Decomposition furnace coal feeding quantity control system and method based on decomposition furnace outlet temperature prediction

Abstract

The invention relates to the technical field of cement firing, and discloses a decomposition furnace coal feeding quantity control system and method based on decomposition furnace outlet temperature prediction, a characteristic point standard set obtained based on a time-delay cross-correlation method is utilized to construct a historical training sample, a model is trained according to the historical training sample, and a decomposition furnace outlet temperature prediction model is obtained; establishing a mapping relation from the coal air combination factor to the outlet temperature of the future decomposition furnace; constructing a tailing coal feeding quantity control model to deduce a tailing coal feeding quantity given value of the decomposing furnace in the next time step on the basis of predicting the outlet temperature of the decomposing furnace; real-time data corresponding to the feature point standard set are obtained, and coal feeding amount adjustment is conducted according to the given value, obtained through the model, of the tailing coal feeding amount of the decomposing furnace in the next time step; and performing decomposition furnace outlet temperature prediction effect monitoring and tailing coal feeding quantity optimization effect monitoring according to the operation condition after the tailing coal feeding quantity is adjusted, and dynamically updating the decomposition furnace outlet temperature prediction model and the tailing coal feeding quantity control model according to the monitoring result.

Description

Technical Field

[0001] The invention relates to the technical field of cement burning, and in particular to a control system and method for a coal feeding amount of a decomposition furnace based on prediction of the outlet temperature of the decomposition furnace. Background Art

[0002] As one of the core technologies of new dry process cement production technology, the pre-decomposition step directly determines the decomposition rate of raw materials, and further determines the burden of rotary kiln clinker burning, thus affecting the quality of cement clinker. In addition, the coal consumption of this step generally accounts for about 60% of the coal required for the entire cement burning, which directly affects the coal consumption index per ton of clinker.

[0003] In order to make the outlet temperature feedback value of the decomposition furnace reach and stabilize at the set target value, the coal feeding amount of the decomposition furnace needs to be properly controlled. However, the internal structure of the decomposition furnace is complex, the physical and chemical reactions are intertwined, the nonlinear characteristics and random characteristics are obvious, and the multiple variables in the process affect each other, which brings challenges to the control of the coal feeding amount of the decomposition furnace: (1) The current method of controlling the coal feeding amount based on the real-time feedback value of the outlet temperature of the decomposition furnace does not fully consider the time delay of the physical and chemical reactions in the decomposition furnace. It is often adjusted only after the obvious deviation between the outlet temperature feedback value of the decomposition furnace and the target value is observed, resulting in a lag in the adjustment of the coal feeding amount of the decomposition furnace; (2) The current method of adjusting the coal feeding amount based on the prediction of the outlet temperature of the decomposition furnace still has room for improvement. The deviation between the predicted value of the outlet temperature of the decomposition furnace and the actual feedback value is still large, resulting in the accuracy of the coal feeding amount control of the decomposition furnace needs to be improved; (3) When the raw material feeding amount or air supply volume of the burning system undergoes a step adjustment, the coal feeding amount of the decomposition furnace is not accurate enough to track such changes, resulting in a short-term sharp rise or fall in the outlet temperature feedback value of the decomposition furnace, which increases the volatility of the outlet temperature of the decomposition furnace and affects the stability of the decomposition furnace working condition. Summary of the invention

[0004] The present invention aims to provide a decomposition furnace coal feeding control system and method based on the prediction of the decomposition furnace outlet temperature, constructs a decomposition furnace outlet temperature prediction model with more accurate prediction, and calculates in advance the gap between the future decomposition furnace outlet temperature and the outlet temperature target value to achieve the purpose of adjusting the coal feeding amount in advance. Through continuous monitoring and dynamic adjustment of the decomposition furnace outlet temperature prediction model and the coal feeding amount control model, the entire coal feeding amount control system has the adaptive ability to track changes in the firing system.

[0005] The basic scheme provided by the present invention is: a method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the outlet temperature of the decomposition furnace, the method comprising the following steps:

[0006] S100, obtaining a characteristic point standard set based on a time-delay cross-correlation method, obtaining corresponding historical data of the characteristic point standard set and performing preprocessing to obtain historical training samples;

[0007] S200, using historical training samples to perform model training, obtain a decomposition furnace outlet temperature prediction model, so as to establish a mapping relationship from the coal-air combination factor to the future decomposition furnace outlet temperature;

[0008] S300, based on the deviation distribution statistics and deviation trend statistics between the predicted value and the target value of the outlet temperature of the calciner, combined with the step mutation statistics of the raw meal feeding amount, a tail coal feeding amount control model is constructed to infer the given value of the tail coal feeding amount of the calciner in the next time step based on the prediction of the outlet temperature of the calciner;

[0009] S400, after the system is deployed, the data corresponding to the characteristic point standard set is obtained in real time, and the tail coal feeding amount of the decomposition furnace is obtained by using the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model, and the given value of the tail coal feeding amount of the decomposition furnace in the next time step is input into the distributed control system for execution to adjust the tail coal feeding amount of the decomposition furnace;

[0010] S500, during the operation of the system, the decomposition furnace outlet temperature prediction effect and the tail coal feeding amount optimization effect are monitored according to the operating conditions after the tail coal feeding amount is adjusted, and the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model are dynamically updated with the monitoring results.

