Grate cooler working condition identification method
Through the combination of Double-BIRCH clustering and GBDT algorithm, the problem of inaccurate division of grate cold machine working conditions in the existing technology is solved, and the accurate identification and optimization control of the complex working conditions of grate cold machine is realized, which significantly improves the accuracy of recognition Rate.
Patent Information
- Application Number
- CN202411063544.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-12-06
AI Technical Summary
The existing single-layer clustering method cannot accurately divide the working conditions of the grate cold machine, making it difficult to achieve accurate evaluation and optimization control of the operating status of the grate cold machine.
The double-layer Double-BIRCH clustering method and gradient-enhanced decision tree algorithm GBDT are used to preprocess, cluster and multiple iterative training of the working condition data to build a strong classifier to identify the working condition of the grate cold machine.
The accurate identification of complex and variable working conditions of the grate cold machine is achieved, which significantly improves the accuracy of working conditions and the accuracy of identification, and can quickly identify the differences between large and small working conditions.
Smart Images

Figure CN119089315A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of grate cooler working condition identification methods, in particular to a grate cooler working condition identification method. Background Art
[0002] The smooth operation of the grate cooler is directly related to the energy efficiency of cement production, and accurate control of its working condition is the key to improving cement quality and cooling efficiency. During the operation of the grate cooler, there are problems such as frequent operating fluctuations and nonlinearity, which cause various working conditions of the grate cooler to appear alternately. In order to evaluate the operating status of the grate cooler and optimize the control goals for complex working conditions, it is necessary to divide the working conditions accordingly, but the existing single-layer clustering method cannot accurately divide the working conditions. Summary of the invention
[0003] The present invention provides a method for identifying the working condition of a grate cooler, so as to solve the problem that the single-layer clustering method in the prior art cannot accurately divide the working condition of the grate cooler.
[0004] In order to achieve the above object, the technical solution adopted by the present invention is:
[0005] A method for identifying a grate cooler operating condition comprises the following steps:
[0006] Step 1, obtaining the working condition data of various working conditions of the grate cooler, selecting key parameters from the working condition data of various working conditions as detection parameters, and constructing a working condition data set based on the detection parameters;
[0007] Then, after preprocessing the data in the working condition data set, the working condition labels of all the data in the working condition data set are obtained;
[0008] Step 2: Using the gradient boosting decision tree algorithm GBDT to perform multiple iterations of training based on the working condition data set with working condition labels obtained in step 1, a strong classifier consisting of multiple trained CART trees is obtained as a working condition recognition model;
[0009] Step 3: input the working condition data to be identified of the grate cooler into the working condition identification model obtained in step 2, and the working condition identification module predicts the working condition label as the working condition identification result.
[0010] In the further step 1, the preprocessing is to standardize the data in the working condition data set.
[0011] In a further step 1, the Double-BIRCH algorithm is used to cluster the data in the preprocessed working condition data set, thereby obtaining the working condition labels of all the data in the working condition data set.
[0012] Compared with the prior art, the present invention has the following advantages:
[0013] 1. Compared with the conventional working condition division algorithm, the present invention can accurately divide the large working condition that is greatly different from the normal fluctuating working condition data through the double-layer Double-BIRCH clustering method, and can also accurately divide the small working condition that is relatively small different from the normal fluctuating working condition data.
[0014] 2. Compared with conventional classification methods, the present invention uses the gradient boosting decision tree algorithm GBDT to train the obtained working condition labels to obtain the corresponding working condition recognition model. It has a significant effect on the accuracy of grate cooler working condition recognition and can quickly and accurately recognize the complex and changeable working conditions of the grate cooler. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 This is a diagram of the working condition division results of the Double-BIRCH algorithm in an embodiment of the present invention.
[0016] Figure 2 It is a principle block diagram of the division and identification of the working conditions of the grate cooler in the embodiment of the present invention. DETAILED DESCRIPTION
[0017] The present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0018] like Figure 2 As shown, this embodiment discloses a method for identifying the working condition of a grate cooler, comprising the following steps:
[0019] Step 1: Obtain the operating data of various operating conditions of the grate cooler, select key parameters from the operating data of various operating conditions as detection parameters, and construct an operating data set based on the detection parameters.
[0020] In this embodiment, the operating data of the grate cooler in five working conditions, namely, material cut-off condition, material reduction condition, large peak grate pressure condition, small peak grate pressure condition, and grate pressure stable fluctuation condition, are used as examples for explanation. First, according to the process mechanism of the cement burning system, the grate pressure difference ΔP, the kiln main machine current I hzy and raw material feed rate W as detection parameters, the operating condition data set X is composed of the selected detection parameter data, where ΔP = PP s , P is the actual grate pressure, P s Set the value for the grate down pressure.
