Method and system for predicting the external temperature of a rotary kiln shell

Abstract

The external temperature (T n A method (100) for predicting a temperature (t), t≧0, comprising a training feature providing step (110) for providing a training feature dataset (Equation I) and a step of providing a training external temperature dataset (T n (t), t<0) and at least one model A for at least one slice n of the rotary kiln. n a model setup step (130) for setting up at least one algorithm A for at least one slice n of the rotary kiln (1) from a machine learning algorithm; n an algorithm selection step (150) for selecting at least one algorithm A n an algorithm selection step (150) for at least one slice n of the rotary kiln (1), wherein the algorithm selection step (150) includes model parameters (Equation II) and hyperparameters (Equation III); n , and at least one algorithm A for at least one slice n of the rotary kiln (1) in the trained state. n the training feature dataset (Equation IV) and the training external temperature dataset (T n an algorithm training step (170) for training by using at least a portion of the time series (t), t<0), wherein the algorithm training step (170) is based on supervised learning; and a predictive model A comprising at least one algorithm A for at least one slice n of the rotary kiln (1) in the trained state. n *, and a prediction feature providing step (180) for providing a prediction feature dataset (Equation V), and an external temperature dataset (Tn (t), t≧0) is used as the prediction model A n *, and a temperature prediction step (190) for making a prediction based on the prediction feature dataset (Equation V).

Description

[Technical Field]

[0001] The present invention relates to a method and system for predicting the external temperature of a rotary kiln shell for at least one slice n of the rotary kiln. [Background technology]

[0002] WO 2020 / 188549(A2) relates to a method and system for monitoring and optimizing the operation of an alumina rotary kiln and discloses a temperature prediction unit 116 configured to predict gas and solids temperature profiles at preselected locations within the alumina rotary kiln 102 over a predetermined time period using preprocessed real-time data and thermal models stored in a database 126, where the thermal models may include first-principles-driven models, data-driven models or knowledge-driven models.

[0003] EP 1228016 B1 relates to a kiln plant control system and discloses the use of a control matrix containing values that determine the relationship between operating parameters and plant measurements, such as the hood temperature (Thood) as disclosed in FIG. 6.

[0004] J. Xu, D. Fu, L. Shao, X. Zhang and G. Liu, “A Soft Sensor Modeling of Cement Rotary Kiln Temperature Field Based on Model-Driven and Data-Driven Methods,” (in IEEE Sensors Journal, vol. 21, no. 24, pp. 27632-27639, 15 Dec. 2021, doi:10.1109 / JSEN.2021.3116937) discloses a soft sensor that takes axial air velocity, swirling air velocity, coal mass flow rate, material mass flow rate, secondary air temperature, and x, y, and z coordinates as inputs and outputs the temperature at a specific location inside the kiln.

[0005] During the operation of a rotary kiln, it is essential to monitor its condition with respect to the external temperature of the rotary kiln shell. If this temperature becomes too high, significant damage to the kiln can occur, for example, due to (partial) melting of the shell in areas where the external temperature exceeds a certain threshold. Such excessive temperatures can occur, for example, due to wear of the inner refractory, which can also result in increased heat load on the shell. Wear of the inner refractory in a rotary kiln cannot be easily monitored during operation; internal inspection can only be performed after the kiln is shut down. Therefore, knowledge of various wear mechanisms and other processes in rotary kilns is limited. Furthermore, there may be other influences on the outer shell temperature aside from refractory wear, such as areas in the rotary kiln with higher internal temperatures, caused by exothermic chemical reactions or directly in the vicinity of the burner, e.g., in the flame zone. In this case, the temperature may also depend on the fuel, fuel mixture, or fuel amount used. Currently, the outer shell temperature is typically monitored by several temperature sensors. Whenever the temperature reaches a certain upper temperature threshold, the kiln is shut down and the refractory lining inside the kiln is renewed. Furthermore, it is difficult to predict the outer shell temperature for future production, for example, in the following days or weeks. Therefore, for safety reasons, the upper temperature threshold is set to a value significantly below the shell failure temperature. Rotary kiln campaigns are often terminated prematurely without utilizing the refractory lining to its full potential, thereby increasing operating costs and reducing productivity. In other words, if the external kiln temperature could be predicted, it would be possible to keep the rotary kiln in operation for a longer period before the refractory lining needs to be renewed. Furthermore, a lower temperature threshold is also of interest to operators, since, for example, a sudden and significant drop in temperature (e.g., a temperature peak down) could lead to increased thermal shock and, consequently, increased refractory wear.

[0006] The inventors have recognized that a reliable prediction of the external temperature of a rotary kiln shell for at least one slice n of the rotary kiln should be achieved by using a model based on a machine learning algorithm trained with a training feature dataset and a training external temperature dataset. Such a trained machine learning algorithm can reliably predict the external temperature of the rotary kiln shell for at least one slice n of the rotary kiln in the near future. This reliable prediction can arise from the fact that a machine learning algorithm trained in this manner does not require any prior knowledge to be provided to such an algorithm or model. Thus, prior art models based on arbitrary physical models, thermodynamic variables, first-principles-driven models, data-driven models, or knowledge-driven models frequently fail in practical applications, which the inventors believe is due to a lack of a complete understanding of the complexities of all processes and wear mechanisms occurring in a rotary kiln. The use of a machine learning algorithm in the context of the present invention does not require any input regarding such internal wear mechanisms or processes, as the resulting machine learning model is achieved by training the model with a training feature dataset and a training external temperature dataset. The inventors further recognized that such models can be trained to very high accuracy if the training is performed using a large external temperature data set with respect to such data having high spatial resolution outside the rotary kiln shell. The inventors recognized that such a large temperature data set can preferably be provided by using one or more infrared cameras pointed at the rotary kiln shell and capable of providing multiple temperature readings during operation of the rotary kiln. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] International Publication No. 2020 / 188549(A2) [Patent Document 2] European Patent No. 1228016(B1) [Non-patent literature]

[0008] [Non-Patent Document 1] J. Dec.15,2021,doi:10.1109 / JSEN.2021.3116937) Summary of the Invention [Problem to be solved by the invention]

[0009] It is therefore an object of the present invention to provide a method and system for predicting the external temperature of a rotary kiln shell for at least one slice n of the rotary kiln. [Means for solving the problem]

[0010] This object is achieved by a method according to claim 1, a prediction unit according to claim 12, a prediction system according to claim 14 and a rotary kiln system according to claim 15.

[0011] The core idea of the present invention is based on the finding that machine learning can be used to determine the future external temperature of a rotary kiln shell for at least one slice n of the rotary kiln. This can be achieved, inter alia, by selecting a machine learning algorithm, feeding the algorithm selected features of at least one slice n of the rotary kiln, and further training the machine learning algorithm to obtain a model. The model can then be used to predict the external temperature for at least one slice n of the rotary kiln based on the predictive feature dataset.

[0012] The term "predict" as used herein is defined as a prior declaration or indication, in other words, an indication of the temperature of an external temperature data set (T n Predicting (t), t≧0) involves foreseeing (future) external temperature data sets in advance.