[0011] In order to implement the method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the outlet temperature of the decomposition furnace, a method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the outlet temperature of the decomposition furnace is also provided; the system comprises:

[0012] A historical training sample acquisition unit is used to obtain a characteristic point standard set based on a time-delay cross-correlation method, obtain corresponding historical data of the characteristic point standard set and perform preprocessing to obtain historical training samples;

[0013] The decomposition furnace outlet temperature prediction model establishment unit is used to use historical training samples to perform model training to obtain a decomposition furnace outlet temperature prediction model to establish a mapping relationship from the coal-air combination factor to the future decomposition furnace outlet temperature;

[0014] The tail coal feeding amount control model establishment unit is used to construct a tail coal feeding amount control model based on the deviation distribution statistics and deviation trend statistics between the predicted value and the target value of the decomposition furnace outlet temperature, as well as the step mutation statistics of the raw meal feeding amount, so as to infer the tail coal feeding amount set value of the decomposition furnace in the next time step based on the prediction of the decomposition furnace outlet temperature;

[0015] The system operation unit is used to obtain the data corresponding to the characteristic point standard set in real time, and obtain the tail coal feeding amount of the decomposition furnace at the next time step by using the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model, and input the given value into the distributed control system for execution to adjust the tail coal feeding amount of the decomposition furnace;

[0016] The model dynamic update unit is used to monitor the decomposition furnace outlet temperature prediction effect and the tail coal feeding amount optimization effect according to the operating conditions after the tail coal feeding amount is adjusted during the system operation, and dynamically update the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model based on the monitoring results.

[0017] The working principle and advantages of the present invention are:

[0018] Firstly, an accurate prediction model for the outlet temperature of the decomposition furnace is established, which follows the changes in operating conditions. Then, based on the deviation between the predicted temperature value and the target value, as well as the step change in the raw meal feeding amount, a tail coal feeding amount control model is established. The feeding amount is optimized based on the deviation between the predicted value and the target value of the outlet temperature and the step change in the raw meal feeding amount, so that the feedback value of the outlet temperature of the decomposition furnace in the future is close to and stable at the preset target value.

[0019] Compared with the existing technology, this solution fully considers the time delay characteristics of the decomposition furnace combustion process. From the entire process of factor selection, model establishment and model update, the predicted value, target value and feedback value of the outlet temperature are used to build the corresponding data relationship based on the time delay characteristics. While improving the accuracy of the decomposition furnace outlet temperature prediction, it can accurately and timely adjust the coal feeding amount, so that the entire system has the adaptive ability to track changes in the burning system.

[0020] This scheme obtains the main factors and delays that affect the decomposition furnace outlet temperature through mechanism analysis, and establishes a more accurate prediction model for the decomposition furnace outlet temperature in combination with a special architecture. In actual application, the average absolute error between the temperature prediction value and the actual feedback value is no more than 2°C, which improves the prediction effect of the decomposition furnace outlet temperature in the cement burning system; based on the deviation distribution statistics and deviation trend statistics between the decomposition furnace outlet temperature prediction value and the target value, as well as the raw material feed amount step mutation statistics, the change of the coal feeding amount is jointly controlled, taking into account both the static deviation value and the dynamic deviation change trend, and the raw material feed amount step mutation factor, so that the change of the coal feeding amount is more targeted and in line with the combustion regulation delay operation characteristics; at the same time, based on the decomposition furnace outlet temperature prediction value, target value and feedback value, the effect of the decomposition furnace outlet temperature prediction model and the coal feeding amount control model is monitored to achieve fine-tuning and updating of the model, so that the model has adaptive adjustment capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a design diagram of the control system architecture of the coal feeding amount of the decomposition furnace based on the prediction of the decomposition furnace outlet temperature provided by an embodiment of the present invention;

[0022] Figure 2 A schematic flow chart of a method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature provided by an embodiment of the present invention;

[0023] Figure 3 This is a relationship diagram between coal air and decomposition furnace outlet temperature in an existing cement burning system;

[0024] Figure 4 A structural diagram of a decomposition furnace outlet temperature prediction model provided by an embodiment of the present invention;

[0025] Figure 5 A flow chart of fine-tuning the decomposition furnace outlet temperature prediction model provided in an embodiment of the present invention;

[0026] Figure 6 A flow chart of a dynamic update of a coal feeding control model for a calciner provided in an embodiment of the present invention;

[0027] Figure 7 A schematic diagram of the structure of a decomposition furnace coal feeding control system based on decomposition furnace outlet temperature prediction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following is a further detailed description through specific implementation methods:

[0029] The embodiment is basically as shown in the attached Figure 1 , Figure 2 As shown in the figure: This scheme follows the principle of model predictive control and uses the data collected by the cement burning distributed control system DCS (Distributed Control System) to establish a decomposition furnace coal feeding control method based on the prediction of the decomposition furnace outlet temperature. The scheme is generally divided into a training stage, a reasoning stage and an updating stage: the training stage focuses on the construction and optimization of the decomposition furnace outlet temperature prediction model, the reasoning stage focuses on the design of the decomposition furnace coal feeding adjustment model, and the updating stage focuses on dynamically updating the decomposition furnace outlet temperature prediction model and the coal feeding control model according to the actual commissioning effect, so that the entire system has the adaptive ability to track changes in the burning system.

[0030] Specifically, the method comprises the following steps:

[0031] S100, obtaining a standard set of feature points based on a time-delay cross-correlation method, obtaining historical data corresponding to the standard set of feature points and performing preprocessing to obtain historical training samples.

[0032] Specifically, in the training stage, in order to establish an accurate prediction model for the decomposition furnace outlet temperature, the main influencing factors of the decomposition furnace outlet temperature feedback value are first analyzed to determine the input feature points.

