method can improve the temperature stability in the furnace.

Description

Technical Field

[0001] This invention relates to the field of furnace temperature control technology, and in particular to a method and system for coordinated control of furnace temperature in a petroleum coke calcining furnace. Background Technology

[0002] Currently, petroleum coke calciners operate under complex conditions involving continuous dynamic changes in material movement, flame position, and multiple temperature segments, making it difficult to accurately capture heat transfer paths and intensities in real time. A lag exists between fuel adjustment and temperature response, causing adjustments to often fall behind actual needs, resulting in fluctuating furnace temperatures and localized over- or under-burning. Therefore, a technology that combines data acquisition and monitoring control functions is urgently needed to achieve coordinated and stable control of multi-segment furnace temperatures.

[0003] In one existing technology, the calcining furnace control system is equipped with several thermocouples along the furnace axis to measure the temperature of each section in real time and feed it back to the controller. The controller compares the temperature of each measuring point with the preset target temperature value. When the temperature of a certain section deviates from the target, it directly adjusts the opening of the corresponding fuel supply valve proportionally according to the magnitude of the deviation. At the same time, the total fuel injection is monitored by a flow meter, and the operator manually adjusts the total amount setting based on experience to maintain overall thermal balance. This technology relies on real-time temperature deviation signals for feedback adjustment, without establishing a dynamic model of heat transfer within the furnace, nor predicting or compensating for the lag process from fuel changes to temperature response. Existing technologies rely on simple temperature feedback and experience-based adjustments, which cannot capture dynamic heat transfer and response lag, resulting in sluggish regulation, large temperature fluctuations, and a tendency for localized overburning.

[0004] Therefore, existing technologies struggle to maintain stable furnace temperatures. Summary of the Invention

[0005] This invention provides a method and system for coordinated temperature control in a petroleum coke calcining furnace to improve the temperature stability inside the furnace.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for coordinated temperature control of a petroleum coke calcining furnace, comprising: Acquire temperature distribution, fuel dosage, material movement velocity, and convective radiation values; Based on the temperature distribution and the fuel dosage, the furnace temperature is predicted using a pre-built temperature prediction model to obtain a predicted temperature sequence. Based on the predicted temperature sequence and the preset target temperature distribution, local anomaly features are extracted to obtain the maximum deviation vector, and fuzzy logic incremental correction is performed based on the maximum deviation vector to obtain the adjusted dosage. Based on the adjusted dosage, heat deviation analysis is performed by constructing theoretical heat release curves and actual heat release curves to obtain heat deviation vectors. Then, heat transfer characteristics are matched based on the heat deviation vectors to obtain heat transfer correction coefficients. Based on the heat transfer correction coefficient, dynamic thermal resistance correction is performed to obtain an updated thermal resistance matrix. Based on the updated thermal resistance matrix, an online overburn risk assessment is performed to generate an optimized allocation scheme. According to the optimized allocation scheme, the position of the combustion flame is adjusted to obtain the adjusted flame position, and the dynamic heat absorption of the material is simulated based on the adjusted flame position and the material movement speed to determine the change in the heat absorption rate. Based on the change in the heat absorption rate and the convective radiation value, a pure time lag effect diagnosis is performed to obtain the compensation time, and dynamic time compensation is performed based on the compensation time to generate a compensation input sequence. Based on the compensation input sequence, instruction coordination and arbitration are performed to generate a low-level execution signal, and closed-loop furnace temperature control is performed based on the low-level execution signal to output a stable furnace temperature state.

[0007] Secondly, the present invention provides a system for coordinated temperature control of a petroleum coke calcining furnace, comprising: The data acquisition module is used to acquire temperature distribution, fuel dosage, material movement speed, and convective radiation values. The prediction module is used to predict the furnace temperature based on the temperature distribution and the fuel dosage using a pre-built temperature prediction model, and obtain a predicted temperature sequence. The correction module is used to extract local anomaly features based on the predicted temperature sequence and the preset target temperature distribution, obtain the maximum deviation vector, and perform fuzzy logic incremental correction based on the maximum deviation vector to obtain the adjusted dosage. The matching module is used to perform heat deviation analysis by constructing a theoretical heat release curve and an actual heat release curve based on the adjusted dosage, to obtain a heat deviation vector, and to perform heat transfer characteristic matching based on the heat deviation vector to obtain a heat transfer correction coefficient. The optimization module is used to perform dynamic thermal resistance correction based on the heat transfer correction coefficient, obtain an updated thermal resistance matrix, and perform online overburn risk assessment based on the updated thermal resistance matrix to generate an optimized allocation scheme. The adjustment module is used to adjust the position of the combustion flame according to the optimized allocation scheme, obtain the adjusted flame position, and perform dynamic heat absorption simulation of the material based on the adjusted flame position and the material movement speed to determine the change in heat absorption rate. The compensation module is used to perform pure time lag effect diagnosis based on the change in heat absorption rate and the convective radiation value, obtain the compensation time, and perform dynamic time compensation based on the compensation time to generate a compensation input sequence. The output module is used to perform instruction coordination and arbitration based on the compensation input sequence, generate a low-level execution signal, and perform closed-loop furnace temperature control based on the low-level execution signal to output a stable furnace temperature state.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains real-time data from multiple dimensions such as temperature distribution, fuel addition, material movement speed and convective radiation value, and combines them with a dynamic heat and temperature prediction model to accurately predict the temperature inside the furnace. It also uses local abnormal area location and fuzzy logic incremental correction to achieve fine control of the temperature distribution of the calcining furnace, thereby improving the stability and response speed of temperature control.

[0009] (2) This invention generates an optimized allocation scheme by analyzing heat deviation and matching heat transfer characteristics, combined with dynamic thermal resistance correction and online overburning risk assessment. It can dynamically adjust the position of the combustion flame and perform dynamic heat absorption simulation of materials, thereby reducing the risk of overburning and improving the thermal efficiency and product quality consistency of the calcination process.

[0010] (3) Based on the pure hysteresis effect diagnosis of the heat absorption rate change and the convective radiation value, the present invention generates the underlying execution signal through dynamic time compensation and command coordination arbitration, realizes the self-adaptation and collaborative optimization of the closed-loop furnace temperature control system, enhances the robustness of the system to changes in operating conditions, and finally outputs a stable and reliable furnace temperature state, providing reliable support for the intelligentization and energy saving of petroleum coke calcination process. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the process for the coordinated temperature control method of a petroleum coke calcining furnace provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the petroleum coke calcining furnace temperature coordinated control system provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides a method for coordinated temperature control of a petroleum coke calcining furnace, comprising the following steps: S11, to obtain temperature distribution, fuel dosage, material movement velocity and convective radiation values; S12, Based on the temperature distribution and the fuel addition amount, the furnace temperature is predicted using a pre-built temperature prediction model to obtain a predicted temperature sequence; S13, Based on the predicted temperature sequence and the preset target temperature distribution, local anomaly features are extracted to obtain the maximum deviation vector, and fuzzy logic incremental correction is performed based on the maximum deviation vector to obtain the adjusted dosage. S14. Based on the adjusted dosage, heat deviation analysis is performed by constructing the theoretical heat release curve and the actual heat release curve to obtain the heat deviation vector, and heat transfer characteristic matching is performed based on the heat deviation vector to obtain the heat transfer correction coefficient. S15. Based on the heat transfer correction coefficient, perform dynamic thermal resistance correction to obtain an updated thermal resistance matrix, and perform online overburn risk assessment based on the updated thermal resistance matrix to generate an optimized allocation scheme. S16, According to the optimized allocation scheme, the position of the combustion flame is adjusted to obtain the adjusted flame position, and the dynamic heat absorption of the material is simulated based on the adjusted flame position and the material movement speed to determine the change in the heat absorption rate; S17. Based on the change in heat absorption rate and the convective radiation value, perform pure time lag effect diagnosis to obtain the compensation time, and perform dynamic time compensation based on the compensation time to generate a compensation input sequence. S18, according to the compensation input sequence, perform instruction coordination arbitration, generate a low-level execution signal, and perform closed-loop furnace temperature control according to the low-level execution signal to output a stable furnace temperature state.