[0013] The term "machine learning," as used herein, includes the use and development of computer systems that can learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze patterns in data and draw inferences.

[0014] The term "machine learning algorithm," as used herein, is defined as an algorithm used in machine learning applications (an algorithm is a set of rules that must be followed when solving a particular problem). Examples of such machine learning algorithms include DT (decision tree regression), RF (random forest regression), SVR (support vector regression), and NN (neural network). Machine learning algorithms build models based on sample data, known as training data, to make predictions or decisions without being explicitly programmed to do so. Generally, in slice n, the associated algorithm is A n It is expressed as:

[0015] The term "supervised learning," also referred to as supervised training, as used herein may include the use of labeled data sets to train an algorithm to classify data or accurately predict outcomes. Supervised learning, also known as supervised machine learning, is a subcategory of machine learning. Supervised learning is a machine learning task that learns a function that maps inputs to outputs based on example input-output pairs. In the context of the present invention, the input-output pairs may include feature values for at least one slice n of the rotary kiln and temperature data for at least one slice n of the rotary kiln.

[0016] The term "model parameters," as used herein, may include parameters in a model that must be determined by performing a training step, such as supervised learning (by using input-output pairs). These are the fitted parameters of the model. In general, slice n contains: Specific algorithm A represented by JPEG2025526272000002.jpg850 n There can be K sets of model parameters.

[0017] The term "hyperparameters", as used herein, may include adjustable parameters that can be adjusted to obtain a particular model. These are parameters whose values are used to control the learning process. In contrast, the values of model parameters are derived through training. The resulting model depends on the actual hyperparameters used. For example, the hyperparameters depend on the actual machine learning algorithm used. In the case of DT, the hyperparameters may include the tree depth or splitting criteria. In general, for slice n, Specific algorithm A represented by JPEG2025526272000003.jpg1053 n There can be a set of J hyperparameters.

[0018] The term "machine learning model," as used herein, may include a data file containing data that has been trained to achieve a good fit between input-output pairs and the output, generated by using a machine learning algorithm with its respective hyperparameters after supervised training.

[0019] The term "rotary kiln" ("Drehrohrofen"), as used herein, refers to a cylindrical vessel having a longitudinal axis L, which is slightly inclined relative to the horizontal, and which rotates about its longitudinal axis L. The cylindrical vessel has an (outer) shell (called the "rotary kiln shell"), usually made of metal, which is lined with refractory bricks on the inside (i.e., between the inside of the rotary kiln shell and the longitudinal axis L of the rotary kiln). The rotary kiln shell has a surface (called the "surface of the rotary kiln shell") facing away from the longitudinal axis L of the rotary kiln, and the rotary kiln shell has an external temperature (the "external temperature of the rotary kiln shell"). The cylindrical vessel is usually heated (fired) by a burner at the lower end of the cylindrical vessel. As the cylindrical vessel rotates, the material moves gradually downward from the top to the bottom. The material is continuously stirred and mixed while being heated by the hot flame / hot gas from the burner. Typical applications of rotary kilns include increasing the temperature of material, for example, for firing in a continuous process.

[0020] The term "external temperature of the surface of the rotary kiln" as used herein includes a temperature measurement outside the rotary kiln at the surface of the rotary kiln shell. This temperature can be measured at this exterior of the object by a temperature measurement device, such as a thermal sensor(s), or non-contact, such as an infrared camera(s).

[0021] The term "slice n of a rotary kiln," as used herein, includes a portion of the rotary kiln, i.e., a portion of the cylindrical vessel. This portion is preferably a cylinder, for example a right cylinder, whereby the cylinder defining the slice is cut from the cylindrical vessel. In other words, the slice is preferably a cylinder whose longitudinal axis coincides with the longitudinal axis L of the cylindrical vessel. The slices are numbered such that slice n is the nth (e.g., 1st, 2nd, 3rd, ...) slice of the rotary kiln. A slice can be very short (i.e., the slice has a short longitudinal dimension), for example, in the width (to be understood as the dimension of a single brick along the longitudinal axis L in the installed form) (e.g., 200 mm) or a fraction of the width of a single brick (e.g., 1 / 10 of the brick width, 20 mm, etc.), and preferably includes at least one single temperature measurement. In some areas, the slices may be, for example, up to 5 (such as 1000 mm) or 10 bricks wide (such as 2000 mm).

[0022] The term "characteristics of at least one slice n of a rotary kiln," as used herein, includes any input values or data extracted from a rotary kiln production and / or laboratory information management system. For example, the characteristics may include process characteristics such as flame temperature, inlet chamber temperature, rotary kiln torque, energy consumption, and chemical characteristics such as alkali content, chloride content, sulfur content, etc. of the hot meal and / or clinker.

[0023] The term "temperature data" as used herein includes, for example, a machine-readable data structure in the form of a single temperature value, or a set of temperature values in the form of a field / vector, or a matrix containing temperature values.

[0024] A prediction unit is understood to mean one or more devices for carrying out the respective method steps described below, and for this purpose comprising separate electronic components for processing signals, or which are partly or completely implemented as a computer program in a computer.

[0025] In a first aspect of the present invention, the object is to determine the external temperature (T n (t), t≧0), the method comprising: (a) a training feature dataset ( a training feature providing step for providing JPEG2025526272000004.jpg810(t), m=1,...,M*, t<0); (b) The external temperature (T n (t, t<0) is the training external temperature dataset (T n (t), t<0) training temperature providing step; (c) at least one model A for at least one slice n of the rotary kiln; n * A model setup step for setting up (i) at least one algorithm A for at least one slice n of the rotary kiln from a machine learning algorithm; n an algorithm selection step for selecting at least one algorithm A n an algorithm selection step, where (ii) at least one algorithm A for at least one slice n of the rotary kiln; n , and at least one algorithm A for at least one slice n of the rotary kiln in the trained state. n The resulting training feature dataset ( JPEG2025526272000005.jpg46(t), m=1,…,M, t<0, M≦M*) and a training external temperature dataset (T n (t), t<0), (iii) an algorithm training step, wherein the algorithm training step is based on supervised learning; (iv) thereby generating at least one algorithm A for at least one slice n of the rotary kiln in the trained state; n , and a set of hyperparameters Prediction Model A including JPEG2025526272000006.jpg629 n * can be obtained a model setup step, including (d) a predicted feature dataset (preferably comprising values of the M features for at least one slice n of the rotary kiln) JPEG2025526272000007.jpg46(t), m=1,...,M, t≧0), (e) an external temperature data set (T n (t), t≧0) is used as the prediction model A n *By using the prediction feature dataset ( a temperature prediction step for predicting based on JPEG2025526272000008.jpg46(t), m=1,...,M, t≧0); It has.

[0026] Preferably, the method comprises measuring the external temperature (T n(t), t≧0) (i.e., the external temperature of the surface of the rotary kiln). Preferably, at least one slice n of the rotary kiln may cover all of a particular zone of the rotary kiln, such as the exit zone, lower transition zone, combustion zone, upper transition zone, safety zone, calcination zone, or entrance zone. Alternatively, the method may predict the external temperature (T n (t), t≧0, n=1,…,N).

