Multi-model fusion sintering end point temperature prediction method

Abstract

The invention relates to the field of iron ore sintering production process prediction and control, in particular to a multi-model fusion sintering end point temperature prediction method. The method comprises the following steps: firstly, performing Spearman level correlation analysis on currently obtained cleaned data, and extracting strong and weak characteristics of the correlation; secondly, clustering the time series data by using a fuzzy C-means clustering algorithm, and dividing the data into different categories; and for each clustering category, respectively constructing a long-short-term memory network for training, and forming a plurality of prediction models based on different data features. And finally, fusing the prediction results of the plurality of LSTMs in a fuzzy membership weighting mode by using the time sequence data of each air bellow, so as to obtain a final sintering end point temperature prediction value. According to the method, through organic fusion of multiple models, high-accuracy and high-robustness result prediction is realized, managers and workers are helped to accurately grasp the production condition, the sintering result is maintained to be stable, and the production efficiency and economic benefits of enterprises are improved.

Description

Technical Field

[0001] The invention relates to the field of prediction and control of iron ore sintering production process, and in particular to a method for predicting sintering end point temperature by integrating multiple models. Background Art

[0002] The sintering endpoint temperature prediction is to use the currently acquired data to predict the future sintering endpoint temperature of the machine during machine operation, so as to guide the staff to adjust the machine production and maintain the stability of the entire production process. The model algorithm is used to predict the sintering endpoint temperature of the production process, so that managers and staff can have a deeper understanding of the production status, maintain the stability of the sintering endpoint temperature, and improve enterprise efficiency and benefits.

[0003] As a traditional industrial process, the sintering process has always attracted the attention of scholars. It is a time-consuming process with complex procedures. Due to the large number of parameters, it is difficult for operators to adjust the operating status in real time. If there is a problem with the production process, a batch of products will be wasted at best, and the operation of the entire machine will be problematic at worst. This has hindered the improvement of sintered ore quality and the saving of production costs. Therefore, the development of the prediction of the result of sintering endpoint temperature has important economic value.

[0004] When the entire production process deviates from the normal sintering trend due to environmental, material and other reasons, how to use the current known data to predict the future sintering endpoint temperature and provide the staff with information on the current production line status is an urgent problem that needs to be solved. Summary of the invention

[0005] In order to solve the problem of low detection accuracy of sintering endpoint temperature and difficulty in knowing the status of the production line, the present invention provides a method for predicting sintering endpoint temperature by multi-model fusion, which mainly includes:

[0006] S1: preprocessing the time series data of sintering endpoint and the time series data of detection parameters;

[0007] S2: Use the Spearman rank correlation model to analyze the correlation strength of various time series data and build a feature library;

[0008] S3: Based on the constructed feature library, each time series data is clustered using fuzzy C-means clustering;

[0009] S4: A multi-model fusion long short-term memory network is established based on the clustering results. The results output by multiple long short-term memory networks are weighted summed according to the membership of different clusters based on the data, and then the sintering endpoint temperature is predicted.

[0010] A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0011] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above method.

[0012] A computer program product comprises a computer program or instructions, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0013] The technical solution provided by the present invention has the following beneficial effects: the present invention uses the currently obtained cleaned data to perform Spearman rank correlation analysis to extract correlation features. Then, the time series data is clustered using the fuzzy C-means clustering algorithm to divide the data into different categories. Then, for each clustering category, a long short-term memory network (LSTM) model is constructed for training to form multiple prediction models based on different data features. Finally, using the time series data of each bellows, the prediction results of multiple LSTM models are fused by fuzzy membership weighting to obtain the final sintering endpoint temperature prediction value. The present invention extracts feature correlation by using the Spearman rank correlation analysis algorithm, thereby eliminating interference features and making the prediction results more accurate. FCM is used to cluster the data to obtain membership, and the result is obtained by weighted summation of multiple long short-term memory networks, which has good robustness and stability. Through the organic fusion of multiple models, high-accuracy and high-robustness result prediction is achieved, so that managers and staff can have a deeper understanding of the production status, maintain the stability of the sintering results, and improve enterprise efficiency and benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0015] Figure 1 It is a flow chart of a method for predicting sintering endpoint temperature by integrating multiple models in an embodiment of the present invention;

[0016] Figure 2 Schematic diagram of feature correlation ranking calculated by Spearman rank correlation analysis in an embodiment of the present invention;