[0014] In step S11, the temperature distribution, fuel dosage, material movement speed, and convective radiation value are obtained.

[0015] Specifically, during the operation of the petroleum coke calciner, temperature distribution data is collected in real time by multiple thermocouple arrays installed at preset positions along the axial and radial axes of the furnace body. Each thermocouple corresponds to a spatial coordinate point, and its output temperature value constitutes a temperature distribution vector, the dimension of which is equal to the total number of thermocouples. Fuel dosage data is collected by a mass flow meter installed on the fuel pipeline, and its output is the mass of fuel entering the furnace per unit time. The material movement velocity is measured by an encoder installed at the feed end of the furnace body. This encoder is linked to the pushing mechanism, and the velocity is converted into the linear velocity of the material movement through pulse counting. Convective radiation values ​​are collected by a radiative heat flux meter installed at the observation window on the furnace wall. This instrument directly outputs the radiative heat transfer flux inside the furnace.

[0016] The acquisition of the above four types of data was completed through standard industrial communication protocols. The temperature distribution vector was stored as an array, with each element corresponding to a temperature value at a fixed spatial location. Fuel dosage was used as a scalar value in the calculations, material velocity was represented as a scalar value indicating the speed of material movement within the furnace, and convective radiation was used as a scalar input reflecting the intensity of radiation within the furnace. All data was synchronized using a unified timestamp, and the sampling interval was set and kept fixed according to process requirements to ensure the consistency of the time-series data.

[0017] In step S12, based on the temperature distribution and the fuel dosage, the furnace temperature is predicted using a pre-built temperature prediction model to obtain a predicted temperature sequence.

[0018] In one implementation, the temperature distribution is mapped into a spatial distribution matrix to obtain a spatial temperature matrix; Multiply the fuel dosage by a preset unit calorific value to obtain the heat input variable; Based on the spatial temperature matrix and the heat input variables, vector concatenation is performed to obtain the system state vector; Based on the system state vector, the furnace temperature is predicted using a pre-built temperature prediction model to obtain a predicted temperature sequence, wherein the pre-built temperature prediction model includes the regional thermal resistance matrix inside the furnace.

[0019] In one implementation, the historical system state vector and historical temperature sequence are obtained; The historical system state vector is input into the input layer of the initially constructed neural network model for training, and the predicted temperature sequence output by the output layer of the neural network model is obtained. Substitute the predicted temperature sequence and the historical temperature sequence into the loss function to calculate the loss value; The gradient of the output layer of the neural network model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data. Based on the gradient data and the preset learning rate, the parameters of each layer of the neural network model are updated using the gradient descent method. The parameters of each layer are iteratively updated until the number of training iterations of the neural network model is greater than a preset number of iterations, or the loss value of the neural network model is less than a preset loss threshold. At this point, the training is considered complete, and the trained temperature prediction model is obtained.

[0020] Specifically, the temperature distribution is derived from the temperature distribution vector obtained in step S11, which contains temperature values ​​from multiple spatial measuring points. The fuel dosage, also derived from step S11, represents the mass of fuel entering the furnace per unit time. First, the temperature distribution vector is rearranged into a two-dimensional matrix, called the spatial temperature matrix, according to the spatial arrangement of thermocouples within the furnace. The rows and columns of this matrix correspond to the axial and radial coordinates of the furnace body, and each matrix element stores the measured temperature value at the corresponding coordinate point. Simultaneously, the fuel dosage is multiplied by a preset unit calorific value. This unit calorific value is obtained through laboratory calorific value determination of the fuel sample used; its value is fixed. The resulting heat input variable is a scalar value reflecting the current total heat input.

[0021] The spatial temperature matrix is ​​expanded row-wise into a one-dimensional vector, which is then concatenated with the heat input variables to form the system state vector. The concatenation method involves adding the heat input variables as additional elements to the end of the expanded one-dimensional temperature vector. The dimension of the system state vector is equal to the total number of spatial temperature measurement points plus one. This vector comprehensively represents the current thermal state within the furnace and the external energy input. The purpose of constructing the system state vector is to integrate the spatially distributed temperature information and the total heat input into a unified mathematical expression, serving as the input to the temperature prediction model.

[0022] The historical data required for training the temperature prediction model comes from the system state vector collected and stored over a long period during the stable operation of the calcining furnace, along with the corresponding actual temperature sequences at multiple future sampling times. The historical temperature sequences are formed by arranging the actual temperature distribution vectors collected at fixed time intervals after each historical system state vector. The data acquisition covered various typical operating conditions to ensure the model's generalization ability. Before training, the historical data was normalized, mapping temperature values ​​and heat input variables to zero and one respectively, to accelerate model convergence.

[0023] The temperature prediction model is a feedforward neural network with one hidden layer. The number of nodes in the input layer is equal to the dimension of the system state vector. The number of nodes in the hidden layer is determined by testing the model performance with different numbers of nodes (e.g., from ten to one hundred) on the training set, and selecting the number of nodes that minimizes the root mean square error on the validation set as the final value. The hidden layer uses a linear rectified function as its activation function. The number of nodes in the output layer is equal to the length of the future temperature sequence to be predicted, which is determined by dividing the required prediction step size by the sampling interval. The output layer uses a linear activation function.

[0024] During model training, the historical system state vector is input into the network, undergoes a linear transformation from the input layer to the hidden layer, and is processed by an activation function to obtain the hidden layer output. The hidden layer output is then subjected to another linear transformation to obtain the predicted temperature sequence of the network output. The mean square error between the predicted temperature sequence and the corresponding historical temperature sequence is calculated as the loss value. The partial derivative of the loss function with respect to the network output is used to obtain the output layer error, which is multiplied by the derivative of the output layer activation function (which is a constant here) to obtain the output layer gradient. Here, the predicted temperature sequence refers to the predicted values ​​of the full distribution of furnace temperature at future consecutive prediction steps. When extracting local anomaly features in step S13, the temperature distribution vector of the first prediction time in the sequence, i.e., the most recent future time, is used. Unless otherwise specified, the comparisons and calculations related to temperature distribution in subsequent steps refer to the temperature distribution vector based on the current real-time measurement or the most recent future prediction.

[0025] The output layer gradient propagates back to the hidden layers using the chain rule. The output layer gradient is obtained by multiplying the output layer gradient by the transpose of the weight matrix from the hidden layer to the output layer, and then by the derivative of the hidden layer activation function at the corresponding input value (the derivative of a linear rectified function is one when the input is greater than zero and zero when it is less than or equal to zero). Based on the hidden layer gradient and the input data, the gradients of the hidden layer weights and biases are calculated. Similarly, based on the output layer gradient and the hidden layer output, the gradients of the output layer weights and biases are calculated. A fixed learning rate is preset, which is selected after trying several candidate values ​​(e.g., 0.01, 0.001) in the early stages of training and observing the descent rate and stability of the loss function in the first few iterations.

[0026] Gradient descent is used for parameter updates. The current value of each layer's parameter is subtracted from the product of its gradient and the learning rate to obtain the updated parameter value. The training process is iterative, with each iteration using all or part of the training data to calculate the gradient and update the parameters. The preset maximum number of training iterations is set to a large number (e.g., 10,000), and the preset loss threshold is set by analyzing historical training records, taking 120% of the training loss value corresponding to the model's best performance on the validation set. When the actual number of iterations exceeds the maximum number of training iterations, or the current loss value falls below the loss threshold, training stops, the network parameters at this point are saved, and the trained temperature prediction model is obtained.