[0027] Preferably, the method is performed at a specific time (here t=0) and is able to predict the external temperature at that time or in the future (t≧0).

[0028] Preferably, a training feature dataset ( JPEG2025526272000009.jpg46(t), m=1,...,M*, t<0) is provided in the training feature providing step. JPEG2025526272000010.jpg46(t) are collected in the past, i.e., at a specific time t<0, i.e., before the execution of the method. The number M* of such features can be in the range 5-500, preferably 10-400, most preferably 50-300. JPEG2025526272000011.jpg46(t) is preferably selected from experimental values or properties of clinker, kiln feed, hot meal, fuel type, or from kiln operating parameters such as pressure or flame temperature. Training Feature Dataset JPEG2025526272000012.jpg46(t) is a specific point in time Contains the time series of the values of each feature in JPEG2025526272000013.jpg849. JPEG2025526272000014.jpg727 is preferably selected from a time interval extending from just before the method is performed to a particular point in time in the past, for example at least the last 50 days, preferably at least the last 100 days, more preferably at least the last 300 days, and most preferably the last 500 days.

[0029] Preferably, a training external temperature data set (T) is provided, which includes external temperature data of the rotary kiln shell for at least one slice n of the rotary kiln. n (t), t<0) is provided in the training temperature providing step. n (t) is collected in the past, i.e., at a certain time t<0, i.e., before the execution of the method. n The external temperature data set T(t) may include a single temperature value for each slice n of the rotary kiln at a particular time t, or may include multiple temperature values for each slice n of the rotary kiln at a particular time t. n (t) is the time JPEG2025526272000015.jpg527 contains a time series of temperature values. JPEG2025526272000016.jpg415 is preferably selected from a time interval from just before the execution of the method to a particular point in time in the past, for example at least the last 50 days, preferably at least the last 100 days, more preferably at least the last 300 days, and most preferably the last 500 days. The point in time t is preferably selected from a training feature dataset JPEG2025526272000017.jpg46(t) are selected from the same time interval as for the external temperature data set T n (t) and the training feature dataset JPEG2025526272000018.jpg46(t) contains the temperature data and the values of M* features taken at the same time point t. In other words, For JPEG2025526272000019.jpg963, at the same time t k in JPEG2025526272000020.jpg46(t k ) and T n (t k ) pairs exist.

[0030] Preferably, a pre-processing step is performed for pre-processing of at least one dataset, and the at least one dataset is a training feature dataset ( JPEG2025526272000021.jpg46(t), m=1,…,M*, t<0), training external temperature dataset (T n (t), t<0), is selected from the list.

[0031] Preferably, the training feature dataset ( JPEG2025526272000022.jpg46(t), m=1,…,M*, t<0) and / or the training external temperature dataset (T n The preprocessing step for preprocessing at least one dataset, such as (t), t<0), includes checking the data format and data structure of the dataset, and / or checking the consistency of the dataset, and / or extracting outliers from the dataset, and / or removing duplicates in the dataset, and / or imputing missing values. This preprocessing allows achieving faster convergence of the algorithm training step, thus saving calculation time.

[0032] Preferably, the pre-processing step for pre-processing at least one dataset includes scaling of values in the dataset. This scaling significantly speeds up the convergence of the algorithm training step, thereby saving additional computation time. Furthermore, scaling is performed on the training feature dataset ( This results in a significant reduction in the memory required to store and process JPEG2025526272000023.jpg46(t), m=1,...,M*, t<0).

[0033] Preferably, at least one model A for at least one slice n of the rotary kilnn *Model setup steps are performed to set up the

[0034] Preferably, at least one model A for at least one slice n of the rotary kiln n *Model setup step to set up the scoring function E for at least one slice n of the rotary kiln n The method further includes a scoring function setup step for setting up (t), where the scoring function is selected from RMSE (Root Mean Square Error), trend, and weighted scoring functions. n (t) is at least one algorithm A selected in the algorithm selection step. n and the training external temperature dataset (T n (t), t<0). In other words, the scoring function E n (t) is a set of algorithms A n (during the training phase, at t<0) is used to calculate the training external temperature dataset (T n (t), t<0) to provide a score. n (t), t<0) is the goal of the algorithm training step, so the scoring function E n (t) is the state of a specific algorithm A n Preferably, the scoring function E n (t) is all Regarding JPEG2025526272000024.jpg527, JPEG2025526272000025.jpg1896 is selected as the RMSE calculated at time t. Preferably, the scoring function E n (t) is all Regarding JPEG2025526272000026.jpg527, JPEG2025526272000027.jpg18125 is selected as the trend calculated at time t. Preferably, the scoring function E n (t) is all Regarding JPEG2025526272000028.jpg527, JPEG2025526272000029.jpg14152 is selected as the weighted scoring function calculated at time t.

[0035] Preferably, at least one model A for at least one slice n of the rotary kiln n * Model setup step for setting up at least one algorithm A for at least one slice n of the rotary kiln from the machine learning algorithm n an algorithm selection step for selecting at least one algorithm A n is the model parameter ( JPEG2025526272000030.jpg912) and hyperparameters ( JPEG2025526272000031.jpg1016) and includes an algorithm selection step. Algorithm A n is preferably a training feature dataset ( Input such as JPEG2025526272000032.jpg46(t) is output A n ( The image is set up to serve the image (JPEG2025526272000033.jpg1087).

[0036] Preferably, in the algorithm selection step, the selected machine learning algorithm is selected from any of tree-based machine learning algorithms such as decision trees (DTs) and random forests (RFs), vector-based machine learning algorithms such as gradient boosting regression and support vector regression (SVR), neural network-based algorithms such as simple neural networks (NNs), and ensemble machine learning algorithms such as voting regression.

[0037] Preferably, at least one model A for at least one slice n of the rotary kiln n The model setup step for setting up M* further comprises an algorithm feature selection step prior to a subsequent algorithm training step, wherein the algorithm feature selection step preferably comprises a training feature dataset ( a selected training feature dataset ( t ) ( m = 1, ..., M*, t < 0) preferably comprising values of M features for at least one slice n of the rotary kiln; JPEG2025526272000035.jpg46(t), m=1,...,M, t<0, M≦M*). Preferably, the selected training feature dataset ( JPEG2025526272000036.jpg46(t), m=1,...,M, t<0, M≦M*) is a training feature dataset ( JPEG2025526272000037.jpg46(t), m=1,...,M*, t<0). Preferably, the reduced number of M features is less than 50% of the number M*, more preferably less than 40%, and most preferably less than 30%. Preferably, the algorithmic feature selection step includes at least one algorithm selected from correlation-based feature selection (CBFS), mutual information-based feature selection (MIFS), sequential forward selection (SFS), and sequential backward selection (SBS). This step reduces the execution time of the algorithmic training step in all cases. Furthermore, this step may reduce the number of features used in subsequent algorithmic training steps in such a way that the accuracy of the results of the subsequent training steps is not compromised and perhaps even improved.

