Method and system for optimizing target value of outlet temperature of decomposing furnace of cement sintering system

Abstract

The invention relates to the technical field of cement firing, and discloses a cement firing system decomposing furnace outlet temperature target value optimization method, which reasonably determines input parameters based on production data accumulated in a cement plant distributed control system, and fuses autoregression and classification models to obtain a target value. A high-quality historical target value adjustment data training model is screened out according to rules, a decomposition furnace outlet temperature target value optimization model is generated, input parameter values of the current moment and the historical time period of the decomposition furnace and a previous decomposition furnace outlet temperature target value are input in real time, and a decomposition furnace outlet temperature target value is obtained through the constructed optimization model. And outputting an adjusting action type and a relative front value adjusting quantity of the target value of the outlet temperature of the decomposing furnace under the current working condition, and directly adjusting the target value of the outlet temperature of the decomposing furnace according to the output adjusting action type and the relative front value adjusting quantity so as to realize one-time in-place adjustment of the target value of the outlet temperature of the decomposing furnace. Therefore, the purposes of automation, intelligence and rapid and accurate adjustment effect of the target value adjustment means of the outlet temperature of the decomposing furnace are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of cement burning, and in particular to a method and system for optimizing a target value of a decomposition furnace outlet temperature in a cement burning system. Background Art

[0002] The pre-decomposition step is the core of the new dry process cement production technology. The outlet temperature of the decomposition furnace is an important process parameter in this step, which directly affects the output, quality and energy consumption of cement. When the outlet temperature of the decomposition furnace is too high, it is easy to cause crusting at the feed port of the preheater or the inner wall of the cyclone, affecting the normal transportation and suspension preheating of the raw material; when the outlet temperature of the decomposition furnace is too low, it may cause the decomposition rate of the cement raw material to decrease and increase the burning load of the rotary kiln. Therefore, in order to produce cement with high output, high efficiency, energy saving and environmental protection, it is of great significance to optimize the target value of the outlet temperature of the decomposition furnace.

[0003] However, the internal structure of the suspended preheater and decomposition furnace is complex, the steps are cumbersome, the physical and chemical reactions are intertwined, the nonlinear and random characteristics are obvious, and the number of internal measurement and control points is small. Multiple variables in the process influence each other, which brings challenges to the optimization of the target value of the decomposition furnace outlet temperature.

[0004] Currently, the determination of the target temperature value at the outlet of the decomposition furnace is still mainly determined by the operator based on experience. On the one hand, the temperature target value fluctuates greatly due to the differences in operating levels and operating habits of different operators. On the other hand, the timeliness of adjusting the target temperature value at the outlet of the decomposition furnace to respond to changes in the firing conditions depends on the sensitivity of manual operation, which can easily lead to a lag in the adjustment of the target temperature value at the outlet of the decomposition furnace.

[0005] In addition to manually setting the target value, patent CN116048163A discloses a cement kiln decomposition furnace temperature setting optimization method. Based on the principle of thermodynamics, the recommended strategy for the decomposition furnace outlet temperature setting value is obtained by using the kiln main current, secondary air temperature, kiln condition comprehensive index, and head-to-tail coal ratio, and the priority of each strategy and the adjustment range of the decomposition furnace outlet temperature setting value under the strategy are classified and discussed. This method belongs to the rule model and requires the adjustment strategy to be defined in advance, but it is impossible to accurately find the appropriate temperature target value for the working conditions that are not defined in advance. Summary of the invention

[0006] The present invention aims to provide a method for optimizing the target value of the outlet temperature of the decomposition furnace of a cement burning system, which can sensitively monitor the operation changes of the burning system and directly predict the adjustment action type and relative adjustment amount of the target value of the outlet temperature of the decomposition furnace at the current moment, so as to achieve the purpose of automating and intelligently adjusting the target value of the outlet temperature of the decomposition furnace and achieving fast and accurate adjustment effect.