[0033] The outlet temperature of the decomposition furnace is the product of the combined effect of the tail coal feed amount, raw meal feed amount, tertiary air temperature and other coal-air combined factors, among which the raw meal feed amount and tail coal feed amount are the most important. Figure 3As shown in the figure, the interdependence of "coal, material and air" is explained, where the single arrow line represents the dependence of the decomposition furnace outlet temperature feedback value on factors such as "coal, material and air", and the double arrow line represents the dependence of the tail coal feeding amount on the decomposition furnace outlet temperature target value and "material and air" and other factors.

[0034] The outlet temperature feedback value of the decomposition furnace is a representation of the gas heat after each physical and chemical reaction process in the decomposition furnace. From the perspective of combustion and heat exchange, the main factors affecting the outlet temperature feedback value of the decomposition furnace are the following two categories: (1) The type of heat provided, including the amount of tail coal fed. The flameless combustion of tail coal is the main source of heat in the decomposition furnace; the tertiary air temperature. The tertiary air brings part of the clinker heat recovered by the grate cooler; the gas temperature of the kiln tail smoke chamber. The gas in the kiln tail smoke chamber brings part of the waste heat after the rotary kiln is fired; the C5 discharge pipe temperature, which indicates the degree of preheating of the raw material entering the decomposition furnace; Generally, under the condition of keeping other factors unchanged, the above factors are positively correlated with the outlet temperature feedback value of the decomposition furnace. (2) The type of heat consumption, mainly the amount of raw material entering the decomposition furnace. The decomposition of raw material is an endothermic reaction and is the main heat consumption process in the decomposition furnace. In addition, the value of the high-temperature fan frequency determines the overall air volume of the firing system and also affects the combustion of pulverized coal in the decomposition furnace. If the high-temperature fan frequency is too low, the oxygen supply is insufficient, resulting in incomplete combustion and possible reduction of the atmosphere. If it is too high, it will lead to excess air, incomplete combustion, and reduce the thermal efficiency of the decomposition furnace. Finally, the relevant operating data in the rotary kiln and grate cooler, such as the rotary kiln current, the kiln head fire-watching TV temperature, the grate pressure under the grate of the grate cooler, etc., may also have an uncertain impact on the feedback value of the decomposition furnace outlet temperature.

[0035] After the above mechanism analysis, this scheme first selects characteristic points 1 to 17 as shown in Table 1 to further analyze their relationship with the feedback value of the decomposition furnace outlet temperature and screen reasonable characteristic points.

[0036] Serial number Point Description Data Unit 1 Feedback value of raw material feed measured by rotor scale Tons / hour (t / h) 2 Feedback value of coal feeding amount to the calciner measured by rotor scale Tons / hour (t / h) 3 Feedback value of outlet pressure of magnetic suspension blower for coal feeding in calciner Pa 4 Feedback value of pressure of coal feeding and injection pipe of calciner Pa 5 Three-stage air temperature feedback value Celsius (℃) 6 Tertiary air branch pipe 1 pressure feedback value Pa 7 Tertiary air branch pipe 2 pressure feedback value Pa 8 Feedback value of gas temperature in the kiln tail smoke chamber Celsius (℃) 9 Gas pressure feedback value of kiln tail smoke chamber Pa 10 Preheater C5A feed pipe temperature feedback value Celsius (℃) 11 Preheater C5B feed pipe temperature feedback value Celsius (℃) 12 High temperature fan frequency feedback value percentage(%) 13 Rotary kiln current feedback value An (A) 14 Fire temperature of kiln head fire TV Celsius (℃) 15 Watch the fire on the kiln head and the clinker temperature Celsius (℃) 16 Pressure under the left grate of the grate bed of the grate cooler Millibar (mbar) 17 Pressure under the right side of the grate bed of the grate cooler Millibar (mbar) 18 Decomposition furnace outlet temperature feedback value Celsius (℃) 19 Decomposition furnace outlet temperature target value Celsius (℃)

[0037] Since the combustion in the decomposition furnace is a time-continuous reaction process, there are time delays of different scales from raw material feeding, tail coal feeding, gas entry, flameless combustion to the arrival of combustion exhaust gas at the decomposition furnace outlet. However, traditional signal correlation analysis methods, such as Pearson coefficient and Spearman coefficient, statically calculate the global synchronization of two variables without considering the chronological relationship between variables. Therefore, they cannot meet the needs of this solution for time delay mining of various data in the combustion process of the decomposition furnace.

[0038] In order to overcome the defects of the conventional method, this scheme adopts the correlation calculation method based on time-delay cross-correlation to analyze the relationship between each characteristic point and the feedback value of the decomposition furnace outlet temperature: Let p t represents the feedback value sequence of the decomposition furnace outlet temperature over a period of time, wt Represents the feedback value sequence of a certain feature point in the same time period. The original data sequence of the feature point is moved by using different offsets τ in turn to obtain the shifted data sequence w t+τ , and calculate the Pearson correlation coefficient c between the offset data sequence and the decomposition furnace outlet temperature feedback value sequence τ . By comparing the correlation coefficients under different offsets, the correlation degree of the two signals can be explored, and the offset that produces the maximum correlation can also be found. The offset value can be approximated as the time delay estimate of the two signals. From the process flow analysis, it can be seen that there is a causal relationship between the feedback value of each feature point and the feedback value of the outlet temperature of the decomposition furnace, so the offset of each feature point data should be negative. This scheme is based on selecting feature points with obvious correlation, that is, selecting points with a bell-shaped Pearson coefficient distribution under different offset conditions. For feature points 1-17 in Table 1, the cross-correlation between their data and the feedback value of the outlet temperature of the decomposition furnace under different time lags is calculated. It is found that the pressure under the grate on the left side of the grate bed of the grate cooler and the pressure under the grate on the right side of the grate bed of the grate cooler have a low correlation with the outlet temperature of the decomposition furnace. Therefore, the feature points 13-17 in Table 1 are removed, and finally the feature points 1-12 in Table 1 are selected to form a standard set of feature points as model input parameters.