[0038] Preferably, at least one model A for at least one slice n of the rotary kiln n * Model setup step for setting up at least one algorithm A for at least one slice n of the rotary kiln n , and at least one algorithm A for at least one slice n of the rotary kiln in the trained state. n The resulting training feature dataset ( JPEG2025526272000038.jpg46(t), m=1,…,M, t<0, M≦M*) and a training external temperature dataset (T n (t), t<0). Preferably, the algorithm training step is based on supervised learning. Preferably, at least one algorithm A for at least one slice n of the rotary kiln in the trained state is n , and a set of hyperparameters ( Prediction model A including JPEG2025526272000039.jpg629 n* is obtained. The inventors believe that such use of a machine learning algorithm, especially when trained by supervised learning, can provide a method for predicting the external temperature (T n We have found that this results in very good predictions of the time series (t), t≧0). The algorithm training step preferably uses Algorithm A n The training feature dataset ( JPEG2025526272000040.jpg46(t, m=1,…,M, t<0, M≦M*) is the training external temperature dataset (T n Algorithm A for approximating (t), t<0) n Algorithm A n A set of K model parameters ( JPEG2025526272000041.jpg527). In other words, the scoring function E n (t) is algorithm A n A set of K model parameters ( The scoring function is optimized (here, minimized) by the variation of the scoring function threshold E n (t) <E min So, when the scoring function reaches a certain (minimum) goal, Algorithm A n A set of K model parameters ( The fluctuation of JPEG2025526272000043.jpg527) is stopped, and algorithm A n is the trained state, i.e., algorithm A n But Algorithm A n The final set of K model parameters ( JPEG2025526272000044.jpg527). Therefore, the final set of K model parameters ( JPEG2025526272000045.jpg527), and algorithm A n ( The set of hyperparameters for JPEG2025526272000046.jpg11109 ( At least one algorithm A for at least one slice n of the rotary kiln in the trained state, including JPEG2025526272000047.jpg629 n Prediction model A including n * may be obtained.

[0039] Preferably, a predicted feature dataset ( JPEG2025526272000048.jpg46(t), m=1,...,M, t≧0). Preferably, a predicted feature dataset ( The M features of JPEG2025526272000049.jpg46(t), m = 1, ..., M, t ≥ 0) are stored in the training feature dataset ( JPEG2025526272000050.jpg46(t), consisting of the same M features (m=1,…,M*, t<0).

[0040] Preferably, the predicted feature dataset ( The prediction feature providing step for providing a prediction feature dataset ( JPEG2025526272000051.jpg46(t), m=1,...,M, t≧0) is performed by generating a prediction feature dataset ( JPEG2025526272000052.jpg46(t), m=1,…,M, t≧0), or a training feature dataset ( JPEG2025526272000053.jpg46(t), m=1,…,M, t<0, M≦M*) based on the continuation of values from the predicted feature dataset ( JPEG2025526272000054.jpg46(t), m=1,…,M, t≧0), or a training feature dataset ( JPEG2025526272000055.jpg46(t), m=1,…,M, t<0, M≦M*) based on a subset of values from the feature dataset ( JPEG2025526272000056.jpg46(t), m=1,...,M, t≧0). Preferably, the method includes providing a predicted feature dataset ( The prediction feature providing step for providing the training feature dataset (JPEG2025526272000057.jpg46(t), m=1,...,M, t≧0) is JPEG2025526272000058.jpg46(t), m=1,…,M, t<0, M≦M*) in the same format and order as the predicted feature dataset ( JPEG2025526272000059.jpg46(t), m=1,...,M, t≧0).

[0041] Preferably, at least one model A for at least one slice n of the rotary kiln n The model setup step involves setting up an external temperature data set (T) for at least one slice n of the rotary kiln. n (t), t≧0) is used as the prediction model A n *By using the prediction feature dataset ( JPEG2025526272000060.jpg46(t), m=1,...,M, t≧0). Preferably, the predicted external temperature data set (T n The external temperature data set (T (t), t≧0) contains temperature data in the future, i.e., after the execution of the method (at t=0), i.e., at time t≧0. n (t), t≧0) is a specific time point JPEG2025526272000061.jpg848 Contains a time series of temperature values. JPEG2025526272000062.jpg415 is preferably selected from a time interval from just before the method is performed to a particular point in time in the future, preferably at least 5 days in the future, more preferably at least 10 days in the future, and most preferably 30 days in the future.

[0042] Preferably, at least one model A for at least one slice n of the rotary kiln n *Model setup step to set up the scoring function E for at least one slice n of the rotary kiln n and setting up at least one scoring function A for at least one slice n of the rotary kiln. n The algorithm setup step is repeated for different sets of hyperparameters (i) to set up the scoring function E n (i) Different models with different results n (i) and then the scoring function E n (i*) Model A has the best results n (i*) is prediction model A n * is selected.

[0043] Preferably, the training external temperature data set (T n (t), t<0) comprises temperature data for at least one slice n of the rotary kiln obtained by at least one temperature monitoring unit, such as an infrared camera. Preferably, the temperature monitoring unit (20) comprises at least one of an infrared camera (25), a discrete temperature measurement probe, such as a thermocouple, or an optical temperature measurement device, such as a fiber optic temperature sensor.

[0044] Preferably, the method according to the first aspect of the invention is carried out in parallel for N slices of the rotary kiln. Thus, preferably, the method according to the first aspect of the invention comprises measuring the external temperature (T n (t), t≧0), where for each slice n, a respective method claim is made. Thus, preferably, at least one slice n comprises at least 100 slices, more preferably at least 500 slices, and most preferably at least 1000 slices.

[0045] Preferably, the training external temperature data set (T n (t), t<0) includes averaged external temperature data within at least one slice n of the rotary kiln, preferably by averaging the external temperature data along the circumference of at least one slice n of the rotary kiln. This allows training the model algorithm without having to obtain parameters of the rotation of the rotary kiln in the training feature data set.

[0046] In a second aspect of the present invention, the object is to determine the external temperature (T n (t), t≧0), the prediction unit (a) Rotary kiln training feature dataset ( a training feature interface for receiving JPEG2025526272000063.jpg46(t), m=1,...,M*, t<0); (b) a training external temperature data set (T n A training temperature interface for receiving (t), t<0); (c) an external temperature data set (T n (t), t ≥ 0) and a temperature prediction interface Equipped with The prediction unit is (d) From the training feature interface, a training feature dataset ( JPEG2025526272000064.jpg46(t), m=1,…,M*, t<0), (e) Training external temperature dataset (T n (t), t<0), (f) at least one model A for at least one slice n of the rotary kiln; n *of, (i) at least one algorithm A for at least one slice n of the rotary kiln from a machine learning algorithm; n and selecting at least one algorithm A n is the model parameter ( JPEG2025526272000065.jpg57) and hyperparameters ( JPEG2025526272000066.jpg69), and (ii) at least one algorithm A for at least one slice n of the rotary kiln; n , and at least one algorithm A for at least one slice n of the rotary kiln in the trained state. n The resulting training feature dataset ( JPEG2025526272000067.jpg46(t), m=1,…,M, t<0, M≦M*) and a training external temperature dataset (T n (t), t<0), (iii) at least one algorithm A n training is based on supervised learning; (iv) at least one algorithm A for at least one slice n of the rotary kiln in the trained state; n , and a predictive model A with set hyperparameters n *Getting Set up by (g) a predicted feature dataset (preferably including values of the M features for at least one slice n of the rotary kiln) JPEG2025526272000068.jpg46(t), m=1,...,M, t≧0), (h) an external temperature data set (T n (t), t≧0) is used as the prediction model A n *By using the prediction feature dataset ( Predict based on JPEG2025526272000069.jpg46(t), m=1,…,M, t≧0 It is programmed as follows.