[0007] The basic scheme provided by the present invention is: a method for optimizing the target value of the outlet temperature of a decomposition furnace in a cement burning system, the method comprising:

[0008] S100, setting input parameters and output parameters of the optimization model for the target value of the decomposition furnace outlet temperature, wherein the output parameters include the adjustment action type and the adjustment amount relative to the previous value, and the previous target value of the decomposition furnace outlet temperature is the previous value;

[0009] S200, filtering the historical sample data of the decomposition furnace based on preset rules, wherein the historical sample data of the decomposition furnace includes the input parameter data and output parameter data of the decomposition furnace corresponding to several good adjustments of the outlet temperature target value during the operation of the decomposition furnace in the past period of time;

[0010] S300, constructing an autoregressive classification fusion model based on a neural network, training the fusion model using historical sample data of the decomposition furnace, and generating a decomposition furnace outlet temperature target value optimization model; the decomposition furnace outlet temperature target value optimization model is used to fit the functional relationship between the input parameter and the output parameter;

[0011] S400, using the decomposition furnace outlet temperature target value optimization model, inputting the decomposition furnace input parameter values ​​at the current time and historical time period and the previous decomposition furnace outlet temperature target value in real time, and outputting the adjustment action type and the adjustment amount relative to the previous value;

[0012] S500, performing a threshold comparison judgment on the relative previous value adjustment amount outputted in S400, and determining according to the judgment result whether the target value of the decomposition furnace outlet temperature is adjusted in place at one time based on the adjustment action type outputted in S400 and the relative previous value adjustment amount.

[0013] The present invention is based on a method for optimizing a target value of a decomposition furnace outlet temperature of a cement burning system, and also provides a system for optimizing a target value of a decomposition furnace outlet temperature of a cement burning system, the system comprising:

[0014] A setting module is used to set the input parameters and output parameters of the decomposition furnace;

[0015] An acquisition module is used to set preset rules and filter the historical sample data of the decomposition furnace based on the preset rules;

[0016] The model module is used to construct an autoregressive classification fusion model based on a neural network, and use the historical sample data of the decomposition furnace to train the fusion model, generate a decomposition furnace outlet temperature target value optimization model and send it to the calculation module;

[0017] The data module is used to collect the input parameter data of the decomposition furnace at the current time and historical time period and the target value of the outlet temperature of the previous decomposition furnace in real time, and input them into the calculation module;

[0018] The calculation module is used to receive the input parameter values ​​of the current time and historical time period and the previous decomposition furnace outlet temperature target value input by the data module in real time, and use the decomposition furnace outlet temperature target value optimization model to output the adjustment action type and the relative adjustment amount of the previous value of the decomposition furnace outlet temperature target value under the current working condition;

[0019] The adjustment module is used to compare and judge the threshold value of the relative previous value adjustment amount obtained by the calculation module output, and determine whether to adjust the decomposition furnace outlet temperature target value in one go based on the output adjustment action type and the relative previous value adjustment amount according to the judgment result.

[0020] The working principle and advantages of the present invention are: input parameter data of the decomposition furnace at the current time and historical time period and the previous decomposition furnace outlet temperature target value are input in real time, and a constructed decomposition furnace outlet temperature target value optimization model is used to output the adjustment action type and relative previous value adjustment amount of the decomposition furnace outlet temperature target value under the current working conditions, and based on the threshold comparison result of the relative previous value adjustment amount, it is judged whether the output adjustment action type and relative previous value adjustment amount are used to adjust the decomposition furnace outlet temperature target value in one go.

[0021] When studying the optimization method for the target value of the decomposition furnace outlet temperature, the existing technology faces the challenge of optimizing the target value brought about by the complexity of the production system structure. The conventional idea is to directly predict the target value by constructing the system operating conditions to avoid constructing a complex relationship between several operating parameters and the target value, and at the same time to avoid the complexity of the model by outputting an independent target value.

[0022] Compared with the existing technology, this scheme breaks through the conventional idea of ​​building a relationship between the system operating conditions and the target value of the outlet temperature of the decomposition furnace, and uses historical good manual adjustment actions to reconstruct the input-output relationship. The autoregressive and classification models are integrated in a data model-based + rule model-assisted manner. The deep neural network is trained based on high-quality manual adjustments selected from historical data. There is no need to define the working conditions in advance, and it can fully adapt to all operating conditions of the system. It does not rely on the experience and habits of the operator. By sensitively monitoring the changes in the working conditions of the burning system and directly inputting the operating parameters, it can simulate the good manual adjustment method to output the adjustment direction and relative adjustment amount, and determine the adjustment timing by the relative adjustment amount threshold. At the same time, it answers three core questions: "When to adjust the outlet temperature target value", "How to adjust (increase or decrease)", and "How much is the adjustment range", so as to achieve the purpose of automating and intelligently adjusting the means of adjusting the outlet temperature target value of the decomposition furnace, and achieving the purpose of fast and accurate adjustment effect. Through the special construction of the autoregressive classification fusion model of this scheme, the adjustment action type and relative adjustment amount are predicted at the same time. Compared with the traditional model that only predicts an independent target value, the prediction accuracy of the outlet temperature target value of the decomposition furnace can be effectively improved. In addition, by setting preset rules, the effect of optimizing the target value of the decomposition furnace outlet temperature can be evaluated using the qualified proportion of raw material decomposition rate and the standard coal consumption of tailings. Adjustments with better evaluation can further achieve model optimization and improve model accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic flow chart of a method for optimizing a target value of a decomposition furnace outlet temperature in a cement burning system provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of the structure of a neural network model for adjusting and predicting the target value of the decomposition furnace outlet temperature provided in the initial design of an embodiment of the present invention;