[0021] Then, considering that the selected detection parameters have different dimensional attributes, this embodiment performs standardized preprocessing on the data in the working condition data set. The standardized preprocessing is shown in formula (1):
[0022]
[0023] In formula (1), x iis the feature in the original data, which is the data in the working condition data set X in this embodiment; i ′ is the value after standardization; E is the mean of the feature; σ is the standard deviation of the feature.
[0024] Finally, the Double-BIRCH algorithm is used to cluster the data in the preprocessed working condition data set, thereby obtaining the working condition labels of all data in the working condition data set. The process is as follows:
[0025] Step S1: Parameter initialization.
[0026] Create an empty clustering feature tree, set the clustering tree radius threshold T and the maximum number of child nodes B; determine the expected number of clusters k1 in the first layer and the expected number of clusters k2 in the second layer.
[0027] Step S2: Insert data.
[0028] Read the first b sample data points from the working condition data set X to generate an initial working condition cluster Q CFb , and put the current sample data point into it, and make it subordinate to the root node.
[0029] Step S3: Insert cluster.
[0030] Read the next sample data point X from the working condition data set X i (i=b+1), calculate the sample data point X i The Euclidean distance from the current leaf node cluster centroid c is used to select the cluster with the closest distance and insert it. The calculation formula of the cluster centroid c is shown in formula (2):
[0031]
[0032] In formula (2), LS represents the sum of the feature vectors with sample operating condition data in this cluster feature, which can be expressed as X j represents the jth sample data point; N represents the number of sample points in the cluster.
[0033] Step S4: Calculate the cluster radius.
[0034] Calculate the cluster radius R of the current sample data i , if R i If it is less than or equal to the threshold T, go to step S7; otherwise, go to step S5. i The calculation formula is shown in formula (3):
[0035]
[0036] In formula (3), SS represents the sum of squares of the feature vectors of the sample operating condition data in this cluster feature, which can be expressed as
[0037] Step S5: Split the cluster.
[0038] If the number of clusters under the current leaf node is less than the maximum number of cluster features of the leaf node, a new cluster is generated and the process goes to step S7; otherwise, the process goes to step S6.
[0039] Step S6: Internal node splitting.
[0040] Select the two cluster centroids with the farthest Euclidean distance under the internal nodes as the cluster centroids of the new nodes, and add the next data point X i Insert the clusters closest to the centroids of the two newly added leaf nodes to form new clusters, and then go to step S7.
[0041] Step S7: Update parameters.
[0042] Complete the sample data X i After the insertion, update the parameters of the current cluster, the leaf nodes to which the cluster belongs, and the internal nodes on the leaf nodes.
[0043] Step S8: Continue processing.
[0044] If all sample data points have been inserted into the clustering tree, go to step S9; otherwise, set i=i+1 and return to step S2.
[0045] Step S9: Check and merge the number of clusters.
[0046] According to the number of clusters in the clustering feature tree, the following judgments are made:
[0047] If the number of clusters is less than the preset number of clusters k1, reinitialize the clustering parameters to improve the segmentation ability of the clustering tree; if the number of clusters is greater than or equal to the preset number of clusters k1, decide whether to merge them based on the Euclidean distance between the cluster centroids and the cluster density to achieve the required number of clusters k1, and assign corresponding cluster labels to the data in these k1 clusters.
[0048] Step S10: Select the data marked with the labels that need to be clustered again from the k1 cluster labels obtained in the first layer of clustering as the input of the second layer of clustering, repeat steps 2 to 9, and obtain k2 clusters and assign corresponding cluster labels to them respectively.
[0049] In this embodiment, according to the working condition data of the five working conditions obtained in step 1, k1=3, k2=3. The selected working condition data is preprocessed as the input of the Double-BIRCH algorithm. After the above steps, the grate cooler working condition division of the Double-BIRCH algorithm can be completed, and the clustering labels L0, L1, L2, L3, L4 of the corresponding working conditions can be obtained. The clustering results are as follows: Figure 1 L0 and L1 represent the material cutting condition and material reducing condition in the large working condition, respectively, and L2, L3, and L4 represent the normal fluctuation condition, large "peak shape" condition, and small "peak shape" condition in the small working condition, respectively.