[0047] Rotary kiln training feature dataset ( The training feature interface for receiving JPEG2025526272000070.jpg46(t), m=1,...,M*, t<0) may be any interface capable of receiving a training feature dataset from a feature providing unit, for example from a source such as a memory or a server, or directly from a production system, etc.

[0048] A training external temperature data set (T nThe training temperature interface for receiving the temperature data set (T (t), t<0) may be any interface that can receive a temperature data set, for example from a source such as a memory or a server, or directly from a unit comprising a temperature monitoring unit, for example an infrared camera. Preferably, the prediction unit receives a training external temperature data set (T n Let (t), t<0) be a training external temperature dataset (T n The temperature monitoring unit may further include a temperature monitoring unit for providing a training temperature interface for receiving (t), t<0).

[0049] The external temperature data set (T n The temperature prediction interface for providing (t), t≧0) may be any interface that can provide the temperature dataset to a source, such as a memory or a server, or directly to a temperature prediction display device, such as a display or monitor.

[0050] Any interface is understood to mean one or more devices for receiving or providing a data set as described above, and for this purpose, either equipped with separate electronic components for processing the data, or partly or fully implemented as a computer program in a computer.

[0051] The prediction unit according to the second aspect of the invention is preferably programmed to perform any or all of the steps according to the first aspect of the invention.

[0052] In a third aspect of the present invention, the object is achieved by providing a prediction system for predicting the external temperature (Tn(t), t≧0) of a rotary kiln shell for at least one slice n of the rotary kiln, the prediction system comprising: a prediction unit according to any of the second aspects of the invention; and A training feature dataset ( a feature providing unit for providing the training feature input (t), m=1,...,M*, t<0) to the training feature interface of the prediction unit; A training external temperature data set (T) containing temperature data of the outer surface of the rotary kiln shell for at least one slice n of the rotary kiln n a temperature monitoring unit for providing (t), t<0) to a training temperature interface of the prediction unit; The external temperature data set (T) for at least one slice n of the rotary kiln provided by the temperature prediction interface of the prediction unit n (t, t≧0) and a temperature prediction display device for displaying information based on the temperature prediction Equipped with.

[0053] A training feature dataset ( The feature providing unit for providing the training feature data set (JPEG2025526272000072.jpg46(t), m=1,...,M*, t<0) to the training feature interface of the prediction unit provides the rotary kiln features to the training feature data set ( The feature providing unit may be a source such as a memory / data acquisition unit connected to a production and / or laboratory information management system, etc., which may provide the feature data set ( t ), m=1,...,M*, t<0) from the production information management system and / or laboratory information management system of the rotary kiln, preferably including values of M* features of the rotary kiln. It is preferable to receive JPEG2025526272000074.jpg46(t), m=1,...,M*, t<0).

[0054] A training external temperature data set (T n The temperature monitoring unit for providing the temperature data set (T (t), t<0) to the training temperature interface of the prediction unit may comprise a discrete temperature measurement probe such as a thermocouple or an optical temperature measurement device such as a fiber optic temperature sensor. Preferably, the temperature monitoring unit comprises an infrared camera, preferably at least one infrared camera. Preferably, the infrared camera records the external kiln shell temperature and the temperature monitoring unit generates a training external temperature data set (T n (t), t<0). Preferably, the temperature monitoring unit comprises a fiber optic temperature sensor. Preferably, the fiber optic temperature sensor is in thermal contact with the rotary kiln shell. Preferably, the temperature monitoring unit provides a temperature data set (T) comprising external temperature data of the rotary kiln shell for N slices n (n=1,...,N) of the rotary kiln, preferably for at least N=100 slices, more preferably for at least N=500 slices, and most preferably for at least N=1000 slices. n (t), t<0).

[0055] The external temperature data set (T) for at least one slice n of the rotary kiln provided by the temperature prediction interface of the prediction unit n The temperature prediction display device for displaying information based on (t), t≧0) may be a display, a monitor, or the like.

[0056] Exemplary embodiments of the invention will now be explained in more detail by means of the figures. [Brief explanation of the drawings]

[0057] [Figure 1] 1 shows a schematic diagram of a first embodiment of the present invention. [Figure 2] 1 shows a schematic representation of the second and third embodiments of the present invention. [Figure 3] An exemplary calculation is shown in which a training external temperature data set (Tn(t), t<0) is averaged within at least one slice n of the rotary kiln. [Figure 4] The results of the prediction of the external temperature (Tn(t), t ≥ 0) of the rotary kiln shell are shown. DETAILED DESCRIPTION OF THE INVENTION

[0058] Figure 1 shows the external temperature (T n 1 shows a schematic diagram of a first embodiment of the present invention, a method (100) for predicting the mean temperature (t), t≧0. First, a training feature dataset ( , M*, t<0). In one embodiment, 150 such features (M*=150) are generated from the time series JPEG2025526272000075.jpg46(t), m=1,...,M*, t<0, at a particular time t<0, i.e., before the method (100) is performed. JPEG2025526272000076.jpg46(t), m = 1, ..., M*. In one example, each time series for each feature m contains values for the last 360 days at 1 hour intervals (thus providing k = 8640 values per feature m in this example). All features are collected simultaneously, so the time series are: JPEG2025526272000077.jpg527 at the same time, JPEG2025526272000078.jpg is synchronized in that it contains values for each of the 835 features m.

[0059] Then, a training temperature providing step (120) provides a training external temperature data set (T) containing temperature data of the outer surface (2) of the rotary kiln shell (3) for slice n of the rotary kiln (1). n (t), t<0). In a specific example, the training external temperature data set (T n (t), t<0) contains external temperature data for slice n of the rotary kiln (1) obtained by at least one temperature monitoring unit (20), which in this example is an infrared camera (25). The infrared camera may acquire external temperature data of the rotary kiln shell (3) for several slices n of the rotary kiln. This is shown in FIG. 3, where the temperature data JPEG2025526272000079.jpg1021 is collected in a particular segment q (q=1,...,Q) for slice n (n=1,...,N). The segments extend along the z direction (extending along the direction of the longitudinal axis L) and are aligned with the azimuthal angle JPEG2025526272000080.jpg811 is a specific portion of the rotary kiln shell (3) including a specific range. The training external temperature data set (T n (t), t<0) comprises external temperature data averaged within at least one slice n of the rotary kiln (1), and thus in this example, a training external temperature data set (T n (t), t<0) is the average temperature T n (t), and therefore JPEG2025526272000081.jpg1574.