[0025] Figure 3 A schematic diagram of the structure of the autoregressive and classification fusion model provided by an embodiment of the present invention;

[0026] Figure 4 A curve chart of the model training loss value provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following is a further detailed description through specific implementation methods:

[0028] Embodiment 1

[0029] Basically as attached Figure 1 As shown: A method for optimizing the target value of the outlet temperature of a decomposition furnace in a cement burning system, the method comprising the following steps:

[0030] S100, setting input parameters and output parameters of a decomposition furnace outlet temperature target value optimization model;

[0031] Specifically, since the long- and short-term change trends of multi-dimensional historical operating data such as kiln current and smoke chamber gas are mainly considered when manually adjusting the target value of the decomposition furnace outlet temperature, this scheme uses the cross-correlation algorithm to set the input parameters within the empirical parameter range. The pseudo code for using the cross-correlation algorithm to screen features within the empirical parameter range is as follows:

[0032] Input: The candidate feature set R = {r1, r2, ..., r m}, where m is the number of candidate features; the threshold θ of the mutual correlation coefficient of retained features;

[0033] Initialization: S = {};

[0034] For i in R:

[0035] Calculate r i Cross-correlation with the decomposition furnace outlet temperature c i , c i = cov(r i , decomposition furnace outlet temperature) / (σ ri ·σ 分解炉出口温度 ), where cov is the covariance of two variables, σ ri is the feature r i The standard deviation, σ 分解炉出口温度 is the standard deviation of the decomposition furnace outlet temperature;

[0036] If c i ≥θ, the feature r i Add to S; otherwise, skip feature r i ;

[0037] Output: The retained feature set S = {s1, s2, ..., s n}, where n is the number of retained features.

[0038] The determined input parameters are shown in Table 1:

[0039] Table 1 Input parameters

[0040]

[0041]

[0042] The output parameters include the type of adjustment action and the adjustment amount relative to the previous value, with the previous decomposition furnace outlet temperature target value being the previous value; the decomposition furnace outlet temperature target value and the adjustment trend of the target value are learned directly from the accumulated historical sample data without the need for additional manual annotation information.

[0043] S200, filtering the historical sample data of the decomposition furnace based on preset rules, wherein the historical sample data of the decomposition furnace includes the input parameter data and output parameter data of the decomposition furnace corresponding to several good adjustments of the outlet temperature target value during the operation of the decomposition furnace in the past period of time;

[0044] Specifically, in the massive historical production data of at least 2 years, the adjustment effect of the target value of the decomposition furnace outlet temperature is different each time. In order to ensure the effect of model training, it is necessary to select data with better adjustment effect as samples to improve the output accuracy of the model.

[0045] Based on the historical operation status after each adjustment of the target value of the decomposition furnace outlet temperature, the preset rules are used to determine whether to retain the data corresponding to this adjustment as a sample. The preset rules include:

[0046] (1) After the target value of the decomposition furnace outlet temperature is adjusted, the mean raw material decomposition rate is ≥ the raw material decomposition rate threshold; the raw material decomposition rate threshold can be set to 90% to balance the sample data quantity and quality;

[0047] (2) For the portion exceeding the raw meal decomposition rate threshold, the following formula is used to further calculate the raw meal decomposition rate marginal benefit per unit of coal fed to the decomposition furnace:

[0048] Marginal benefit = (average raw material decomposition rate - raw material decomposition rate threshold) / average coal feeding amount of decomposition furnace

[0049] Sort the marginal benefits from large to small, set the marginal benefit threshold, and select the decomposition furnace outlet temperature target value adjustment set corresponding to the marginal benefit higher than the threshold to form several good adjustments of the outlet temperature target value, that is, select the decomposition furnace outlet temperature target value adjustment that makes the raw material decomposition rate qualified and the marginal benefit higher as the good adjustment, and use the decomposition furnace operation data corresponding to the good adjustment as the decomposition furnace historical sample data; the decomposition furnace historical sample data includes the decomposition furnace input parameter data and output parameter data corresponding to several good adjustments of the outlet temperature target value in the operation of the decomposition furnace in the past period of time. The number of retained samples can be controlled by setting the marginal benefit threshold.