[0039] The corresponding historical data of the feature point standard set is obtained and preprocessed so that the constructed historical training samples meet the requirements of model training.

[0040] Specifically, the historical data is reasonably acquired according to the actual system operation time, and 2 years may be preferably selected. In this embodiment, data from 2022.08.01 to 2024.08.01 are acquired.

[0041] Since the meaning, dimension, and range of each feature point data are not consistent, and some data are abnormal, they need to be preprocessed before use, which mainly includes the following two steps:

[0042] (1) Using the box plot method to handle outliers: The outlier detection method based on the 3σ rule of normal distribution or z-score is based on the assumption that the data obeys the normal distribution. However, actual data often does not strictly obey the normal distribution. The standard for judging outliers is based on the mean and standard deviation of the data. However, the mean and standard deviation have low tolerance and outliers themselves will have a greater impact on them. Obviously, the effectiveness of using this method to judge outliers in non-normal distribution data is limited. The box plot method relies on actual data. On the one hand, it does not need to assume in advance that the data obeys a specific distribution form and does not make any restrictive requirements on the data. It can truly and intuitively show the original appearance of the data shape. On the other hand, the standard for judging outliers in the box plot is based on quartiles and interquartile ranges. Quartiles have a certain tolerance to outliers, and the result of identifying outliers is relatively objective. Therefore, this scheme adopts the box plot method, and counts the lower quartile Q1 and upper quartile Q3 of each feature point data respectively, calculates the interquartile range IQR=Q3-Q1, and regards the data with values ​​outside the range of [Q1-1.5IQR,Q3+1.5IQR] as outliers. There are two cases for outliers: directly delete the outliers that appear continuously, and replace them with the mean of the two adjacent time steps for the outliers that appear occasionally.

[0043] (2) Data normalization: The dimensions and value ranges of each feature point are different. Directly using the original data for modeling will cause the model to be affected by the value range of the value itself. Therefore, it is necessary to normalize the data of each point to the same measurement scale. This solution uses the maximum and minimum normalization method to normalize the data of each point to the range of [0,1]. The normalization formula is as follows:

[0044] Normalized value = (original value - minimum value) / (maximum value - minimum value)

[0045] The corresponding data of the standard set of feature points No. 1-12 selected in Table 1 are selected from the full amount of data sampled each time to form a data vector, and the data during the kiln shutdown period is excluded. After the above-mentioned two-step preprocessing of outliers and normalization, the historical training samples are obtained, and the training set, validation set and test set are divided according to the ratio of 7:1:2.

[0046] S200, using historical training samples to perform model training, obtain a decomposition furnace outlet temperature prediction model, so as to establish a mapping relationship from the coal-air combination factor to the future decomposition furnace outlet temperature.

[0047] Specifically, through deep learning and mining of historical training samples, an accurate mapping from coal-air combination factors to the future decomposition furnace outlet temperature is established. The mathematical expression of the decomposition furnace outlet temperature prediction model is shown in the following formula:

[0048] y t =f(xt-L ,…,x t-1 ,x t )

[0049] Among them, x t The characteristic vector of the coal-air combination factor corresponding to the current time step t can be simply expressed as x t =(m t ,r t ,w t ), m t is the tail coal feeding amount at time step t, r t is the raw material feed amount at time step t, w t is the system wind and other variables at time step t, L represents the length of the historical feature sequence that needs to be introduced in addition to the features of the current time step t, f represents the nonlinear mapping, and y t A label sequence representing the predicted values ​​of the decomposition furnace outlet temperature at a number of future time steps relative to the current time step t.

[0050] Since there are different time delay relationships from factors such as "coal wind" to the feedback value of the decomposition furnace outlet temperature, there is a time limit for the model to accurately predict the future decomposition furnace outlet temperature. In theory, it is impossible to accurately predict the future outlet temperature that exceeds the shortest time extension. In this embodiment, the characteristic point with the shortest time delay with the decomposition furnace outlet temperature feedback value is the decomposition furnace coal feeding and coal injection pipe pressure feedback value, and the shortest time delay is 64 seconds (sampling frequency is once every 2 seconds, that is, 32 time steps). The prediction time length cannot exceed the shortest time delay to ensure that the data collection is reasonable. Therefore, in this embodiment, the decomposition furnace outlet temperature prediction model sets the prediction time step length of the future outlet temperature to 30, that is, the decomposition furnace outlet temperature is predicted for the next 1 minute, y t A label sequence that can represent the decomposition furnace outlet temperature feedback values ​​for the next 30 time steps t+1, t+2, …, t+30.

[0051] For a better understanding, it is necessary to explain that there are three temperature values ​​in the whole process: the predicted value, which is the "guessed temperature value" after the model receives the working condition characteristics and performs calculations; the feedback value, which is the actual temperature value during the system operation; the target value, which is the temperature target value set for the system. When training the model, because the future real feedback value has not yet been generated, and the feedback value is needed as a label to guide the model to model the mapping relationship from working condition characteristics to temperature, the historical data is used to complete the model training. If the model is well trained, the predicted value can be approximated as the feedback value in the inference stage, and the future temperature situation can be predicted in advance. After the model is trained and deployed to the system, and the real feedback value is generated, the real feedback value is used for the subsequent process.