[0027] The trained model receives the current system state vector as input, performs forward propagation calculations, and directly outputs a predicted temperature sequence. This sequence is a one-dimensional vector, whose elements represent the predicted temperature values ​​of all spatial measurement points at each future prediction time, arranged in temporal and spatial order. This predictive capability enables the control system to sense temperature change trends in advance, laying the foundation for prediction-based feedforward and feedback coordinated regulation, thereby overcoming the problems of untimely adjustment and temperature fluctuations caused by the lag in thermal processes in traditional control.

[0028] It should be noted that the regional thermal resistance matrix is ​​not directly derived from physical measurements, but is introduced as a trainable parameter matrix within the neural network model and learned together. In the initial stage of model construction, based on the physical structure of the calcining furnace, its internal space is divided into several continuous regions, for example, divided into several segments along the axial direction. Each segment is considered a region, and a matrix is ​​initialized. Its diagonal elements are initialized to a baseline thermal resistance value estimated based on the thermal conductivity of a typical insulation material in that region, while the off-diagonal elements are initialized to zero or a very small value, representing the initial coupling thermal resistance between regions. This matrix is ​​defined as a physical knowledge layer parameter of the model. After training... During forward propagation, before entering the fully connected layer, the system state vector undergoes a pre-defined operation with the matrix. For example, the portion of the state vector representing the average temperature of each region is used to construct a vector, the product function of the vector and the matrix is ​​calculated, and the result is then concatenated with the original state vector. This embeds the physical constraints of thermal resistance into the network in a structured manner. During backpropagation, the parameters of the matrix are updated along with the other weight parameters of the neural network using gradient descent. In this way, the trained model not only learns the dynamic patterns in the data, but its internal regional thermal resistance matrix is ​​also adaptively adjusted to a parameterized representation that best reflects the actual heat transfer characteristics inside the furnace.

[0029] In step S13, local anomaly features are extracted based on the predicted temperature sequence and the preset target temperature distribution to obtain the maximum deviation vector, and fuzzy logic incremental correction is performed based on the maximum deviation vector to obtain the adjusted dosage.

[0030] In one implementation, a temperature deviation matrix is ​​obtained by performing point-by-point difference calculation based on the predicted temperature sequence and the preset target temperature distribution. Calculate the Euclidean norm of the temperature deviation matrix to obtain the overall deviation norm; When the overall deviation norm exceeds the preset overall deviation norm threshold, the data of the continuous region with the largest absolute value in the temperature deviation matrix is ​​extracted to form the maximum deviation vector; When the overall deviation norm does not exceed the preset overall deviation norm threshold, the fuel dosage remains unchanged; Based on the maximum deviation vector and the preset deviation correction rule base, fuzzy inference calculation is performed using a fuzzy logic control algorithm based on triangular membership function and centroid method to obtain the fuel correction increment. The adjusted dosage is obtained by performing a weighted summation calculation based on the fuel correction increment and the fuel addition amount.

[0031] Specifically, the predicted temperature sequence comes from the predicted temperature values ​​at multiple future moments output in step S12, and this sequence is arranged in temporal and spatial order. The preset target temperature distribution is a fixed vector with the same dimension as the temperature distribution vector, and each element represents the expected temperature value at the corresponding spatial measurement point. First, a point-by-point difference calculation is performed, subtracting the corresponding elements of the predicted temperature distribution vector at the first future moment in the predicted temperature sequence from the target temperature distribution vector to obtain a temperature deviation matrix. The row and column structure of this matrix is ​​consistent with the spatial temperature matrix, and each element value represents the deviation between the predicted temperature and the target temperature at that location; a positive value indicates that the predicted temperature is too high, and a negative value indicates that the predicted temperature is too low.

[0032] The Euclidean norm of the temperature deviation matrix is ​​calculated to quantify the overall deviation level. The specific calculation process involves treating each element of the temperature deviation matrix as a vector, calculating the Euclidean norm of that vector, i.e., squaring each element value, summing all the squared values, and then taking the square root of the sum. The result is the overall deviation norm. Specifically, the preset overall deviation norm threshold is determined through historical data analysis. A large amount of historical operating data with stable furnace temperatures and qualified product quality is collected, and the Euclidean norm of the temperature deviation matrix corresponding to each batch of data is calculated. The 90th percentile of all historical norm values ​​is taken as the threshold. This threshold represents the acceptable upper limit of overall deviation and is used to determine whether regulatory intervention is needed.

[0033] When the calculated overall deviation norm exceeds a preset threshold, the system determines that a temperature anomaly exists and needs to extract local anomaly features. In the temperature deviation matrix, the element with the largest absolute value is found, and then, using this element as the center, the system expands to its adjacent positions in all directions, extracting consecutive adjacent elements with the same deviation sign (both positive or both negative) to form a submatrix. Expanding this submatrix row-wise yields a one-dimensional vector, which is the maximum deviation vector. This vector represents the information of the most severe local deviation region between the current predicted temperature and the target temperature. If the overall deviation norm does not exceed the threshold, no further correction is performed, and the current fuel dosage is maintained unchanged. In this case, the adjusted fuel dosage is equal to the current fuel dosage.

[0034] The calculation of fuel correction increments relies on a pre-defined deviation correction rule base. This rule base is constructed based on a large amount of historical debugging data and simulation experimental data. The construction process involves extracting the maximum deviation vector sample from historical operating data, calculating the average deviation and distribution range of each sample, and recording the actual fuel adjustment amount that stabilizes the furnace temperature for that sample. Next, cluster analysis is performed on the numerical ranges of the average deviation, distribution range, and fuel adjustment amount, dividing continuous values ​​into several intervals, each assigned a linguistic value. For example, the average deviation is categorized as "negative large," "negative small," "zero," "positive small," and "positive large." Then, for each historical data point, its average deviation, distribution range, and fuel adjustment amount are categorized into the corresponding linguistic values, forming an initial rule of "if...then...". Finally, the initial rules with the same preconditions are statistically analyzed, and the conclusion with the highest frequency is selected as the final rule, thus establishing a complete fuzzy rule base. The core of the rule base is a set of fuzzy rules, each in the form of "if the average deviation of the maximum deviation vector is a certain linguistic value and its spatial distribution range is a certain linguistic value, then the fuel correction increment is a certain linguistic value." The average deviation and distribution range are used as input fuzzy variables, and their universe of discourse is determined based on the statistical range of the historical maximum deviation vector, and divided into multiple fuzzy subsets, each corresponding to a triangular membership function. The fuel correction increment is used as the output fuzzy variable, and its universe of discourse is determined based on the range of historically effective fuel adjustment amounts, similarly divided into multiple fuzzy subsets and assigned triangular membership functions.

[0035] In the fuzzy inference process, firstly, the average value of all elements of the input maximum deviation vector is calculated as the actual average deviation value, and the number of spatial locations contained in this vector is calculated as the actual distribution range value. Triangular membership functions are used to calculate the membership degree of these two actual values ​​to each input fuzzy subset. Then, based on the fuzzy rules in the rule base, the minimum value operation is used to determine the activation strength of each rule. For activated rules, the membership function of the output fuzzy subset of its conclusion part is truncated according to the activation strength.

[0036] The defuzzification process employs the centroid method. The membership functions of all activated and truncated output fuzzy subsets are superimposed along the vertical axis to obtain a total, irregularly shaped output fuzzy set, which represents a fuzzy distribution over the fuel correction increment universe. The coordinates of the geometric centroid of the region covered by this fuzzy distribution on the horizontal axis (i.e., the fuel correction increment universe) are calculated. Specifically, the universe is discretized into a large number of densely spaced points. For each point, its membership degree is taken. The horizontal coordinates of all points are multiplied by their respective membership degrees, summed, and then divided by the sum of the membership degrees. The resulting quotient is the horizontal coordinate of the geometric centroid. This horizontal coordinate is a specific numerical value, serving as the final fuel correction increment output. The fuel correction increment is a scalar value; a positive value indicates that fuel needs to be increased, and a negative value indicates that fuel needs to be decreased.