[0050] S300, constructing an autoregressive classification fusion model based on a neural network, training the fusion model using historical sample data of the decomposition furnace, and generating a decomposition furnace outlet temperature target value optimization model; the decomposition furnace outlet temperature target value optimization model is used to fit the functional relationship between the input parameter and the output parameter.

[0051] S301, construct model training samples with the decomposition furnace input parameter data and output parameter data corresponding to several good adjustments in the decomposition furnace historical sample data; use the mini-batch training method based on stochastic gradient descent, use a small batch sample consisting of several samples (64 in this scheme) for each iteration, select the model with the smallest model loss function as the best model after multiple iterations, and complete the training of the constructed autoregressive classification fusion model. Among them, the time series operating condition data of the current moment and the past history totaling 5 minutes of each good adjustment is used as a sample feature input. The selection setting of 5 minutes takes into account the acquisition delay of each parameter. The conventional data delay is 3-5 minutes. Selecting the maximum delay can produce samples with complete features but redundant. However, the training of neural networks supports the introduction of redundant data, and key features and redundant features can be distinguished according to the weights of neurons.

[0052] First, the box plot method is used to filter out abnormal data in the input parameter data of the historical sample data: the lower quartile Q1 and upper quartile Q3 of the sample data are calculated, and then the interquartile range IQR = Q3-Q1 is calculated, and only the data within the interval [Q1-1.5IQR, Q3+1.5IQR] is retained; then the maximum and minimum normalization is used to unify the feature dimensions, and the data of each output parameter is normalized to the range of [0,1]. The normalization formula is as follows:

[0053] Normalized value = (original value - minimum value) / (maximum value - minimum value)

[0054] At each time step, the output parameters are selected to form a data vector, and the feature vector at the current time t is x t This scheme uses the feature matrix composed of the vectors of the current moment of each well-adjusted history and the past 5 minutes as the sample X t ={x t-L ,…,x t-1 ,x t}, L represents the historical length of sample usage. All constructed samples are divided into training set, validation set and test set in a ratio of 7:1:2.

[0055] S302, constructing an autoregressive classification fusion model based on a neural network.

[0056] This scheme does not directly regress and predict the target temperature value of the decomposition furnace outlet at a certain moment, but further considers the type of temperature target value adjustment action, that is, temperature target value reduction, maintenance, and increase. The conventional method of directly predicting the adjustment range using only the regression model has been found to be poor in effect through experiments, and the MSE loss function is difficult to converge; after adding the classification loss (cross entropy loss), the model considers the adjustment direction while considering the adjustment range, and the comprehensive loss converges significantly in the experiment. Therefore, when the regression loss is difficult to further reduce, by introducing the classification loss of the temperature target value adjustment action type, the prediction accuracy of the decomposition furnace outlet temperature target value can be effectively improved in theory.

[0057] Therefore, this scheme constructs an autoregressive classification fusion model, including a classification model and a regression model, in which the classification model and the regression model share the convolution layer parameters, and the gru layer and the dense layer are modeled separately. Taking the operating condition characteristics at the current moment and the historical preset time and the previous decomposition furnace outlet temperature target value as input, the classification predicts the adjustment action of the decomposition furnace outlet temperature target value at the current moment, including lowering, maintaining, and increasing, and at the same time, regression predicts the relative amount of the target value adjustment.

[0058] The modeled mapping relationship is as follows:

[0059] y′ t =f reg (X t ,y′ t-1 ,z′ t-1 )

[0060] z′ t =f cls (X t ,y′ t-1 ,z′ t-1 )

[0061] Among them, y′ t represents the adjustment amount of the target value of the decomposition furnace outlet temperature predicted by the model at time t relative to the previous value, z′ t represents the probability distribution vector of each adjustment action output by the model at time t, f reg is the nonlinear mapping function represented by the regression model, f cls It is a nonlinear function characterized by the classification model; Xt represents the input matrix composed of the input parameter data at the current moment and within the historical preset time, and the preset time can be selected as 5 minutes.