[0052] The structure of the decomposition furnace outlet temperature prediction model in this scheme adopts the iTransformer architecture, which is currently the best architecture in time series prediction. It uses an encoder consisting of an embedding layer, a multi-layer self-attention layer, a normalization layer, and a forward calculation layer to extract deep self-attention representations from the current and historical "coal wind" and other operating condition feature sequences, and directly outputs the predicted value of the decomposition furnace outlet temperature in the future several time steps through a fully connected layer.

[0053] The structure of the prediction model for the outlet temperature of the decomposition furnace is as follows: Figure 4 As shown in the figure, the "embedding layer" in the model performs a linear transformation on the vector formed by the features of each measurement point at the time step. The "self-attention layer" calculates the relationship between the features after embedding and obtains the deep attention features by weighted summation of the features. The function of the "normalization layer" is to scale the feature values ​​of each sample according to the mean and standard deviation. The "forward calculation layer" performs a more complex nonlinear transformation on the input through a two-layer fully connected network with a Relu activation function, such as Figure 4 As shown, a module is composed of a self-attention layer, a normalization layer, a forward calculation layer, and a normalization layer in sequence. The module is stacked several times in the same form to form the backbone of the prediction model. The fully connected output layer directly predicts the decomposition furnace outlet temperature of several future time steps through several fully connected neurons. In this embodiment, it is found that the benefits brought by stacking more than 6 times are not obvious, and the effect of the decomposition furnace outlet temperature prediction model is not significantly improved. Considering the limitations of computing resources and training time, 6 stackings are selected, and 30 fully connected neurons are used to directly predict the decomposition furnace outlet temperature of the next minute (corresponding to the next 30 time steps).

[0054] The training loss function of the model is the mean square error (MSE) between the predicted value and the actual feedback value of the decomposition furnace outlet temperature. temp The evaluation index of the model prediction effect is the mean absolute error MAE temp , let y′ t is the outlet temperature of the decomposition furnace predicted by the model at time step t in the next 1 minute, then the mean square error MSE predicted by the model is temp And the mean absolute error MAE temp The definition is as follows, where ‖·‖ 1 represents the 1-norm, ‖·‖ 2 represents the 2-norm.

[0055] MSE temp =‖y t -y′ t ‖ 2 / 60

[0056] MAE temp =‖y t -y′ t ‖1 / 60

[0057] S300, based on the deviation distribution statistics and deviation trend statistics between the predicted value and the target value of the decomposition furnace outlet temperature, as well as the step mutation statistics of the raw meal feeding amount, a tail coal feeding amount control model is constructed to infer the given value of the tail coal feeding amount of the decomposition furnace in the next time step based on the prediction of the decomposition furnace outlet temperature.

[0058] Specifically, the algorithm process is as follows:

[0059] (1) At time step t, the operating condition feature matrix x of the current and historical L time steps is t-L ,…,x t-1 ,x t Input it into the trained decomposition furnace outlet temperature prediction model f, and output the predicted value y′ of the decomposition furnace outlet temperature in the next minute t ;

[0060] (2) Order Represents the target value of the decomposition furnace outlet temperature, and calculates the deviation sequence between the temperature prediction value and the target value Then calculate the deviation distribution index of the deviation sequence, that is, the mean μ of the statistical deviation sequence t and standard deviation σ t ; Calculate the deviation change trend index, that is, use the least squares method to fit the deviation sequence to obtain the slope k representing the deviation change trend t ;

[0061] (3) Let the tail coal feeding amount feedback value at the current time step t be m t , the given value of the coal feeding amount of the calciner in the next time step is recorded as The given value of the decomposition furnace tail coal in the next time step is calculated according to the following formula:

[0062]

[0063] Among them, θ β ,θ γ To adjust the step size hyperparameter; in this embodiment, during the model training phase, θ β The preset initial value is 0.01, θ γ The preset initial value is 0.1. The initial value can be determined based on historical data mining or expert experience, and dynamically adjusted according to the monitoring effect after operation.

[0064] α t , β t , γ t Calculate the updated coefficients for the adaptive condition, where α t The coefficient that represents the adjustment of tail coal feed amount due to a sudden step change in raw meal feed amount is set according to the following formula:

[0065]

[0066] β t To characterize the coefficient of adjusting the tail coal feeding amount according to the deviation distribution index between the predicted value and the target value of the decomposition furnace outlet temperature, the value of this item is related to the mean value of the deviation μ t The absolute value of is proportional to the standard deviation σ t The adjustment direction of the coal feeding amount is inversely proportional to the deviation mean μ t The signs of β are opposite, then t The value of is set according to the following formula:

[0067]

[0068] γ t In order to characterize the coefficient of adjusting the tail coal feeding amount according to the deviation trend index between the predicted value and the target value of the decomposition furnace outlet temperature, when the slope of the deviation k t When the absolute value of is not higher than the slope threshold, this item is set to 0. If it is higher than the slope threshold, it means that the deviation between the predicted value of the decomposition furnace outlet temperature and the target value has a significant expansion trend. The value of this item is proportional to the absolute value of the slope, and the sign of this value is opposite to the sign of the slope. Then γ t According to the following formula, the slope threshold can be set to 0.2:

[0069]

[0070] S400, after the system is deployed, the corresponding data of the characteristic point standard set is obtained in real time, and the tail coal feeding amount of the decomposition furnace is given by the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model, and the given value is input into the distributed control system to adjust the tail coal feeding amount of the decomposition furnace.

[0071] Specifically, after the system is put into operation, the real-time data corresponding to the characteristic point standard set is input into the decomposition furnace outlet temperature prediction model to obtain a sequence of decomposition furnace outlet temperature prediction values ​​for several future time steps, and the sequence of decomposition furnace outlet temperature prediction values ​​is input into the tail coal feeding amount control model, and according to the algorithm in S300, a given value of the decomposition furnace tail coal feeding amount for the next time step is obtained, and the decomposition furnace tail coal feeding amount is adjusted according to this value.