[0037] After obtaining the fuel correction increment, it is weighted and superimposed with the current fuel injection amount obtained in step S11 to obtain the adjusted injection amount. The superposition calculation formula is that the adjusted injection amount equals the current fuel injection amount plus the fuel correction increment multiplied by a preset weighting coefficient. This preset weighting coefficient is fixed at 0.7, the purpose of which is to make the control action smooth and avoid system oscillation caused by excessively large single correction increments. The weighting coefficient value is determined by simulation analysis of historical control processes, selecting the value that minimizes system response overshoot and shortens the adjustment time.

[0038] In step S14, based on the adjusted dosage, a heat deviation analysis is performed by constructing a theoretical heat release curve and an actual heat release curve to obtain a heat deviation vector. Then, a heat transfer characteristic matching is performed based on the heat deviation vector to obtain a heat transfer correction coefficient.

[0039] In one implementation, fuel supply is executed according to the adjusted feed rate, furnace radiation intensity is obtained, and thermal inversion calculations are performed based on a multispectral temperature measurement inversion algorithm using the least squares method to obtain a real-time heat release curve. Calculate the heat accumulation of the adjusted dosage within a preset time window, generate a theoretical heat release curve, and align it with the real-time heat release curve to obtain an aligned theoretical curve and an aligned real-time curve. The heat deviation vector is generated by calculating the point-by-point difference between the alignment theoretical curve and the alignment real-time curve. Based on the heat deviation vector, multi-scale wavelet decomposition is performed using the Mallat algorithm to obtain the nonlinear fluctuation component; Based on the nonlinear fluctuation component, heat transfer characteristics are matched using a preset heat transfer attenuation coefficient table to obtain the heat transfer correction coefficient.

[0040] Specifically, the adjusted feed rate comes from the output of step S13. The fuel supply valve opening is controlled according to the adjusted feed rate, and fuel of the corresponding flow rate is fed into the furnace. Simultaneously, raw data of furnace radiation intensity is acquired using a multispectral radiation thermometer installed at a specific observation position in the furnace. This data is a sequence of radiation intensity measurements under multiple narrow bands. A multispectral temperature measurement inversion algorithm based on the least squares method is used to calculate the real-time heat release curve. The specific process of this algorithm is as follows: An equation for the radiation transfer of the furnace medium is established. The multi-band radiation intensity measurements are used as known quantities. The temperature distribution and emissivity distribution to be solved are parameterized, and a sum-of-squares function of the difference between the parameters and the measured values ​​is constructed. The parameter values ​​are adjusted through an iterative optimization algorithm to minimize the sum-of-squares function value. The corresponding parameter solution is the inverted temperature field. Then, based on this temperature field and the known furnace geometry and medium properties, the sequence of the real-time total heat release rate of the entire furnace over time is calculated, i.e., the real-time heat release curve.

[0041] The theoretical heat release curve is generated based on the adjusted fuel dosage. The adjusted fuel dosage sequence is multiplied by the unit calorific value of the fuel to obtain the heat input sequence per unit time. This sequence is then accumulated within a preset time window. The accumulation method starts from the current moment and proceeds backwards for a fixed time length, summing the heat input values ​​at each sampling moment within the window. This yields the cumulative heat value at each time point from the start of the window to the current moment. Arranging these cumulative values ​​in chronological order constitutes the theoretical heat release curve. This curve represents the theoretical cumulative heat that the furnace should release under ideal complete combustion and instantaneous heat transfer conditions.

[0042] It should be further explained that the theoretical heat release curve is actually the desired furnace heat release rate curve. The corresponding theoretical heat input power sequence is obtained by multiplying the adjusted feed rate sequence by the unit calorific value of the fuel. Considering that there is a dynamic process from fuel injection to complete combustion, a first-order inertial element can be introduced to smooth the heat input power sequence. The time constant of this inertial element is determined by fuel characteristic experiments. The power sequence after smoothing is the theoretical heat release curve.

[0043] The theoretical heat release curve and the real-time heat release curve are time-aligned. The alignment method uses the moment of fuel adjustment as the zero point and shifts the time axes of the two curves so that their starting points correspond to the same physical moment. The aligned theoretical heat release curve is called the aligned theoretical curve, and the aligned real-time heat release curve is called the aligned real-time curve. The number of data points on the two curves is adjusted to be consistent using a linear interpolation method.

[0044] The heat deviation vector is generated by calculating the point-by-point difference between the aligned theoretical curve and the aligned real-time curve. Specifically, the cumulative heat value at each time point on the aligned theoretical curve is subtracted from the cumulative heat value at the corresponding time point on the aligned real-time curve, resulting in a series of differences. These differences, arranged in chronological order, constitute the heat deviation vector. Positive values ​​indicate that the actual heat release is lower than the theoretical value, while negative values ​​indicate that the actual heat release is higher than the theoretical value.

[0045] The heat deviation vector is decomposed into nonlinear fluctuation components using the Mallat algorithm at multiple scales. This algorithm uses a set of pre-selected low-pass and high-pass filters to convolve the heat deviation vector, decomposing the signal into approximate and detail components at different frequency scales. After a predetermined number of decomposition levels (e.g., three levels), a series of sub-signals representing fluctuation characteristics at different time scales are obtained. High-frequency detail components are extracted and merged to form the nonlinear fluctuation component. This component mainly contains the heat release fluctuation characteristics caused by the nonlinearity, hysteresis, and disturbances of the heat transfer process.

[0046] A pre-defined heat transfer attenuation coefficient table is used for heat transfer characteristic matching. This table is constructed based on historical data and heat transfer simulations. During construction, computational fluid dynamics simulations are used to simulate the deviation between the theoretical and actual measured heat release curves of the furnace under different furnace conditions (such as different material filling rates and different furnace wall ash thicknesses). This deviation is then subjected to the same wavelet decomposition as described above to extract its nonlinear fluctuation components. Then, based on the statistical characteristics (such as energy distribution and zero-crossing rate) of different furnace condition parameters and the extracted nonlinear fluctuation components, a mapping relationship is established and stored in the form of a lookup table, which is the heat transfer attenuation coefficient table. Each row in the table corresponds to a set of furnace condition parameters and fluctuation characteristics, and is associated with a normalized heat transfer correction coefficient value.

[0047] The heat transfer characteristic matching process compares the currently calculated nonlinear fluctuation component with the records in the heat transfer attenuation coefficient table. The comparison method involves calculating the Euclidean distance between the statistical eigenvector of the current fluctuation component and the eigenvectors of each record in the table, and selecting the heat transfer correction coefficient value corresponding to the record with the smallest distance as the matching result. This coefficient is a scalar value between zero and one, used to quantitatively characterize the degree of attenuation of the current actual heat transfer process relative to the ideal state.

[0048] In step S15, dynamic thermal resistance correction is performed based on the heat transfer correction coefficient to obtain an updated thermal resistance matrix, and an online overburn risk assessment is performed based on the updated thermal resistance matrix to generate an optimized allocation scheme.

[0049] In one implementation, the heat transfer correction coefficient is multiplied by the regional thermal resistance matrix of the temperature prediction model to obtain an updated thermal resistance matrix. Based on the updated thermal resistance matrix, the heat conduction is solved using a finite volume algorithm to obtain the nodal temperature vector; Based on the node temperature vector, the net radiative heat transfer is calculated using the radiation angle coefficient algorithm to obtain the heat load index; When the heat load index does not exceed the preset heat load index threshold, the fuel distribution ratio is maintained; When the heat load index exceeds the preset heat load index threshold, the fuel allocation ratio is reduced to generate an optimized allocation scheme.