[0062] Since the model simultaneously models the classification of the adjustment action and the regression of the adjustment amount, the loss function of the model training in this scheme is the weighted sum of the classification loss function and the regression loss function:

[0063] loss = α·loss cls +(1-α)·lossreg

[0064] Among them, α is the weighted harmonic coefficient, loss cls is the classification loss function, loss = α·loss cls +(1-α)·loss reg is the regression loss function; the value range of α is (0,1), and α indicates the degree of attention paid to the classification of the adjustment action. The larger the value, the more the weighted loss pays attention to the classification, and vice versa. α is obtained through experiments, ranging from 0.1 to 0.9, with an interval of 0.1, and the training loss size of different α is compared, and the best one is selected. After experiments, 0.6 is preferably selected.

[0065] Classification loss function loss cls Using cross entropy, as shown in the following expression:

[0066]

[0067] Where n represents the total number of samples (i.e. the total number of historical good adjustments screened out), z t It represents the actual adjustment action of the target temperature value at the outlet of the decomposition furnace at time t (in this embodiment, it is assumed that each historical adjustment is adjusted in place once), and is encoded as a vector in a one-hot manner, that is, a decrease in the temperature target value is encoded as [1,0,0], a remain unchanged value is encoded as [0,1,0], and an increase is encoded as [0,0,1].

[0068] Regression loss function loss = α·loss cls +(1-α)·loss reg Using the root mean square logarithmic error, the introduction of log operation is conducive to amplifying the difference between the true value and the predicted value, as shown in the following expression:

[0069]

[0070] Among them, y t Represents the actual value of the target value of the decomposition furnace outlet temperature at time t relative to the previous value adjustment, y′ t It represents the adjustment amount of the target value of the decomposition furnace outlet temperature predicted by the model at time t relative to the previous value.

[0071] The structure of the neural network model for adjusting the target value of the decomposition furnace outlet temperature in the initial design is as follows: Figure 2As shown in the figure, "conv1x1" and "conv3x3" represent convolution operations with kernel sizes of 1x1 and 3x3, respectively. Convolutional networks are a type of feedforward neural network that imitates biological vision mechanisms. They can extract local texture features of each part of the input data in sequence with a small number of parameters and in a convolutional calculation manner. The purpose of designing two different sizes of convolutional kernels is to learn features at different scales, thereby generating rich and differentiated features.

[0072] also, Figure 2 The GRU (gated recurrent unit) network structure in the paper is a variant of the Recurrent Neural Network (RNN). RNN is a type of recursive neural network that takes sequence data as input, recurses in the direction of sequence evolution, and all nodes are connected in a chain. It has the characteristics of memory, parameter sharing, and Turing completeness, so it has certain advantages in learning the nonlinear characteristics of the sequence. Since the input of the model is current and historical data, that is, the time series data of the working conditions, the GRU model structure is designed to model the relationship between sequences. GRU is a variant of RNN, but the traditional RNN model cannot model sequences that are too long. The length of the time series sequence used in this solution is 5 minutes. RNN cannot model such a long sequence, which will cause problems such as gradient diffusion. Therefore, using GRU to implement model construction can solve the problems of the inability to remember for a long time in traditional RNN and the gradient dissipation in back propagation.

[0073] In order to reduce model parameters, this solution further integrates the above classification and regression models, shares the convolutional layer parameters, and separately models the GRU layer and dense layer belonging to classification and regression. Figure 3 As shown in the figure, with the continuous accumulation of online data, the optimization model can achieve gradual improvement in effect through iterative optimization.

[0074] The traditional autoregressive model inputs the target value of the decomposition furnace outlet temperature at the last moment during training, which is the true value y t-1 However, when the model is actually deployed for reasoning, since there is no real value, the input is the predicted value y′ of the target value adjustment of the decomposition furnace outlet temperature at the previous moment t-1 Therefore, there is a significant difference in the input distribution between the training stage and the inference stage of the model, which leads to inaccurate predictions. As the time series inference proceeds, the prediction error gradually accumulates. Therefore, this scheme adopts the scheduled sampling method during model training to bridge the input difference between the training state and the inference state of the traditional autoregressive model: during model training, the input of the target value of the decomposition furnace outlet temperature at the previous moment is calculated with probability p. t Select the true value y t-1 , with probability 1-pt Select the predicted value y′ t-1 In general, a larger probability p is used in the early stages of training. t The real value of the target temperature at the previous moment is selected as the input with high probability, so as to quickly guide the model from the randomly initialized state to a reasonable state. As the training progresses, p t Gradual decay, this method will gradually use more predicted values ​​of the previous moment as input to solve the problem of inconsistent data distribution. Common probability decay methods include linear decay, exponential decay, directional sigmoid decay, etc. This scheme uses a simple linear decay method to select the probability p of the true value and decay according to the following formula:

[0075] p t =max(p F ,p0-k·t)

[0076] Among them, p0 is the upper limit of the probability value, p F It is the lower limit of the probability value, and k controls the speed at which the probability decays.

[0077] S303: The model uses the training set data to perform parameter training, uses the validation set data to evaluate the effect of the model during the training process, and selects the model with the smallest loss as the target model based on the weighted loss function on the validation set. Figure 4 As shown in the figure, the training set loss continues to decrease with the increase in the number of training rounds, and the validation set loss first decreases and then increases. Therefore, this scheme selects the intermediate model of the 62nd round of training as the target model. The validation set loss of the model after this round of training is the smallest.

[0078] S304, model training is completed, and a decomposition furnace outlet temperature target value optimization model is generated; the decomposition furnace outlet temperature target value optimization model is used to fit the functional relationship between the input parameters and the output parameters.

[0079] S400, using the decomposition furnace outlet temperature target value optimization model, real-time input of the decomposition rate at the current time and historical time period input parameter values ​​and the previous decomposition furnace outlet temperature target value, output to obtain the decomposition furnace outlet temperature target value under the current working conditions adjustment action type and relative to the previous value adjustment amount.

[0080] Specifically, in this scenario, the model predicts that the interval for adjusting the target value of the decomposition furnace outlet temperature is 1 minute, that is, an inference is made every minute to determine whether adjustment is needed, which solves the problem that humans cannot monitor and adjust online for a long time.

[0081] S500, deploying the model to the server, and adjusting the target value of the decomposition furnace outlet temperature in place at one time according to the adjustment action type and the relative adjustment amount to the previous value obtained in S400;

[0082] Specifically, the output relative previous value adjustment amount is compared and judged by the threshold value, and it is determined whether the target value of the decomposition furnace outlet temperature is adjusted in place at one time based on the output adjustment action type and the relative previous value adjustment amount according to the judgment result. If it is not lower than the threshold value, the target value of the decomposition furnace outlet temperature is adjusted in place at one time with the received adjustment action type and the relative previous value adjustment amount; if it is lower than the threshold value, the target value of the decomposition furnace outlet temperature is kept unchanged. Considering the actual production, the threshold value can be 1°C.

[0083] In order to implement the above method, the present invention also provides a cement burning system decomposition furnace outlet temperature target value optimization system, the system comprises

[0084] A setting module is used to set the input parameters and output parameters of the decomposition furnace;

[0085] An acquisition module is used to set preset rules and filter the historical sample data of the decomposition furnace based on the preset rules;

[0086] The model module is used to construct an autoregressive classification fusion model based on a neural network, and use the historical sample data of the decomposition furnace to train the fusion model, generate a decomposition furnace outlet temperature target value optimization model and send it to the calculation module;

[0087] The data module is used to collect the input parameter data of the decomposition furnace at the current time and historical time period and the target value of the outlet temperature of the previous decomposition furnace in real time, and input them into the calculation module;

[0088] The calculation module is used to receive the input parameter values ​​of the current time and historical time period and the previous decomposition furnace outlet temperature target value input by the data module in real time, and use the decomposition furnace outlet temperature target value optimization model to output the adjustment action type and the relative adjustment amount of the previous value of the decomposition furnace outlet temperature target value under the current working condition;

[0089] The adjustment module is used to compare and judge the threshold value of the relative previous value adjustment amount obtained by the calculation module output, and determine whether to adjust the decomposition furnace outlet temperature target value in one go based on the output adjustment action type and the relative previous value adjustment amount according to the judgment result.