[0072] S500, during the operation of the system, the decomposition furnace outlet temperature prediction effect and the tail coal feeding amount optimization effect are monitored according to the operating conditions after the tail coal feeding amount is adjusted, and the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model are dynamically updated with the monitoring results.

[0073] Specifically, Figure 5As shown, the decomposition furnace outlet temperature prediction effect is monitored in the update phase to determine whether the average absolute error between the predicted value given by the decomposition furnace outlet temperature prediction model and the decomposition furnace outlet temperature feedback value exceeds a first threshold value to monitor the decomposition furnace outlet temperature prediction effect.

[0074] During the operation of the system, the DCS system can obtain the feedback value of the decomposition furnace outlet temperature, compare the feedback value with the predicted value given by the decomposition furnace outlet temperature prediction model, and calculate the mean absolute error (MAE) _FK_YC Using the mean absolute error MAE _FK_YC The value of real-time monitoring of the decomposition furnace outlet temperature prediction model; if the mean absolute error MAE _FK_YC is not greater than the first threshold, it means that the prediction effect is in line with expectations and the model does not need to be adjusted; if the mean absolute error MAE _FK_YC If it is greater than the first threshold, it means that the prediction effect does not meet expectations and the model needs to be adjusted. Specifically, the operating data of the time step corresponding to the decomposition furnace outlet temperature feedback value and its forward L time steps are regarded as negative samples. The negative samples are distributed differently from the training samples of the decomposition furnace outlet temperature prediction model, such as Figure 5 As shown in the figure, the parameters of the decomposition furnace outlet temperature prediction model can be fine-tuned by using a forward inference of a negative sample and a reverse feedback of a gradient, and the fine-tuned new model can replace the old model for the prediction inference of the next time step. Such an update design can make the prediction model adaptable to the changes in the working conditions of the firing system or the environment. The fine-tuning process is specifically a training process of the neural network parameters: forward inference inputs a sequence of feature vectors, and uses the model to calculate the output, that is, the future decomposition furnace outlet temperature prediction value; reverse feedback first calculates the mean square error loss MSE between the prediction value and the true feedback value, and then feeds the loss back. During the feedback process, the gradient of the loss value relative to the model parameter is calculated, and each parameter is updated in the opposite direction of the gradient. Finally, the purpose of updating the model as a whole is achieved.

[0075] like Figure 6 As shown in the figure, the optimization effect of tail coal feeding amount is monitored in the updating stage. During the operation of the system, the mean absolute error (MAE) between the feedback value of the decomposition furnace outlet temperature and the target value is calculated. _FK_MB Whether it exceeds the second threshold is used to monitor the optimization effect of the tail coal feeding amount.

[0076] If the mean absolute error MAE _FK_MB is not greater than the second threshold, it indicates that the given coal feeding amount is in line with expectations. _FK_MBIf it is greater than the second threshold, the optimization effect does not meet expectations, and the tail coal feeding amount control model is updated according to the monitoring results of the decomposition furnace outlet temperature prediction effect; because the tail coal feeding amount control model of this scheme is constructed based on the decomposition furnace outlet temperature prediction model, its optimization effect is not good, which may be because the prediction model is inaccurate, or the prediction model is accurate but the tail coal feeding amount control model itself is inaccurate. Therefore, it is necessary to check before adjusting the control model and evaluate the prediction model first. If the prediction model effect is not good, update the prediction model, otherwise the control model is inaccurate, and directly update the control model, thereby making the model tuning of this scheme more accurate.

[0077] Specifically, if MAE _FK_MB If the predicted value is greater than the second threshold, the effect of the decomposition furnace outlet temperature prediction model is evaluated first. If the predicted effect of the decomposition furnace outlet temperature does not meet expectations, the mean absolute error (MAE) between the predicted value and the feedback value of the decomposition furnace outlet temperature is _FK_YC If the value is greater than the first threshold, the decomposition furnace outlet temperature prediction model is fine-tuned and updated according to the above design, and the deviation sequence between the decomposition furnace outlet temperature prediction value and the target value at the current and past L time steps is obtained, and the mean value μ of the deviation distribution is calculated. t , standard deviation σ t and the slope value k of the deviation change t To update the β in the tail coal feed adjustment formula t With γ t , and at the same time obtain the current raw material feed quantity given value and its previous value to update the α in the tail coal feed quantity adjustment formula t , and update the tail coal feeding quantity control model.

[0078] If the predicted effect of the decomposition furnace outlet temperature is in line with expectations, it means that the decomposition furnace coal feeding control algorithm is inaccurate, and the adjustment step size hyperparameter in the coal feeding control model should be adjusted directly. Specifically, the adjustment is to use the least squares method to minimize the mean square error between the decomposition furnace outlet feedback value and the target value through linear fitting, so as to obtain a new adjustment step size hyperparameter θ β ,θ γ .