[0050] Specifically, the heat transfer correction coefficient comes from the output of step S14. The regional thermal resistance matrix of the temperature prediction model is a parameter matrix within the neural network model used to characterize the thermal resistance distribution in different regions of the furnace, and its dimension is consistent with the number of regions divided within the furnace space. The specific process of dynamic thermal resistance correction is to multiply the heat transfer correction coefficient by each element in the regional thermal resistance matrix. This coefficient is a scalar value, and its multiplication operation with the matrix is ​​defined as multiplying each element of the matrix by the coefficient, resulting in a new matrix, called the updated thermal resistance matrix. This operation is equivalent to scaling the heat transfer capacity of each region within the furnace globally. The scaling ratio is determined by the heat transfer correction coefficient. A coefficient less than one indicates a decrease in actual heat transfer capacity, requiring an increase in the thermal resistance model parameters; a coefficient greater than one indicates an increase in heat transfer capacity, requiring a decrease in the thermal resistance model parameters.

[0051] After obtaining the updated thermal resistance matrix, the heat conduction is solved using a finite volume algorithm. This algorithm first discretizes the furnace space into a large number of non-overlapping small control volumes (elements). The updated thermal resistance matrix provides the material thermal resistance properties of each element or group of elements. The algorithm integrates the heat conduction differential equation over each control volume, transforming the volume integral into an area integral over the surface of each element, i.e., flux summation. For each element, its temperature change equals the sum of all heat fluxes flowing into that element divided by the heat capacity of that element. By establishing equations for all elements in the furnace according to this rule, a large system of linear equations is formed. Solving this system yields the temperature value at the center point of each element, and these temperature values ​​are arranged in order of element number to form a nodal temperature vector.

[0052] Net radiative heat transfer is calculated using the radiation angle coefficient algorithm based on the nodal temperature vector. This algorithm first calculates the radiation angle coefficient between all unit surfaces within the furnace, representing the proportion of radiant energy emitted by one surface that directly reaches another. The calculation is based on the geometric position, normal vector, and visibility of the unit surfaces. For each unit surface, its net radiative heat transfer equals the radiant energy emitted by itself minus the radiant energy received from all other surfaces. Based on the nodal temperature vector, the radiation angle coefficient of all inner surface units within the furnace to the material region surface is calculated using the radiation angle coefficient algorithm. For each material surface unit, the sum of the incident radiant energy from all inner surface units within the furnace is calculated to obtain the local irradiance of that material unit. The average local irradiance of all material surface units is then calculated, and this average value is defined as the heat load index; this index directly reflects the average radiative heating intensity received by the material.

[0053] The preset heat load index threshold was determined through historical data statistics. A large amount of historical operating data was collected where the furnace temperature was normal and no overheating occurred. The heat load index corresponding to each batch of data was calculated, and the 85th percentile of all historical index values ​​was taken as the threshold. This threshold represents the safe upper limit of the furnace radiant heat load under normal operating conditions. The basis for setting the threshold is that, under normal production conditions, the radiant heat load should be maintained below this statistical upper limit to leave a safety margin and prevent localized overheating of materials due to a sudden increase in heat load.

[0054] Online overheating risk assessment is based on a comparison between the heat load index and a preset threshold. When the calculated heat load index does not exceed the preset threshold, the current thermal state inside the furnace is considered safe, the risk of overheating is low, and the system maintains the current fuel distribution ratio unchanged. In this case, the optimized distribution scheme is the current fuel distribution scheme. When the heat load index exceeds the preset threshold, an overheating risk is identified, and the fuel supply intensity needs to be reduced to lower the heat load. The specific method for reducing the fuel distribution ratio is to multiply the current distribution ratio of each fuel supply loop by a uniform attenuation coefficient less than one, such as 0.9, to obtain a new set of proportionally reduced distribution ratio values. This new set of ratio values ​​constitutes the optimized distribution scheme. The attenuation coefficient is a preset fixed value, and its selection principle is to control the single adjustment range of the fuel distribution ratio within a reasonable range to avoid drastic fluctuations in furnace temperature.

[0055] In step S16, the position of the combustion flame is adjusted according to the optimized allocation scheme to obtain the adjusted flame position, and the dynamic heat absorption simulation of the material is performed based on the adjusted flame position and the material movement speed to determine the change in the heat absorption rate.

[0056] In one implementation, the adjusted flame position is obtained by performing a weighted average calculation based on the fuel allocation ratio of each loop in the optimized allocation scheme and the nozzle position coordinates corresponding to each loop. Based on the adjusted flame position, a three-dimensional radiation field is reconstructed using the Monte Carlo ray tracing algorithm to generate a heat flux density matrix. Based on the heat flux density matrix and the material velocity, a motion trajectory model is constructed by numerical integration of the motion trajectory using the fourth-order Runge-Kutta algorithm. Based on the motion trajectory model and the heat flux density matrix, path heat absorption is calculated through numerical integration and differentiation to confirm the change in heat absorption rate.

[0057] Specifically, the optimized allocation scheme comes from the output of step S15 and includes the allocation ratio values ​​of each fuel supply loop. The nozzle position coordinates corresponding to each loop are known fixed parameters, and the physical position of each nozzle in the furnace is represented by three-dimensional spatial coordinates. The combustion flame position adjustment is achieved through weighted average calculation. The specific process is as follows: the allocation ratio value of each fuel loop is used as a weight and multiplied by the three components of the nozzle position coordinate of that loop to obtain the weighted coordinates of the loop's contribution to the flame position. The weighted coordinates of all loops are summed over the three spatial components, and then the sum of each component is divided by the sum of the allocation ratio values ​​of all loops. The final three-dimensional coordinates obtained are the adjusted flame position. This position represents the expected spatial position of the flame center in the furnace under the current fuel allocation scheme.

[0058] Based on the adjusted flame position, a three-dimensional radiation field is reconstructed using a Monte Carlo ray tracing algorithm. The algorithm simplifies the flame into a point radiation source located at the adjusted flame position, with its radiation intensity proportional to the total fuel injection. A large number of virtual energy rays are randomly emitted from this point source into the furnace space. The initial direction of each ray is determined by a random number generator through uniform sampling in spherical space. As the rays propagate within the furnace, the absorption, reflection, or transmission processes that occur when they encounter surfaces are simulated, based on the material properties of the furnace wall and the materials. By tracing the final destination of a large number of rays (e.g., hundreds of thousands), the energy deposited by each ray on tiny volume elements at various locations within the furnace is statistically analyzed. The total energy deposited on each volume element is divided by the volume of that volume element and the total number of rays to obtain the heat flux density value of that volume element. The heat flux density values ​​of all volume elements are arranged into a three-dimensional array according to their spatial location, forming the heat flux density matrix. This matrix characterizes the intensity of heat flux distribution caused by flame radiation at any spatial location within the furnace.

[0059] It should be noted that the above simplification of the flame as a point source and the material motion as a uniform particle trajectory is a simplified simulation model used to illustrate the principle of the present invention. In actual system implementation, more accurate models can be used depending on the availability of computing resources. For example, combustion simulation results based on computational fluid dynamics (CFD) can be used as the radiation source term; or the discrete element method (DEM) can be used to simulate the actual motion and accumulation state of the material to obtain a more accurate heat flux density distribution and material heat absorption path.

[0060] The material velocity is derived from the real-time measurement in step S11. Based on the heat flux density matrix and the material velocity, a motion trajectory model is constructed by numerical integration using a fourth-order Runge-Kutta algorithm. This algorithm simplifies the material's motion within the furnace as the uniform motion of a point mass along a preset trajectory. Given the material's inlet and outlet positions and the furnace geometry, the material trajectory is discretized into a series of small time steps. Within each time step, the heat flux density value at the material's current spatial position is obtained from the heat flux density matrix through three-dimensional linear interpolation. Simultaneously, the material velocity provides the displacement along the trajectory tangent within each time step. The fourth-order Runge-Kutta algorithm uses the current displacement and heat flux density value to predict the material's position at the end of the next time step through four weighted average calculations. This process is iterated repeatedly, integrating from the inlet to the outlet, to obtain a sequence of spatial positions corresponding to each discrete time point during the material's passage through the furnace. This sequence constitutes the motion trajectory model.