[0090] The present embodiment provides a method and system for optimizing the outlet temperature target value of a cement burning furnace in a cement burning system. The method uses historical good manual adjustment actions to reconstruct the input-output relationship, and fuses the autoregressive and classification models in a data model-based + rule model-assisted manner. The deep neural network is trained based on the high-quality manual adjustment selected from the historical data. The working conditions do not need to be defined in advance, and the deep neural network can be fully adapted to all operating conditions of the system. The method does not rely on the operator's experience and habits. By sensitively monitoring the changes in the working conditions of the burning system, the operating parameters are directly input to simulate the good manual adjustment method to output the adjustment direction and relative adjustment amount. At the same time, the adjustment timing is determined by the relative adjustment amount threshold. At the same time, the three core questions of "when to adjust the outlet temperature target value", "how to adjust (increase or decrease)", and "how much is the adjustment range" are answered, and the outlet temperature target value of the decomposition furnace is adjusted once, so as to achieve the purpose of automating and intelligently adjusting the outlet temperature target value of the decomposition furnace, and achieving the purpose of fast and accurate adjustment effect. In addition, through the special construction of the autoregressive classification fusion model of this scheme, the adjustment action type and the relative adjustment amount are predicted at the same time, which can effectively improve the prediction accuracy of the outlet temperature target value of the decomposition furnace compared with the traditional model that only predicts an independent target value. In addition, by setting preset rules, the effect of optimizing the target value of the decomposition furnace outlet temperature can be evaluated using the qualified proportion of raw material decomposition rate and the standard coal consumption of tailings. Adjustments with better evaluation can further achieve model optimization and improve model accuracy.

[0091] Embodiment 2

[0092] Different from the first embodiment, there are multiple attempts to adjust the target temperature each time in history. When constructing a data set from historical data, for the case where the target value of the decomposition furnace outlet temperature is adjusted multiple times in a short period of time, multiple adjustments within one hour are combined into one adjustment, and the direction and amplitude of the adjustment are accumulated for the aforementioned training process. This can better utilize historical data to improve the quality of the model and the accuracy of the optimization results.

[0093] 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 optimizing the target value of the outlet temperature of a calciner in a cement burning system, characterized in that: The method comprises: S100, setting input parameters and output parameters of the optimization model for the target value of the decomposition furnace outlet temperature, wherein the output parameters include the adjustment action type and the adjustment amount relative to the previous value, and the previous target value of the decomposition furnace outlet temperature is the previous value; S200, filtering the historical sample data of the decomposition furnace based on preset rules, wherein the historical sample data of the decomposition furnace includes the input parameter data and output parameter data of the decomposition furnace corresponding to several good adjustments of the outlet temperature target value during the operation of the decomposition furnace in the past period of time; S300, constructing an autoregressive classification fusion model based on a neural network, training the fusion model using historical sample data of the decomposition furnace, and generating a decomposition furnace outlet temperature target value optimization model; the decomposition furnace outlet temperature target value optimization model is used to fit the functional relationship between the input parameter and the output parameter; S400, using the decomposition furnace outlet temperature target value optimization model, inputting the input parameter values ​​of the current time and historical time period and the previous decomposition furnace outlet temperature target value in real time, and outputting the adjustment action type and the adjustment amount relative to the previous value; S500, performing a threshold comparison judgment on the relative previous value adjustment amount outputted in S400, and determining according to the judgment result whether the target value of the decomposition furnace outlet temperature is adjusted in place at one time based on the adjustment action type outputted in S400 and the relative previous value adjustment amount.

2. The method for optimizing the target value of the outlet temperature of the calciner in a cement burning system according to claim 1, characterized in that: In S100, the input parameters are filtered using a cross-correlation algorithm within the empirical parameter range; the input parameters include the measured value of the decomposition furnace outlet temperature, the measured value of the decomposition furnace outlet pressure, the measured value of the kiln tail smoke chamber temperature, the measured value of the kiln tail smoke chamber pressure, the measured value of the tertiary air temperature, the measured value of the tertiary air branch pipe 1 pressure, the measured value of the tertiary air branch pipe 2 pressure, the measured value of the preheater cyclone C6A discharge pipe temperature, the measured value of the preheater cyclone C6B discharge pipe temperature, the measured value of the raw material feeding amount into the kiln, the measured value of the decomposition furnace coal feeding amount, the measured value of the decomposition furnace coal feeding and coal injection pressure, the measured value of the high-temperature fan frequency and the decomposition furnace outlet temperature target value.