[0079] Among them, the first threshold and the second threshold are used in the algorithm operation monitoring stage, and are respectively used to evaluate the effects of the decomposition furnace outlet temperature prediction model and the coal feeding control model, that is, whether the corresponding absolute average error exceeds the above-mentioned corresponding thresholds to determine whether the model hyperparameters need to be adjusted. The specific implementation is the dynamic adjustment of the step size hyperparameter. The first threshold and the second threshold are only used as the benchmark for evaluating the model effect, and they themselves will not be adjusted within a certain period of time. The smaller the first threshold and the second threshold are set, the higher the requirements for the prediction model and the control model. Through in-depth analysis, the first threshold, the second threshold, and the slope threshold are set to initial values ​​based on experience at the beginning of the operation. When the algorithm is actually running and reasoning, these thresholds are adjusted through a period of debugging to achieve the best operating condition; in this embodiment, the first threshold is obtained based on the statistics of historical data. Through the prediction of the historical training set by the prediction model, the average absolute deviation of the predicted value and the feedback value of the temperature can be obtained. The statistical mean is 2.7°C and the standard deviation is 0.4°C. The current first threshold is set to 3.9°C = mean + 3 times the standard deviation. According to the normal distribution theory, such a first threshold can theoretically ensure that about 99.87% of the predicted average absolute deviation is within 3.9°C; the second threshold is determined according to the operator's tolerance for the deviation between the feedback value and the target value of the decomposition furnace outlet temperature after the automatic tail coal control is put into use. The current setting of 1.5°C is the upper limit of the effect that can be achieved during manual continuous adjustment. The slope threshold is determined similarly to the second threshold. The threshold set in this way can enable the algorithm to reach the level of manual control when it is running at the current stage after it is put into use, meeting the current system operation requirements.

[0080] When the system is put into operation, the three thresholds will be adjusted during the long-term operation of the model to dynamically adapt to the current situation of the system. A mapping relationship can be constructed between the three thresholds and the degree of effect deviation in the model operation monitoring and the "tolerance" of the decomposition furnace outlet temperature increase and decrease rate, and the threshold adjustment time can be reasonably determined. The degree of effect deviation in the model operation monitoring and the decomposition furnace outlet temperature increase and decrease rate in the historical time period are analyzed, and the thresholds are adjusted based on the mapping relationship. Such dynamic adjustment of the thresholds can ensure that the monitoring dynamics of the model are in line with the current system operation, and the monitoring results are more accurate, while further improving the model's adaptive ability.

[0081] like Figure 7 As shown, in order to implement the above-mentioned decomposition furnace coal feeding amount control method based on the decomposition furnace outlet temperature prediction, this scheme also provides a decomposition furnace coal feeding amount control system based on the decomposition furnace outlet temperature prediction, and the system includes:

[0082] A historical training sample acquisition unit is used to obtain a characteristic point standard set based on a time-delay cross-correlation method, obtain corresponding historical data of the characteristic point standard set and perform preprocessing to obtain historical training samples;

[0083] The decomposition furnace outlet temperature prediction model establishment unit is used to use historical training samples to perform model training to obtain a decomposition furnace outlet temperature prediction model to establish a mapping relationship from the coal-air combination factor to the future decomposition furnace outlet temperature;

[0084] The tail coal feeding amount control model establishment unit is used to construct a tail coal feeding amount control model based on the deviation distribution statistics and deviation trend statistics between the predicted value and the target value of the decomposition furnace outlet temperature, as well as the step mutation statistics of the raw meal feeding amount, so as to infer the tail coal feeding amount set value of the decomposition furnace in the next time step based on the prediction of the decomposition furnace outlet temperature;

[0085] The system operation unit is used to obtain the data corresponding to the characteristic point standard set in real time, and obtain the tail coal feeding amount of the decomposition furnace at the next time step by using the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model, and input the given value into the distributed control system for execution to adjust the tail coal feeding amount of the decomposition furnace;

[0086] The model dynamic update unit is used to monitor the decomposition furnace outlet temperature prediction effect and the tail coal feeding amount optimization effect according to the operating conditions after the tail coal feeding amount is adjusted during the system operation, and dynamically update the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model based on the monitoring results.

[0087] It can be understood that the system can fully execute all the processes in the above-mentioned decomposition furnace coal feeding amount control method based on the prediction of the decomposition furnace outlet temperature, and the specific process will not be repeated.

[0088] The decomposition furnace coal feeding control system and method based on the decomposition furnace outlet temperature prediction provided in this embodiment obtains the main factors and time delays affecting the decomposition furnace outlet temperature through mechanism analysis, and establishes an accurate decomposition furnace outlet temperature prediction model. In actual application, the average absolute error between the temperature prediction value and the true feedback value is no more than 2°C, which meets the decomposition furnace outlet temperature prediction effect in the conventional cement burning system; the iTransformer model is applied to the decomposition furnace outlet temperature prediction, and the prediction effect is improved; the deviation distribution and the deviation change rate distribution of the decomposition furnace outlet temperature prediction value and the target value are used to jointly control the change of the coal feeding amount, considering both the static deviation value and the dynamic deviation change trend; the effects of the decomposition furnace temperature prediction model and the coal feeding amount control model are tracked, and the parameters of the two models are adjusted and updated in time to adapt to changes in working conditions.

[0089] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field are aware of all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement the scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for the ordinary technicians in the relevant field to implement this application. It should be pointed out that for the technicians in this field, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent.

Claims

1. A method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the outlet temperature of the decomposition furnace, characterized in that: The method comprises the following steps: S100, obtaining a characteristic point standard set based on a time-delay cross-correlation method, obtaining corresponding historical data of the characteristic point standard set and performing preprocessing to obtain historical training samples; S200, using historical training samples to perform model training, obtain a decomposition furnace outlet temperature prediction model, so as to establish a mapping relationship from the coal-air combination factor to the future decomposition furnace outlet temperature; S300, based on the deviation distribution statistics and deviation trend statistics of the predicted value and target value of the decomposition furnace outlet temperature, as well as the step mutation statistics of the raw meal feeding amount, a tail coal feeding amount control model is constructed to infer a given value of the tail coal feeding amount of the decomposition furnace in the next time step based on the prediction of the decomposition furnace outlet temperature; S400, after the system is deployed, the data corresponding to the characteristic point standard set is obtained in real time, and the tail coal feeding amount of the decomposition furnace is obtained by using the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model, and the given value of the tail coal feeding amount of the decomposition furnace in the next time step is input into the distributed control system for execution to adjust the tail coal feeding amount of the decomposition furnace; S500, during the operation of the system, the decomposition furnace outlet temperature prediction effect and the tail coal feeding amount optimization effect are monitored according to the operating conditions after the tail coal feeding amount is adjusted, and the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model are dynamically updated with the monitoring results.

2. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 1, characterized in that: In S200, the decomposition furnace outlet temperature prediction model structure adopts the iTransformer architecture with time series prediction.

3. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 1, characterized in that: In S200, the mathematical expression of the decomposition furnace outlet temperature prediction model is shown in the following formula: y t =f(x t-L ,…,x t-1 ,x t ) Among them, x t represents the characteristic vector of the coal-air combination factor corresponding to the current time step t; L represents the length of the historical characteristic sequence pushed forward from the current time step t; f represents nonlinear mapping; y t A label sequence representing the predicted values ​​of the decomposition furnace outlet temperature at a number of future time steps relative to the current time step t.

4. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 1, characterized in that: In S300, the control algorithm formula of the tail coal feeding amount control model is: in, The given value of tail coal feeding amount in the next time step; m t is the feedback value of tail coal feeding amount at the current time step t; θ β ,θ γ To adjust the step size hyperparameter; α t β is the coefficient for adjusting the tail coal feed amount due to a sudden change in the raw meal feed amount; t The coefficient for adjusting the tail coal feeding amount according to the deviation distribution index between the predicted value and the target value of the decomposition furnace outlet temperature is represented; γ t To characterize the coefficient for adjusting the tail coal feeding amount based on the deviation trend index between the predicted value and the target value of the decomposition furnace outlet temperature.

5. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 1, characterized in that: In S500, the prediction effect of the decomposition furnace outlet temperature is monitored by determining whether the average absolute error between the predicted value given by the decomposition furnace outlet temperature prediction model and the decomposition furnace outlet temperature feedback value exceeds a first threshold value; if the average absolute error is greater than the first threshold value, the prediction effect does not meet expectations, and the operating condition data of the current and historical several time steps corresponding to the decomposition furnace outlet temperature prediction value is regarded as a negative sample, and the decomposition furnace outlet temperature prediction model is fine-tuned and updated with the negative sample.

6. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 4, characterized in that: In S500, the optimization effect of the tail coal feeding amount is monitored by whether the average absolute error between the decomposition furnace outlet temperature feedback value and the target value exceeds the second threshold; if the average absolute error is greater than the second threshold, the optimization effect does not meet expectations, and the tail coal feeding amount control model is updated according to the monitoring results of the decomposition furnace outlet temperature prediction effect.

7. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 6, characterized in that: If the monitored decomposition furnace outlet temperature prediction effect does not meet expectations, the decomposition furnace outlet temperature prediction model is updated first, and the deviation sequence of the decomposition furnace outlet temperature prediction value and the target value at the current and historical time steps is obtained, and the β is updated with the mean and standard deviation of the deviation sequence. t , update γ by comparing the deviation slope of the deviation sequence and its threshold t , and at the same time obtain the current raw material feed quantity given value and its previous value to update α t , in order to update the tail coal feeding quantity control model.

8. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 6, characterized in that: If the predicted effect of the monitored decomposition furnace outlet temperature meets expectations, the adjustment step length hyperparameter θ in the tail coal feeding control model is directly adjusted according to the preset method. β ,θ γ .

9. The method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to claim 8, characterized in that: The default method is to use the least squares method to minimize the mean square error between the feedback value and the target value of the decomposition furnace outlet temperature through linear fitting, so as to obtain a new adjustment step size hyperparameter θ β ,θ γ .

10. A coal feeding control system for a decomposition furnace based on the prediction of the outlet temperature of the decomposition furnace, characterized in that: A method for controlling the coal feeding amount of a decomposition furnace based on the prediction of the decomposition furnace outlet temperature according to any one of claims 1 to 9 is implemented; the system comprises: A historical training sample acquisition unit is used to obtain a characteristic point standard set based on a time-delay cross-correlation method, obtain corresponding historical data of the characteristic point standard set and perform preprocessing to obtain historical training samples; The decomposition furnace outlet temperature prediction model establishment unit is used to use historical training samples to perform model training to obtain a decomposition furnace outlet temperature prediction model to establish a mapping relationship from the coal-air combination factor to the future decomposition furnace outlet temperature; The tail coal feeding amount control model establishment unit is used to construct a tail coal feeding amount control model based on the deviation distribution statistics and deviation trend statistics between the predicted value and the target value of the decomposition furnace outlet temperature, as well as the step mutation statistics of the raw meal feeding amount, so as to infer the tail coal feeding amount set value of the decomposition furnace in the next time step based on the prediction of the decomposition furnace outlet temperature; The system operation unit is used to obtain the data corresponding to the characteristic point standard set in real time, and obtain the tail coal feeding amount of the decomposition furnace at the next time step by using the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model, and input the given value into the distributed control system for execution to adjust the tail coal feeding amount of the decomposition furnace; The model dynamic update unit is used to monitor the decomposition furnace outlet temperature prediction effect and the tail coal feeding amount optimization effect according to the operating conditions after the tail coal feeding amount is adjusted during the system operation, and dynamically update the decomposition furnace outlet temperature prediction model and the tail coal feeding amount control model based on the monitoring results.

Images