[0061] Based on the motion trajectory model and the heat flux density matrix, path heat absorption is calculated through numerical integration and differentiation. First, along the spatial position sequence given by the motion trajectory model, the heat flux density value corresponding to each position is extracted from the heat flux density matrix, resulting in a sequence of heat flux density variations along the trajectory. Then, this heat flux density sequence is numerically integrated along the trajectory path. The trapezoidal rule is used for integration, averaging the heat flux density values ​​between two adjacent trajectory points and multiplying this by the path length between the two points to obtain the heat absorption of that small path segment. The heat absorption of all small path segments is accumulated to obtain the total heat absorption of the material from the inlet to the current position. Finally, the total heat absorption sequence is numerically differentiated with respect to time. The central difference method is used for differentiation, subtracting the total heat absorption of the previous moment from the total heat absorption of the next moment and dividing by twice the time step to obtain the instantaneous heat absorption rate at each moment. The change in the heat absorption rate at the current moment compared to the heat absorption rate in the previous control cycle is calculated; this change is the change in heat absorption rate.

[0062] In step S17, based on the change in heat absorption rate and the convective radiation value, a pure time lag effect diagnosis is performed to obtain the compensation time, and dynamic time compensation is performed based on the compensation time to generate a compensation input sequence.

[0063] In one implementation, a comprehensive state vector is generated by time alignment based on the change in the heat absorption rate and the convective radiation value. Based on the comprehensive state vector, the system lag characteristics are identified through impulse response analysis to obtain the lag time and the dominant time. The lag time is then divided by the dominant time to obtain the lag influence coefficient. When the lag effect coefficient exceeds the preset lag effect coefficient threshold, the lag time is multiplied by the preset compensation coefficient to obtain the compensation time; If the hysteresis coefficient does not exceed the preset hysteresis coefficient threshold, the adjusted dosage remains unchanged; The compensation time is superimposed on the time axis of the adjusted dosage to generate a compensation input sequence.

[0064] Specifically, the heat absorption rate change comes from the output of step S16 and is a scalar sequence representing the instantaneous change in the material's heat absorption rate. The convective radiation value comes from the real-time measurement value of step S11 and is a scalar sequence reflecting the combined heat transfer intensity of radiation and convection within the furnace. First, time alignment is performed. Using a unified system sampling time as a reference, interpolation is applied to the heat absorption rate change sequence and the convective radiation value sequence to ensure that the two sequences have the same time points and the same data length. The aligned heat absorption rate change value and convective radiation value are concatenated into a two-dimensional vector at each sampling moment. These vectors, arranged in chronological order, constitute the comprehensive state vector.

[0065] Based on the comprehensive state vector, the system's lag characteristics are identified through impulse response analysis. This method identifies the moment of a step change in fuel dosage in historical data and extracts the comprehensive state vector sequence within a window before and after this moment as the input signal. Simultaneously, the temperature distribution vector change sequence within the corresponding window after this step change is extracted as the output response. The cross-correlation function sequence between the input signal and the output response is calculated; the time shift corresponding to the maximum value of the cross-correlation function is identified as the system's lag time. The dominant time is obtained by analyzing the output response sequence, calculating the time required from the start of the response to reach 63.2% of its steady-state value; this time is defined as the system's dominant time constant. Dividing the identified lag time by the dominant time yields the quotient, called the lag influence coefficient.

[0066] The preset lag effect coefficient threshold is determined through statistical analysis of historical operating data. Historical data on the system's stable adjustment process under various operating conditions are collected, and the lag effect coefficient corresponding to each effective adjustment is calculated. The 75th percentile of all historical coefficient values ​​is taken as the threshold. This threshold represents the boundary of typical lag effects; exceeding this value indicates a significant lag effect requiring compensation. The threshold is set based on the fact that in most successful control cases, the lag effect coefficient remains below this statistical value; exceeding this value increases the risk of delayed control response.

[0067] The preset compensation coefficient is a fixed value of 0.5. This coefficient value is determined based on the stability analysis of the closed-loop system. The system's settling time and overshoot under different compensation coefficients are tested through simulation, and the coefficient value that shortens the settling time without increasing the overshoot is selected.

[0068] When the calculated lag effect coefficient exceeds a preset threshold, the lag effect is determined to be significant, and compensation is required. The compensation time is obtained by multiplying the lag time by the preset compensation coefficient. The calculated compensation time is a time length value in seconds. If the lag effect coefficient does not exceed the preset threshold, the lag effect is determined to be insignificant, and the adjustment dosage output in step S13 remains unchanged. At this time, the compensation input sequence is the adjustment dosage sequence itself.

[0069] When generating the compensation input sequence, the compensation time is superimposed onto the time axis of the adjusted dosage. Specifically, each timestamp in the adjusted dosage sequence is shifted forward by the number of seconds corresponding to the compensation time. For each data point in the sequence, its value remains unchanged, but its occurrence time is advanced. For future time data points that are beyond the current time due to the time shift, their dosage values ​​remain unchanged; for data points needed due to the time shift but not defined in historical time, they are filled with the dosage value of the current time. The new sequence obtained after the time shift is the compensation input sequence.

[0070] It is worth noting that the time compensation logic in this embodiment is essentially feedforward compensation. Specifically, based on the identified lag time and compensation coefficient, their product is calculated to obtain the lead amount. When generating the execution instruction for the current moment, the adjustment injection amount calculated based on the current moment's state is no longer used directly. Instead, a virtual adjustment injection amount is assumed and calculated based on the historical moment, i.e., the current moment minus the lead amount. Since it is impossible to backtest the history, this virtual adjustment injection amount is approximated by querying the record of the control instruction that actually occurred near the current moment minus the lead amount, or by using the stored short-term historical data of the system state for rapid replay calculation. The instruction obtained in this way is the compensation instruction for the lag effect.

[0071] In step S18, instruction coordination arbitration is performed according to the compensation input sequence to generate a low-level execution signal, and closed-loop furnace temperature control is performed according to the low-level execution signal to output a stable furnace temperature state.

[0072] In one implementation, the coordinated control instructions are obtained by decomposing the instructions according to the compensation input sequence using a pre-built instruction hierarchical mapping rule. The coordination control command is input into a preset exception rule base for conflict arbitration to generate a low-level execution signal; Based on the underlying execution signal, the petroleum coke calciner is driven to perform closed-loop furnace temperature control and output a stable furnace temperature state.

[0073] Specifically, the compensation input sequence, derived from the output of step S17, is a time-compensated fuel dosage instruction sequence. Instruction coordination arbitration first decomposes the instructions using a pre-constructed hierarchical mapping rule. The construction process of the pre-constructed hierarchical mapping rule involves collecting historical data from a petroleum coke calciner that has been operating stably for a long period and whose products are qualified, and extracting the correspondence between the total fuel dosage and the actual flow rate of each independent fuel supply loop. This data is then sorted by the total dosage value from smallest to largest, and the total dosage range is evenly divided into several continuous intervals. For all historical data within each interval, the proportion of the actual flow rate of each loop to the total dosage at that moment is calculated, and then the arithmetic mean of these proportions is calculated to obtain the standard allocation coefficient for each loop within that interval. These coefficients satisfy the condition that the sum of all loop coefficients is one. A mapping relationship is established between each total dosage interval and its corresponding set of standard allocation coefficients, and this is stored in the form of a structured lookup table, thus forming the instruction hierarchical mapping rule base.