3. The method for optimizing the target value of the outlet temperature of the calciner in a cement burning system according to claim 1, characterized in that: In S200, the preset rules include: After the target value of the decomposition furnace outlet temperature is adjusted, the mean raw meal decomposition rate is ≥ the raw meal decomposition rate threshold value; For the portion exceeding the raw meal decomposition rate threshold, the following formula is used to further calculate the raw meal decomposition rate marginal benefit per unit of coal fed to the decomposition furnace: Marginal benefit = (raw meal decomposition rate average - raw meal decomposition rate threshold) / average coal feeding amount of decomposition furnace Sort the marginal benefits from large to small, set a marginal benefit threshold, and select the decomposition furnace outlet temperature target value adjustment set corresponding to the marginal benefit higher than the threshold to form several good adjustments of the outlet temperature target value.

4. The method for optimizing the target value of the outlet temperature of the calciner in a cement burning system according to claim 1, characterized in that: The box plot method is used to filter out abnormal data in the input parameter data of the decomposition furnace historical sample data, and then normalize them to construct model training samples with the processed data.

5. The method for optimizing the target value of the outlet temperature of the calciner in a cement burning system according to claim 1, characterized in that: In S300, the autoregressive classification fusion model includes a classification model and a regression model, wherein the classification model and the regression model share convolutional layer parameters, and the gru layer and the dense layer are modeled separately.

6. The method for optimizing the target value of the outlet temperature of the calciner in a cement burning system according to claim 5, characterized in that: The modeled mapping relationship is as follows: y′ t =f reg (X t ,y′ t-1 ,z′ t-1 ) z′ t =f cls (X t ,y′ t-1 ,z′ t-1 ) Among them, y′ t represents the adjustment amount of the target value of the decomposition furnace outlet temperature predicted by the model at time t relative to the previous value, z′ t represents the probability distribution vector of each adjustment action output by the model at time t, f reg is the nonlinear mapping function represented by the regression model, f cls It is a nonlinear function characterized by the classification model; Xt represents the input matrix composed of the input parameter data at the current moment and the historical preset time.

7. The method for optimizing the target value of the outlet temperature of the calciner in a cement burning system according to claim 5, characterized in that: The loss function of model training is the weighted sum of the classification loss function and the regression loss function: loss=α·loss cls +(1-α)·loss reg Among them, α is the weighted harmonic coefficient, loss cls is the classification loss function, loss = α·loss cls +(1-α)·loss reg is the regression loss function.

8. The method for optimizing the target value of the outlet temperature of the decomposition furnace of a cement burning system according to claim 1, characterized in that: In S500, when the output relative to the previous value adjustment amount is lower than the threshold value, the decomposition furnace outlet temperature target value will not be adjusted, and the previous decomposition furnace outlet temperature target value will be kept unchanged. Otherwise, the decomposition furnace outlet temperature target value will be adjusted once and for all according to the adjustment action type and the relative to the previous value adjustment amount obtained by the output in S400.

9. A system for optimizing the target value of the outlet temperature of the decomposition furnace of a cement burning system, characterized in that: A method for optimizing the target value of the outlet temperature of a decomposition furnace in a cement burning system according to any one of claims 1 to 8; the system comprises A setting module is used to set the input parameters and output parameters of the decomposition furnace; An acquisition module is used to set preset rules and filter the historical sample data of the decomposition furnace based on the preset rules; The model module is used to construct an autoregressive classification fusion model based on a neural network, and use the historical sample data of the decomposition furnace to train the fusion model, generate a decomposition furnace outlet temperature target value optimization model and send it to the calculation module; The data module is used to collect the input parameter data of the decomposition furnace at the current time and historical time period and the target value of the outlet temperature of the previous decomposition furnace in real time, and input them into the calculation module; The calculation module is used to receive the input parameter values ​​of the current time and historical time period and the previous decomposition furnace outlet temperature target value input by the data module in real time, and use the decomposition furnace outlet temperature target value optimization model to output the adjustment action type and the relative adjustment amount of the previous value of the decomposition furnace outlet temperature target value under the current working condition; The adjustment module is used to compare and judge the threshold value of the relative previous value adjustment amount obtained by the calculation module output, and determine whether to adjust the decomposition furnace outlet temperature target value in one go based on the output adjustment action type and the relative previous value adjustment amount according to the judgment result.

10. A cement burning system decomposition furnace outlet temperature target value optimization system according to claim 9, characterized in that: If the received relative previous value adjustment amount is not lower than the threshold, the decomposition furnace outlet temperature target value is adjusted once for all using the received adjustment action type and the relative previous value adjustment amount; if it is lower than the threshold, the previous decomposition furnace outlet temperature target value is kept unchanged.

Images