[0017] Figure 3 It is a time series comparison diagram of the prediction model of multi-model fusion in the embodiment of the present invention for predicting the sintering end point temperature in the next 30 minutes;

[0018] Figure 4 It is a schematic diagram of the operation of the hardware device in the embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0020] Example 1

[0021] Please refer to Figure 1 , Figure 1 This is a flow chart of a method for predicting the sintering end temperature by fusion of multiple models in an embodiment of the present invention. First, the present invention marks the original data by obtaining the sintering end temperature. Then, the sintering end temperature is predicted using the acquired time series data of each wind box. The specific steps are as follows:

[0022] (1) Collect sintering production historical data to obtain original sample data

[0023] The sintering production history data is stored in the local database of the industrial computer in the operation room in the form of daily reports. According to the data in the daily reports, the historical data of one month, such as the sintering end point and the exhaust gas temperature of the i-th wind box (i = 1, 2, 3, 5, 7, 9, 11, 13, 15, 17, 18, 19, 20, 21, 22, 23, 24), are collected to form the original sample data. A total of 20 test parameters.

[0024] (2) Preprocessing the time series data of sintering endpoint and detection parameter

[0025] The collected raw sample data are preprocessed, specifically, zero detection data caused by sensor failure and shutdown are eliminated; the data sampling interval is 60 seconds, and for the time series data of the sintering endpoint and the time series data of the detection parameters, the length of each time subsequence is 10 minutes, including 10 data points, and a sample database is established with these sample data;

[0026] (3) Building a feature library

[0027] Now there are 17 bellows with different temperatures, namely T 1 、T 2 、T 3 、T 5 、T 7 、T 9 、T 11 、T 13 、T 15 、T 17 、T 18 、T 19 、T 20 、T 21 、T 22 、T 23 、T 24 ;

[0028] The acquired bellows temperature data is used as the characteristic variable of the sample, and the Spearman rank correlation coefficient is calculated to analyze and determine certain characteristic variables with strong correlation. In this embodiment, the time series data with a Spearman rank correlation coefficient greater than 0.5 is taken as the characteristic variable with strong correlation.

[0029] For two original variables A = [a 1 ,a 2 ,…,a n ] and B = [b 1 ,b 2 ,…,b n ], their levels are α=[α 1 ,α 2 ,…,α n ] and β = [β 1 ,β 2 ,…,β n ], then the Spearman rank correlation is calculated as follows,

[0030]

[0031] Where n is the total number of time series data and ρ is the Spearman rank correlation coefficient.

[0032] The correlation of various types of time series data calculated using the Spearman rank correlation analysis algorithm is used to screen each feature, and all time series data with a correlation greater than 0.5 are used as a feature library.

[0033] (4) Clustering each time series data using fuzzy C-means clustering

[0034] (4-1) From step (2), we get a time subsequence with a length of 10 minutes and a sampling interval of 1 minute. Let a set of time series data be X = {x 1 , x 2 , x 3 , …, x 10}, where each x i ∈R d It is represented as a time series with a length of d (d=1).

[0035] (4-2) For each iteration t = 1, 2, ..., T, we can calculate the membership matrix U at the previous moment. (t-1) Calculate the cluster center matrix V at this moment (t) , where the cluster center is calculated according to the following formula:

[0036]

[0037] is the degree of membership corresponding to the data point at the previous moment, x iis the time series data at that moment, m is the fuzzy coefficient, and C is the number of clusters, which can be evaluated by the fuzzy coefficient.

[0038] (4-3) Since the sampling intervals of the time series data are equal, the Euclidean distance is used to calculate the data point x i To Center The distance matrix D (t) , where the Euclidean distance is calculated as follows:

[0039]

[0040] is the cluster center point corresponding to the data at that moment, x i is the time series data at that moment, and m is the fuzzy coefficient.

[0041] (4-4) The distance matrix D calculated according to step (4-3) (t) Update the membership matrix U at this moment (t) , where the membership degree is calculated as follows:

[0042]

[0043] in, is the Euclidean distance of the data point at that moment, and m is the fuzzy coefficient.

[0044] (4-5) The membership matrix U at this moment calculated according to step (4-4) (t) and the membership matrix U at the previous moment (t-1) Compare, if the maximum number of iterations T or the minimum error is reached, stop; otherwise repeat (4-2), (4-3), (4-4) for iteration. Output the cluster center V and membership matrix U.