[0060] In this example, a preprocessing step (125) is performed to preprocess at least one dataset, in this example two datasets: a training feature dataset ( JPEG2025526272000082.jpg46(t), m=1,…,M*, t<0) and the training external temperature dataset (Tn (t), t<0) are used. Both of these datasets are checked for data format and structure, as well as for dataset consistency. Additionally, both datasets are subjected to outlier extraction, duplicate removal, and missing value imputation. In this example, even the values in the datasets are scaled by normalization and standardization.

[0061] Another step is to install at least one model A for slice n of the rotary kiln (1). n The model setup step (130) involves setting up a scoring function E* for slice n of the rotary kiln (1). This step includes further substeps. In this example, n A scoring function setup step (140) is performed to set up (t). In this example, the scoring function is an RMSE scoring function, so that all Regarding JPEG2025526272000083.jpg527, JPEG2025526272000084.jpg1053. The RMSE scoring function depends on the algorithm (to be selected). JPEG2025526272000085.jpg821) and the provided training external temperature dataset (T n (t), t<0). The scoring function is a function of the algorithm ( The output of JPEG2025526272000086.jpg512 and the training external temperature dataset (T n (t), t<0), and is therefore a measure of how well trained the algorithm is.

[0062] In the algorithm selection step (150), at least one algorithm A for slice n of the rotary kiln (1) is selected. n is selected from a machine learning algorithm, in this example the algorithm is a neural network (NN) algorithm, and this algorithm A nis the model parameter ( JPEG2025526272000087.jpg57) and hyperparameters ( JPEG2025526272000088.jpg69). In this example, the first set of hyperparameters ( JPEG2025526272000089.jpg818) is selected (in this case, this is the optimizer selected as "rmsprop", epochs are chosen to be 100, and batch size is chosen to be 20). These particular choices of hyperparameters ( It was found in techniques for the optimization of JPEG2025526272000091.jpg510, in which the method of grid search (neural network grid search) was used.

[0063] The algorithm feature selection step (160) is performed before the algorithm training step (170), and the algorithm feature selection step (160) includes a training feature dataset ( a selected training feature dataset ( t ) containing the values of M features for at least one slice n of the rotary kiln ( 1 ) from JPEG2025526272000092.jpg46(t), m=1,…,M*, t<0); JPEG2025526272000093.jpg46(t), m=1,...,M, t<0, M≦M*). In this example, by using correlation-based feature selection (CBFS), the above value of M*=150 features for the rotary kiln (1) can be reduced to M=20 features.

[0064] In the algorithm training step (170), algorithm A for slice n of the rotary kiln (1) is n is the selected training feature dataset ( JPEG2025526272000094.jpg46(t), m=1,…,M, t<0, M≦M*) and a training external temperature dataset (T n Algorithm A for slice n of a rotary kiln (1) trained by using (t), t<0) n * is obtained. Training is achieved by supervised learning. Mathematically, this means that the scoring function E (summed over all time points) n (t) is minimized, converges, or at least falls below a certain threshold. JPEG2025526272000095.jpg57) (in this specific example, the optimization achieved a minimum RMSE of 4.05°C). In such a trained algorithm, the output of the algorithm is generally a function of the training external temperature dataset (T n (t), t<0). The optimized model parameters ( JPEG2025526272000096.jpg813) and its set of hyperparameters The algorithm was trained using JPEG2025526272000097.jpg69, and the model / prediction model It is called JPEG2025526272000098.jpg913, which is a combination of the input features and the time point ( The output of a previously trained algorithm (trained on November 21, 2020, t=-10d, d=day) without prior knowledge of the external temperature dataset (from t=-10d to t=0, d=day) is shown in Figure 4 by the dotted line marked "p" for "Prediction" in the region from t=-10d (i.e., from November 21, 2020) to t=0 (i.e., to December 1, 2020). We use this output from the previously trained model as a prior prediction and use it to predict the resulting external temperature dataset (T n(t), the solid line marked "m" in "Measurement" for the region from t=-10d to t=0 (d=day). As can be seen from this comparison, the previously trained model is able to measure the external temperature dataset T n We were able to predict (t) very well indeed.

[0065] Prediction feature dataset ( In a further predictive feature provision step (180) to provide JPEG2025526272000100.jpg46(t), m=1, ..., M, t≧0, the same M=20 features are provided for the future (t=0 to t=+20d, d=days). These features are taken based on the production plan of the rotary kiln (1).

[0066] In the temperature prediction step (190), the external temperature data set (T n (t), t≧0) is the prediction model A n *By using the predicted feature dataset ( JPEG2025526272000101.jpg46(t), m=1,…,M, t≧0), so JPEG2025526272000102.jpg9105. Figure 4 shows the output of the (newly) trained algorithm (trained on December 1, 2020, t=0, dotted line marked with "p" for "prediction"), which is based on the previously trained model plus the obtained external temperature dataset T for the region t=-10d to t=0 (d=day). n (t) is the (additional) training external temperature dataset (T n (t), t<0). This updated and trained algorithm will be used to measure future external temperature data sets (T n (t), from t=0 to t=20d, d=days). After a certain point in time, say December 11, 2020 (thus t=10d), an additional external temperature data set T for the region from t=0d to t=10d (d=days) is used. n(t) is the (additional) training external temperature dataset (T n Another update to this trained model may be made by using (t), t<0).

[0067] A specific predetermined temperature range (in this example, T min =340℃~T max = 370°C). n If the temperature (t), t≧0) is outside this temperature range, an alarm can be triggered to allow the operator to react in a timely manner. In situations where excessive external temperatures are predicted, the operator may change the operating parameters of the kiln. Alternatively, the operator may plan to repair or renew the refractory lining in the kiln well before a critical situation due to high temperatures occurs. Such situations are illustrated in Figure 4, where the first situation occurs at t=5d (marked 1 in Figure 4), where the predicted external temperature data set (T n (t, t≧0) is T min = 340°C, which is only a few degrees Celsius below the lower threshold. In this situation where a temperature peak down is predicted, the operator may change the operating conditions to prevent such a temperature peak down, thereby preventing or at least reducing refractory wear due to thermal shock, and as a result, achieving an extension of the refractory life.

[0068] The second situation is shown in Figure 4, where at t = 12d (marked 2 in Figure 4), the predicted external temperature data set (T n (t, t≧0) is T max= 370 °C by just a few degrees. The operator may react by changing the kiln's operating parameters to prevent this second situation. In even more severe situations, the operator may have to react to prevent damage to the rotary kiln shell (3). The operator's reaction can be planned well before such a situation occurs, which may prevent the situation altogether or allow planned repairs to be scheduled with reduced downtime. In contrast, if such a situation occurs without advance notice, it could potentially lead to emergency shutdowns and unscheduled maintenance with significant downtime and production loss.

[0069] In this regard, Figure 2 also shows the external temperature (T n 1 shows an example of a prediction system 40 having a prediction unit 30 for predicting (t), t≧0. The prediction unit 30 in this example is a computer.