[0074] When decomposing the command, the interval to which the total fuel injection value belongs is found based on the total fuel injection value at each moment in the compensation input sequence. The total injection value is multiplied by the standard allocation coefficient of each loop corresponding to that interval to obtain a set of fuel flow set values ​​for each loop. This set of set values ​​constitutes the coordinated control command.

[0075] The coordinated control commands are input into a pre-defined exception rule base for conflict arbitration. The exception rule base is constructed based on equipment safety constraints and process limitations. The rule base contains multiple logical judgment rules, each in the form of "If the set flow value of a certain loop is greater than its corresponding preset safety upper limit, then modify the set flow value of that loop to the safety upper limit value." The safety upper limit value of each loop is directly obtained from the equipment nameplate parameters and process design documents and stored as a fixed constant in the rule base. The conflict arbitration process iterates through the set flow value of each loop in the coordinated control command, comparing it one by one with the corresponding safety upper limit value in the rule base. If the set value exceeds the upper limit, the output set value of that loop is corrected to the safety upper limit value; otherwise, the original set value remains unchanged. The set of set values ​​of all loops after arbitration and correction constitutes the underlying execution signal.

[0076] Based on the underlying execution signals, the petroleum coke calcining furnace is driven to perform closed-loop furnace temperature control. These signals are sent to the field actuators, primarily the regulating valves in each fuel supply loop, via a standard industrial communication protocol. Each regulating valve receives its corresponding fuel flow setpoint. Its built-in closed-loop controller calculates the valve opening adjustment based on the deviation between the setpoint and the current actual flow rate using a proportional-integral control algorithm, and drives the valve position change to ensure the actual fuel flow tracks the setpoint. Precise control of the fuel flow in each loop achieves the final regulation of the total heat input and its spatial distribution within the furnace. Under the updated fuel supply, the furnace temperature distribution gradually approaches the preset target temperature distribution. When the Euclidean norm between the real-time temperature distribution vector and the target temperature distribution vector is less than the preset overall deviation norm threshold for multiple consecutive sampling periods, the system determines that the furnace temperature has reached a stable state and outputs a stability flag.

[0077] It should be noted that, in determining the overall deviation norm threshold, heat load index threshold, and hysteresis coefficient threshold in this invention, the historical data used should be uniformly the dataset collected during the long-term stable operation of the calcining furnace and the period when the product quality meets the process standards. Selecting specific percentiles, such as the 90th, 85th, and 75th percentiles, as thresholds aims to ensure that the threshold level covers the vast majority of normal operating conditions, thereby setting the control trigger point at the edge of the normal fluctuation range. The specific value of the percentile can be adjusted according to the enterprise's need for a balance between control sensitivity and stability. Increasing the percentile will loosen the threshold and make the system adjustment smoother; decreasing the percentile will tighten the threshold and make the system response more sensitive. Those skilled in the art can select appropriate percentiles by analyzing the distribution of historical data according to actual process requirements.

[0078] Furthermore, the control method described in this invention is executed online in a periodic, rolling manner. In each complete control cycle, for example, with a cycle length of 30 seconds, the system executes the entire process from step S11 to step S18. At the beginning of each cycle, the latest real-time measurement data is acquired, such as temperature distribution and fuel injection amount. The intermediate states, such as the updated thermal resistance matrix and optimized allocation scheme output in the previous cycle, are saved and used as the initial values ​​or references for the calculation in the next cycle, thereby realizing the inheritance and iterative optimization of state information. The compensation input sequence generated in step S18 refers to a set of fuel flow setpoint instructions that are scheduled in advance in time and will be issued in the current cycle. The amount of advance is determined by the compensation time, thereby offsetting the system lag.

[0079] Reference Figure 2 The second embodiment of the present invention provides a petroleum coke calcining furnace temperature coordinated control system, comprising: The data acquisition module is used to acquire temperature distribution, fuel dosage, material movement speed, and convective radiation values. The prediction module is used to predict the furnace temperature based on the temperature distribution and the fuel dosage using a pre-built temperature prediction model, and obtain a predicted temperature sequence. The correction module is used to extract local anomaly features based on the predicted temperature sequence and the preset target temperature distribution, obtain the maximum deviation vector, and perform fuzzy logic incremental correction based on the maximum deviation vector to obtain the adjusted dosage. The matching module is used to perform heat deviation analysis by constructing a theoretical heat release curve and an actual heat release curve based on the adjusted dosage, to obtain a heat deviation vector, and to perform heat transfer characteristic matching based on the heat deviation vector to obtain a heat transfer correction coefficient. The optimization module is used to perform dynamic thermal resistance correction based on the heat transfer correction coefficient, obtain an updated thermal resistance matrix, and perform online overburn risk assessment based on the updated thermal resistance matrix to generate an optimized allocation scheme. The adjustment module is used to adjust the position of the combustion flame according to the optimized allocation scheme, obtain the adjusted flame position, and perform dynamic heat absorption simulation of the material based on the adjusted flame position and the material movement speed to determine the change in heat absorption rate. The compensation module is used to perform pure time lag effect diagnosis based on the change in heat absorption rate and the convective radiation value, obtain the compensation time, and perform dynamic time compensation based on the compensation time to generate a compensation input sequence. The output module is used to perform instruction coordination and arbitration based on the compensation input sequence, generate a low-level execution signal, and perform closed-loop furnace temperature control based on the low-level execution signal to output a stable furnace temperature state.

[0080] It should be noted that the petroleum coke calcining furnace temperature coordinating control system provided in this embodiment of the invention is used to execute all the process steps of the petroleum coke calcining furnace temperature coordinating control method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0081] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0082] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for coordinated temperature control of a petroleum coke calcining furnace, characterized in that, include: Acquire temperature distribution, fuel dosage, material movement velocity, and convective radiation values; Based on the temperature distribution and the fuel dosage, the furnace temperature is predicted using a pre-built temperature prediction model to obtain a predicted temperature sequence. Based on the predicted temperature sequence and the preset target temperature distribution, local anomaly features are extracted to obtain the maximum deviation vector, and fuzzy logic incremental correction is performed based on the maximum deviation vector to obtain the adjusted dosage. Based on the adjusted dosage, heat deviation analysis is performed by constructing theoretical heat release curves and actual heat release curves to obtain heat deviation vectors. Then, heat transfer characteristics are matched based on the heat deviation vectors to obtain heat transfer correction coefficients. Based on the heat transfer correction coefficient, dynamic thermal resistance correction is performed to obtain an updated thermal resistance matrix. Based on the updated thermal resistance matrix, an online overburn risk assessment is performed to generate an optimized allocation scheme. According to the optimized allocation scheme, the position of the combustion flame is adjusted to obtain the adjusted flame position, and the dynamic heat absorption of the material is simulated based on the adjusted flame position and the material movement speed to determine the change in the heat absorption rate. Based on the change in the heat absorption rate and the convective radiation value, a pure time lag effect diagnosis is performed to obtain the compensation time, and dynamic time compensation is performed based on the compensation time to generate a compensation input sequence. Based on the compensation input sequence, instruction coordination and arbitration are performed to generate a low-level execution signal, and closed-loop furnace temperature control is performed based on the low-level execution signal to output a stable furnace temperature state.

2. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 1, characterized in that, The step of predicting the furnace temperature based on the temperature distribution and the fuel dosage using a pre-built temperature prediction model to obtain a predicted temperature sequence includes: The temperature distribution is mapped into a spatial distribution matrix to obtain the spatial temperature matrix; Multiply the fuel dosage by a preset unit calorific value to obtain the heat input variable; Based on the spatial temperature matrix and the heat input variables, vector concatenation is performed to obtain the system state vector; Based on the system state vector, the furnace temperature is predicted using a pre-built temperature prediction model to obtain a predicted temperature sequence, wherein the pre-built temperature prediction model includes the regional thermal resistance matrix inside the furnace.

3. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 2, characterized in that, The process of constructing the temperature prediction model includes: Obtain the historical system state vector and historical temperature sequence; The historical system state vector is input into the input layer of the initially constructed neural network model for training, and the predicted temperature sequence output by the output layer of the neural network model is obtained. Substitute the predicted temperature sequence and the historical temperature sequence into the loss function to calculate the loss value; The gradient of the output layer of the neural network model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data. Based on the gradient data and the preset learning rate, the parameters of each layer of the neural network model are updated using the gradient descent method. The parameters of each layer are iteratively updated until the number of training iterations of the neural network model is greater than a preset number of iterations, or the loss value of the neural network model is less than a preset loss threshold. At this point, the training is considered complete, and the trained temperature prediction model is obtained.

4. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 1, characterized in that, The step of extracting local anomaly features based on the predicted temperature sequence and the preset target temperature distribution to obtain the maximum deviation vector, and then performing fuzzy logic incremental correction based on the maximum deviation vector to obtain the adjusted dosage, includes: Based on the predicted temperature sequence and the preset target temperature distribution, point-by-point difference calculation is performed to obtain the temperature deviation matrix; Calculate the Euclidean norm of the temperature deviation matrix to obtain the overall deviation norm; When the overall deviation norm exceeds the preset overall deviation norm threshold, the data of the continuous region with the largest absolute value in the temperature deviation matrix is ​​extracted to form the maximum deviation vector; When the overall deviation norm does not exceed the preset overall deviation norm threshold, the fuel dosage remains unchanged; Based on the maximum deviation vector and the preset deviation correction rule base, fuzzy inference calculation is performed using a fuzzy logic control algorithm based on triangular membership function and centroid method to obtain the fuel correction increment. The adjusted dosage is obtained by performing a weighted summation calculation based on the fuel correction increment and the fuel addition amount.

5. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 1, characterized in that, The process involves adjusting the dosage, constructing a theoretical heat release curve, and comparing it with the actual heat release curve to analyze the heat deviation, obtaining a heat deviation vector. Then, based on this heat deviation vector, heat transfer characteristics are matched to obtain a heat transfer correction coefficient, including: Fuel supply is adjusted according to the aforementioned dosage, furnace radiation intensity is obtained, and thermal inversion calculations are performed based on a multispectral temperature measurement inversion algorithm using the least squares method to obtain real-time heat release curves. Calculate the heat accumulation of the adjusted dosage within a preset time window, generate a theoretical heat release curve, and align it with the real-time heat release curve to obtain an aligned theoretical curve and an aligned real-time curve. The heat deviation vector is generated by calculating the point-by-point difference between the alignment theoretical curve and the alignment real-time curve. Based on the heat deviation vector, multi-scale wavelet decomposition is performed using the Mallat algorithm to obtain the nonlinear fluctuation component; Based on the nonlinear fluctuation component, heat transfer characteristics are matched using a preset heat transfer attenuation coefficient table to obtain the heat transfer correction coefficient.

6. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 1, characterized in that, The process of performing dynamic thermal resistance correction based on the heat transfer correction coefficient to obtain an updated thermal resistance matrix, and then performing online overheating risk assessment based on the updated thermal resistance matrix to generate an optimized allocation scheme includes: Multiply the heat transfer correction coefficient by the regional thermal resistance matrix of the temperature prediction model to obtain the updated thermal resistance matrix; Based on the updated thermal resistance matrix, the heat conduction is solved using a finite volume algorithm to obtain the nodal temperature vector; Based on the node temperature vector, the net radiative heat transfer is calculated using the radiation angle coefficient algorithm to obtain the heat load index; When the heat load index does not exceed the preset heat load index threshold, the fuel distribution ratio is maintained; When the heat load index exceeds the preset heat load index threshold, the fuel allocation ratio is reduced to generate an optimized allocation scheme.

7. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 1, characterized in that, The step of adjusting the combustion flame position according to the optimized allocation scheme to obtain the adjusted flame position, and performing dynamic heat absorption simulation of the material based on the adjusted flame position and the material movement speed to determine the change in the heat absorption rate, includes: Based on the fuel allocation ratio of each loop in the optimized allocation scheme and the corresponding nozzle position coordinates of each loop, a weighted average calculation is performed to obtain the adjusted flame position; Based on the adjusted flame position, a three-dimensional radiation field is reconstructed using the Monte Carlo ray tracing algorithm to generate a heat flux density matrix. Based on the heat flux density matrix and the material velocity, a motion trajectory model is constructed by numerical integration of the motion trajectory using the fourth-order Runge-Kutta algorithm. Based on the motion trajectory model and the heat flux density matrix, path heat absorption is calculated through numerical integration and differentiation to confirm the change in heat absorption rate.

8. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 1, characterized in that, The process involves diagnosing the pure time lag effect based on the change in the heat absorption rate and the convective radiation value, obtaining the compensation time, and performing dynamic time compensation based on the compensation time to generate a compensation input sequence, including: Based on the change in the heat absorption rate and the convective radiation value, time alignment is performed to generate a comprehensive state vector; Based on the comprehensive state vector, the system lag characteristics are identified through impulse response analysis to obtain the lag time and the dominant time. The lag time is then divided by the dominant time to obtain the lag influence coefficient. When the lag effect coefficient exceeds the preset lag effect coefficient threshold, the lag time is multiplied by the preset compensation coefficient to obtain the compensation time; If the hysteresis coefficient does not exceed the preset hysteresis coefficient threshold, the adjusted dosage remains unchanged; The compensation time is superimposed on the time axis of the adjusted dosage to generate a compensation input sequence.

9. The method for coordinated temperature control of a petroleum coke calcining furnace according to claim 1, characterized in that, The step of performing instruction coordination and arbitration based on the compensated input sequence, generating a low-level execution signal, and performing closed-loop furnace temperature control based on the low-level execution signal to output a stable furnace temperature state includes: Based on the compensation input sequence, the instructions are decomposed using a pre-built instruction hierarchical mapping rule to obtain coordinated control instructions; The coordination control command is input into a preset exception rule base for conflict arbitration to generate a low-level execution signal; Based on the underlying execution signal, the petroleum coke calciner is driven to perform closed-loop furnace temperature control and output a stable furnace temperature state.

10. A system for coordinated temperature control of a petroleum coke calcining furnace, characterized in that, include: The data acquisition module is used to acquire temperature distribution, fuel dosage, material movement speed, and convective radiation values. The prediction module is used to predict the furnace temperature based on the temperature distribution and the fuel dosage using a pre-built temperature prediction model, and obtain a predicted temperature sequence. The correction module is used to extract local anomaly features based on the predicted temperature sequence and the preset target temperature distribution, obtain the maximum deviation vector, and perform fuzzy logic incremental correction based on the maximum deviation vector to obtain the adjusted dosage. The matching module is used to perform heat deviation analysis by constructing a theoretical heat release curve and an actual heat release curve based on the adjusted dosage, to obtain a heat deviation vector, and to perform heat transfer characteristic matching based on the heat deviation vector to obtain a heat transfer correction coefficient. The optimization module is used to perform dynamic thermal resistance correction based on the heat transfer correction coefficient, obtain an updated thermal resistance matrix, and perform online overburn risk assessment based on the updated thermal resistance matrix to generate an optimized allocation scheme. The adjustment module is used to adjust the position of the combustion flame according to the optimized allocation scheme, obtain the adjusted flame position, and perform dynamic heat absorption simulation of the material based on the adjusted flame position and the material movement speed to determine the change in heat absorption rate. The compensation module is used to perform pure time lag effect diagnosis based on the change in heat absorption rate and the convective radiation value, obtain the compensation time, and perform dynamic time compensation based on the compensation time to generate a compensation input sequence. The output module is used to perform instruction coordination and arbitration based on the compensation input sequence, generate a low-level execution signal, and perform closed-loop furnace temperature control based on the low-level execution signal to output a stable furnace temperature state.

Images