[0050] Step 2: Use the gradient boosting decision tree algorithm GBDT to perform multiple iterative training based on the working condition data set with working condition labels obtained in step 1 to obtain a strong classifier composed of multiple trained CART trees as the working condition recognition model. The specific process is as follows:
[0051] Step (2.1), determine the number of working conditions K involved in the identification. During multiple iterations of training, K CART trees are trained in each iteration. Suppose the i-th sample x in the data set with working condition labels is i The probability that (i=1,2,...,N) belongs to working condition k in the mth (m=1,2,...,M) iteration is Where k = 1, 2, ..., K, then the sample x i The probability p of belonging to the kth operating condition in all iterations ik As shown in formula (4):
[0052]
[0053] Step (2.2), for sample x i , according to the probability p of belonging to the kth operating condition in all iterations ik , calculate the corresponding loss function L(y ik ,p ik ) is shown in formula (5):
[0054]
[0055] In formula (5): y ik is the sample x i Indicator of whether it belongs to working condition k (y ik The value is 0 or 1);
[0056] After calculating the loss function, the negative gradient of the sample is calculated based on the loss function. Sample x i The negative gradient r mk It can be expressed as shown in formula (6):
[0057]
[0058] In formula (6): is the sample x i The probability of belonging to operating condition k in the m-1th iteration;
[0059] Step (2.3), for each CART tree, fit it to the negative gradient of the corresponding category, so that the K CART trees obtained in each iteration correspond to one category respectively. After each iteration, the predicted value is added to the previous predicted probability, and the learning rate λ is used to control the update amplitude. For sample x i and category k, the update amplitude control is shown in formula (7):
[0060]
[0061] In formula (7): Indicates the CART tree model fitted for class k in m iterations for sample x i The predicted value of
[0062] Step (2.4), after M iterations, according to the final predicted probability, select the category with the largest predicted probability as the sample x i The final prediction result As shown in formula (8)
[0063]
[0064] The final prediction result based on formula (8) As the output of the GBDT algorithm after M iterations.
[0065] Step 3: input the working condition data to be identified of the grate cooler into the working condition identification model obtained in step 2, and the working condition identification module predicts the working condition label as the working condition identification result.
[0066] In this embodiment, for the working condition identification scenario of the grate cooler, the data of grate pressure difference, kiln main machine current, and raw material feed amount are selected and clustered by Double-BIRCH algorithm to obtain the working condition label as the input of the working condition identification model established by the GBDT algorithm. The number of working conditions involved in the identification is K=5. The working condition labels are randomly shuffled and then divided into training set and test set in a ratio of 7:3. The final predicted working condition label is the model output. Then the GBDT algorithm of this embodiment and other different classification algorithms are selected, among which the learning rate λ of the GBDT algorithm of this embodiment is 0.2, and the other algorithms are all default parameters. Finally, the performance of each algorithm is evaluated by accuracy, recall rate, FI score and running time. The results are shown in Table 1.
[0067] Table 1. Comparison of different classification algorithms applied to grate cooler working condition recognition performance
[0068] Algorithm indicators Accuracy / % Recall / % FI points / % Time / s GBDT 98.05 97.63 97.88 0.916 KNN 94.11 89.14 91.86 0.046 SVM 87.23 82.70 84.40 0.244 BP 88.82 86.37 87.24 2.325
[0069] It can be seen from Table 1 that the GBDT algorithm of this embodiment has a higher accuracy and recall rate, and the running time is also shorter, which can meet the accuracy requirements during the operation of the grate cooler.
[0070] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention, and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction, and such combinations should also be regarded as the contents disclosed in the present disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.
[0071] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by technical personnel in this field should fall within the protection scope of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.
Claims
1. A method for identifying the working condition of a grate cooler. It is characterized in that The following steps are involved: Step 1, obtaining the working condition data of various working conditions of the grate cooler, selecting key parameters from the working condition data of various working conditions as detection parameters, and constructing a working condition data set based on the detection parameters; Then, after preprocessing the data in the working condition data set, the working condition labels of all the data in the working condition data set are obtained; Step 2: Using the gradient boosting decision tree algorithm GBDT to perform multiple iterations of training based on the working condition data set with working condition labels obtained in step 1, a strong classifier consisting of multiple trained CART trees is obtained as a working condition recognition model; Step 3: input the working condition data to be identified of the grate cooler into the working condition identification model obtained in step 2, and the working condition identification module predicts the working condition label as the working condition identification result.
2. A method for identifying a grate cooler operating condition according to claim 1, It is characterized in that In step 1, the preprocessing is to standardize the data in the working condition data set.
3. A method for identifying the working condition of a grate cooler according to claim 1, It is characterized in that In step 1, the Double-BIRCH algorithm is used to cluster the data in the preprocessed working condition data set, thereby obtaining the working condition labels of all data in the working condition data set.
Citation Information
Cited By
Grate cooler equipment parameter optimization method based on clustering algorithm
CN119758914A