[0070] The prediction unit (30) receives the training feature dataset ( JPEG2025526272000103.jpg46(t), m=1,...,M*, t<0), and the feature providing unit is provided with a training feature interface (310) for receiving a feature dataset ( ) including values of M*=150 features of the rotary kiln (1) received from a production information management system and / or a laboratory information management system of the rotary kiln (1). JPEG2025526272000104.jpg46(t), m=1,…,M*, t<0).

[0071] The prediction unit (30) calculates the training external temperature data set (T n(t), t<0), in this example a training external temperature data set (T n (t), t<0) is obtained by at least one temperature monitoring unit (20), here an infrared camera (25). Here, the training temperature interface (320) receives the external temperature data set (T n (t), t<0) are collected over time.

[0072] The prediction unit (30) comprises a temperature prediction interface (390) for providing an external temperature data set (Tn(t), t≧0) for slice n of the rotary kiln (1), which is connected to a temperature prediction display device (490), in this case a display, for displaying information based on the external temperature data set (Tn(t), t≧0) for slice n of the rotary kiln.

[0073] The prediction unit is programmed to perform the steps described above in relation to the first aspect of the invention.

[0074] 2 shows a rotary kiln system (50) including a rotary kiln (1) having a rotary kiln shell (3), which is equipped with a production information management system, a laboratory information management system, and the above-mentioned prediction unit (30). Here, the feature providing unit (10) receives a feature dataset ( ) including values of M*=150 features of the rotary kiln (1) from the production information management system and laboratory information management system of the rotary kiln (1). JPEG2025526272000105.jpg46(t), m=1,...,M*, t<0, the temperature monitoring unit (20) receives the slice n of the rotary kiln and generates an external temperature data set (T n(t), t<0). Also shown in FIG. 2 is a longitudinal axis L of the rotary kiln (1), a z-direction extending along the direction of the longitudinal axis L, and an azimuth angle of the rotary kiln (1). JPEG2025526272000106.jpg56, and slice n (n=1,...,N) of the rotary kiln (1) are also shown.

[0075] The temperature prediction display device (490) of this example presents a graph as shown in FIG. 4 to the operator. Also, an external temperature data set (T n A warning message is also shown if (t), t≧0) is outside the predetermined temperature range.

[0076] In this example, the method according to the first aspect of the invention is performed in parallel on N=3000 slices. [Explanation of symbols]

[0077] 1. Rotary kiln 2. The (outer) surface of the rotary kiln shell 3 Rotary Kiln Shells 10 Feature Providing Unit 20 Temperature Monitoring Unit 25 Infrared Camera 30 prediction units 40 Prediction System 50 Rotary Kiln System 100 ways 110 Training feature provision step 120 training temperature provision steps 125 pre-processing steps 130 Model Setup Steps 140 Scoring Function Setup Steps 150 Algorithm Selection Steps 160 Algorithm Feature Selection Step 170 Algorithm Training Steps 180 Prediction feature provision step 190 Temperature Prediction Steps 310 Training Feature Interface 320 Training Temperature Interface 390 Temperature Prediction Interface 490 Temperature Prediction Display Device L Longitudinal axis of rotary kiln 1 Time t t=t min Start time of the training dataset t=0 End time of training dataset / Start time of prediction t≧0 prediction period n is the slice index, n=1,…,N m feature indices, m=1,…,M q are the indexes of the segments in the slice, q=1,…,Q JPEG2025526272000107.jpg612External temperature of the qth segment in the nth slice of the rotary kiln shell at time t T n (t) A temperature data set containing temperature data for each of N slices, n=1,…,N, of the rotary kiln shell. A n nth slice algorithm JPEG2025526272000108.jpg46(t) A training feature dataset containing M features (m = 1, ..., M) for each of N slices (n = 1, ..., N) of the rotary kiln at time t. JPEG2025526272000109.jpg57 A set of K model parameters for slice n JPEG2025526272000110.jpg956, k=1,…,K JPEG2025526272000111.jpg57 A set of K optimized model parameters for slice n JPEG2025526272000112.jpg531, k=1,…,K JPEG2025526272000113.jpg69 Set of J hyperparameters for slice n JPEG2025526272000114.jpg1058, j=1,…,J E n (t) Scoring function JPEG2025526272000115.jpg56 Azimuth angle of the cylindrical coordinate system of a rotary kiln

Claims

1. External temperature (T) of the rotary kiln shell (3) for at least one slice n of the rotary kiln (1) ｎ A method (100) for predicting (t), t≧0), (a) Training feature dataset ( A training feature provision step (110) to provide (t), m = 1, ..., M*, t < 0), (b) The training external temperature of the rotary kiln shell for at least one slice n of the rotary kiln (1) is set to the training external temperature dataset (T ｎ A training temperature provision step (120) to provide (t), t < 0), (c) At least one predictive model A for at least one slice n of the rotary kiln (1) ｎ A model setup step (130) for setting up *, (i) At least one algorithm A for the at least one slice n of the rotary kiln (1) from a machine learning algorithm ｎ An algorithm selection step (150) for selecting at least one algorithm A ｎ However, the model parameters ( ) and hyperparameters ( The algorithm selection step (150) includes, (ii) The at least one algorithm A for the at least one slice n of the rotary kiln (1) ｎ The at least one algorithm A for the at least one slice n of the trained rotary kiln (1) ｎ The training feature dataset that brings about the above ( (t), m=1, ..., M, t<0, M≦M*) and the training external temperature dataset (T ｎ An algorithm training step (170) for training by using at least a portion of (t), t < 0), (iii) The algorithm training step (170) is based on supervised learning, (iv)The at least one algorithm A for the at least one slice n of the rotary kiln (1) in the trained state ｎ The prediction model A including ｎ * is obtained Model setup step (130), including, (d) Prediction feature dataset ( Step (180) to provide predictive features for (t), m = 1, ..., M, t ≥ 0, (e) External temperature data set (T) for at least one slice n of the rotary kiln (1) ｎ (t), t≧0) is defined in the prediction model A ｎ By using *, the aforementioned predictive feature dataset ( A temperature prediction step (190) to predict based on (t), m = 1, ..., M, t ≥ 0) and A method (100) having the following characteristics.

2. A preprocessing step (125) for preprocessing at least one dataset, wherein the at least one dataset is a training feature dataset ( (t), m=1,...,M*, t<0), training external temperature dataset (T ｎ A preprocessing step (125) selected from the list of (t), t < 0). The method according to claim 1, further comprising:

3. The method according to claim 2, wherein the preprocessing step (125) for preprocessing at least one dataset includes checking the data format and data structure of the dataset, and / or checking the consistency of the dataset, and / or extracting outliers from the dataset, and / or removing duplicates within the dataset, and / or imputing missing values.

4. The method according to claim 2, wherein the preprocessing step (125) for preprocessing at least one dataset includes scaling the values ​​in the dataset.

5. At least one Model A for at least one slice n of the rotary kiln (1) ｎ The aforementioned model setup step (130) for setting up * is, The algorithm feature selection step (160) preceding the algorithm training step (170) is the algorithm feature selection step (160) in which the training feature dataset ( From (t), m=1, ..., M*, t<0), selected training feature dataset ( This includes a selection of M features that provide (t), m = 1, ..., M, t < 0, M ≤ M*), The algorithm feature selection step (160) includes at least one algorithm selected from correlation-based feature selection (CBFS), mutual information-based feature selection (MIFS), sequential forward selection (SFS), and sequential backward selection (SBS). The method according to claim 1, further comprising:

6. The method according to claim 1, wherein in the algorithm selection step (150), the selected machine learning algorithm is selected from among decision trees (DT), random forests (RF), support vector regression (SVR), neural networks (NN), gradient boosting regression, and voting regression.

7. At least one predictive model A for at least one slice n of the rotary kiln (1) ｎ The aforementioned model setup step (130) for setting up * is, Scoring function E for at least one slice n of the rotary kiln (1) ｎ A scoring function setup step (140) for setting up (t), wherein the scoring function is selected from RMSE, trend, weighted scoring function. The method according to claim 1, further comprising:

8. At least one Model A for at least one slice n of the rotary kiln (1) ｎ The model setup step (130) for setting up * is a scoring function E for the at least one slice n of the rotary kiln (1). ｎ The scoring function setup step (140) includes setting up a scoring function for at least one algorithm A for the at least one slice n of the rotary kiln (1). ｎ The algorithm setup step (130) for setting up the scoring function E is repeated for different sets of hyperparameters (i), and the scoring function E ｎ （ｉ） Different models A have different results. ｎ （ｉ） This brings about the scoring function E ｎ （ｉ＊） Model A has the best results. ｎ （ｉ＊） The aforementioned prediction model A ｎ The method according to claim 1, which is selected as *.

9. Predictive feature dataset ( The prediction feature provision step (180) for providing (t), m = 1, ..., M, t ≥ 0) is, (a) Based on the production plan of the rotary kiln (1), the predictive feature dataset ( To provide (t), m = 1, ..., M, t ≥ 0), or (b) The training feature dataset ( Based on the continuation of values ​​from (t), m = 1, ..., M, t < 0, M ≤ M*), the predicted feature dataset ( To provide (t), m = 1, ..., M, t ≥ 0), or (c) The aforementioned training feature dataset ( Based on a subset of values ​​from (t), m = 1, ..., M, t < 0, M ≤ M*), the predicted feature dataset ( Provide (t), m = 1, ..., M, t ≥ 0 The method according to claim 1, including the method described in claim 1.

10. The training external temperature dataset (T) provided in the training temperature provision step (120) ｎ The method according to claim 1, wherein (t), t < 0) includes temperature data of the outer surface (2) of the rotary kiln shell (3) for the at least one slice n of the rotary kiln (1) obtained by at least one temperature monitoring unit (20), the temperature monitoring unit (20) includes at least one of an infrared camera (25), a thermocouple, or an optical fiber temperature sensor.

11. The training external temperature dataset (T) provided in the training temperature provision step (120) ｎ The method according to claim 1, wherein (t), t < 0) includes temperature data averaged within the at least one slice n of the rotary kiln (1).

12. External temperature (T) of the rotary kiln shell (3) for at least one slice n of the rotary kiln (1) ｎ A prediction unit (30) for predicting (t), t≧0), (a) Training feature dataset of the rotary kiln (1) A training feature interface (310) for receiving (t), m = 1, ..., M*, t < 0), (b) A training external temperature dataset (T) including external temperature data of the rotary kiln shell (3) for at least one slice n of the rotary kiln (1) ｎ A training temperature interface (320) for receiving (t), t < 0), (c) External temperature data set (T) for at least one slice n of the rotary kiln (1) ｎ A temperature prediction interface (390) to provide (t), t≧0) and Equipped with, The prediction unit (30) is (d) From the training feature interface (310) the training feature dataset ( (t), m=1, ..., M*, t<0) provided (110), (e) From the training temperature interface (320) to the training external temperature dataset (T ｎ (t), t < 0) provide (120), (f) at least one Model A for at least one slice n of the rotary kiln (1) ｎ *of, (i) At least one algorithm A for the at least one slice n of the rotary kiln (1) from a machine learning algorithm ｎ (150) to select the at least one algorithm A ｎ However, the model parameters ( ) and hyperparameters ( ) including, (ii) The at least one algorithm A for the at least one slice n of the rotary kiln (1) ｎ The at least one algorithm A for the at least one slice n of the trained rotary kiln (1) ｎ The training feature dataset that brings about the above ( (t), m=1, ..., M, t<0, M≦M*) and the training external temperature dataset (T ｎ Training by using at least a portion of (t), t < 0) (170), (iii) The at least one algorithm A ｎ The training (170) is based on supervised learning, (iv) The at least one algorithm A for the at least one slice n of the rotary kiln (1) in a trained state. ｎ , and the set of hyperparameters Predictive model A including ｎ * obtain Set up by (130), (g) Prediction feature dataset ( (t), m = 1, ..., M, t ≥ 0) provided (180), (h) External temperature (T) of the at least one slice n of the rotary kiln (1) ｎ (t), t≧0) is defined in the prediction model A ｎ By using *, the aforementioned predictive feature dataset ( Predict based on (t), m = 1, ..., M, t ≥ 0) (190) A prediction unit (30) is programmed to do so.

13. A preprocessing step (125) for preprocessing at least one dataset, wherein the at least one dataset is a training feature dataset ( The prediction unit (30) according to claim 12, further programmed to perform a preprocessing step (125) selected from (t), m=1, ..., M*, t<0) and a training external temperature dataset (T n(t), t<0).

14. External temperature (T) of the rotary kiln shell (3) for at least one slice n of the rotary kiln (1) ｎ A prediction system (40) for predicting (t), t≧0), (a) The prediction unit (30) according to claim 12, (b) Training feature dataset for at least one slice n of the rotary kiln (1) A feature providing unit (10) for providing (t), m=1, ..., M*, t<0) to the training feature interface (310) of the prediction unit (30), (c) A training external temperature dataset (T) including temperature data of the outer surface (2) of the rotary kiln shell (3) for at least one slice n of the rotary kiln (1) ｎ A temperature monitoring unit (20) for providing (t), t < 0) to the training temperature interface (320) of the prediction unit (30), wherein the temperature monitoring unit (20) includes at least one of an infrared camera (25), a thermocouple, or an optical fiber temperature sensor, (d) The external temperature dataset (T) for at least one slice n of the rotary kiln (1) provided by the temperature prediction interface (390) of the prediction unit (30). ｎ A temperature prediction display device (490) for displaying information based on (t), t≧0) and A prediction system (40) equipped with the following features.

15. (a) A rotary kiln (1) having a rotary kiln shell (3) equipped with a production information management system and / or a laboratory information management system, (b) The prediction system (40) according to claim 14 and Equipped with, The feature providing unit (10) receives a feature dataset from the production information management system and / or laboratory information management system of the rotary kiln (1). The temperature monitoring unit (20) receives an external temperature dataset (T), m = 1, ..., M*, t < 0), and the temperature monitoring unit (20) receives an external temperature dataset (T), which includes external temperature data of the rotary kiln shell (3) for at least one slice n of the rotary kiln (1). ｎ A rotary kiln system (50) that provides (t), t < 0).

Images