[0045] (5) Sintering endpoint temperature prediction experiment

[0046] (5-1) Design the structure of the long short-term memory network, which consists of two hidden layers, an input layer, and an output layer. Set the parameters of each layer to 128 and 64 neurons respectively.

[0047] (5-2) Build the overall model architecture, divide the results obtained from different clustering into training sets and test sets according to the time window, standardize the training set data and construct time series samples; use the time series samples to train the long short-term memory network through the back propagation through time (BPTT) algorithm, use the gradient descent method to update the weights of the long short-term memory network, and use the Adam optimizer to update the network parameters until the set iteration rounds are reached to obtain a trained long short-term memory network.

[0048] (5-3) Train multiple long short-term memory network models separately to obtain different time series feature extraction capabilities. Suppose the current data is X t , whose temporal association weights under different clusters are w 1 ,w 2 ,…,w n , input the data into each long short-term memory network model for multi-step rolling prediction, and obtain the corresponding temperature prediction sequence T 1 ,T 2 ,…,T n , according to the time series association weight, dynamic fusion is performed, and the final prediction result is,

[0049]

[0050] Where T is the predicted sintering end temperature result, avg(T i ) represents the weighted average of the prediction sequence of the i-th LSTM network model, and the weight is adaptively adjusted by the historical prediction error.

[0051] The prediction results of multiple LSTM models are fused by fuzzy membership weighting to obtain the final prediction value of sintering endpoint temperature. Figure 2 Output the prediction result of the sintering end point in the next 30 minutes, as shown in Figure 3 As shown, it can be seen that the method can be used to predict the sintering endpoint temperature well, which has important economic value and application value.

[0052] Example 2

[0053] like Figure 4 As shown, a computer device includes a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to implement the steps of the above method.

[0054] Example 3

[0055] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above method.

[0056] Example 4

[0057] A computer program product comprises a computer program or instructions, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting sintering endpoint temperature by fusion of multiple models, characterized in that: include: S1: preprocessing the time series data of sintering endpoint and the time series data of detection parameters; S2: Use the Spearman rank correlation model to analyze the correlation strength of various time series data and build a feature library; S3: Based on the constructed feature library, each time series data is clustered using fuzzy C-means clustering; S4: A multi-model fusion long short-term memory network is established based on the clustering results. The results output by multiple long short-term memory networks are weighted summed according to the membership of different clusters based on the data, and then the sintering endpoint temperature is predicted.

2. The method for predicting sintering endpoint temperature by multi-model fusion according to claim 1, characterized in that: The preprocessing described in S1 is to remove zero detection data caused by sensor failure and downtime, set the sampling interval of the data, and select the length of each time subsequence and the number of data points.

3. The method for predicting sintering endpoint temperature by multi-model fusion according to claim 1, characterized in that: S2 includes the following processes: S2.1: Get the temperatures of several wind boxes: T1, T2, T3, T5, T7, T9, T 11 , T 13 , T 15 , T 17 , T 18 , T 19 , T 20 , T 21 , T 22 , T 23 , T 24 ; S2.2: When using the Spearman rank correlation model for analysis, the bellows temperature data obtained in S2.1 is used as the characteristic variable of the sample. By calculating the Spearman rank correlation coefficient, the time series data with a Spearman rank correlation coefficient greater than the preset threshold is taken as the characteristic variable of strong correlation; For two original variables A=[a1,a2,…,a n ] and B=[b1,b2,…,b n ], their levels are α=[α1,α2,…,α n ] and β=[β1,β2,…,β n ], then the Spearman rank correlation is calculated as follows, Where n is the total number of time series data, ρ is the Spearman rank correlation coefficient; S2.3: Screen each feature according to the strength of the calculated correlation, and use the time series data with all relevant indicator scores greater than the preset threshold as the feature library.

4. The method for predicting sintering endpoint temperature by multi-model fusion according to claim 1, characterized in that: S3 includes the following processes: S3.1: Let a set of time series data be X = {x1, x2, x3, ..., x 10 }, where each x i ∈R d Represented as a time series of length d; S3.2: For each iteration e = 1, 2, ..., E, according to the membership matrix U of the previous iteration (e-1) Calculate the cluster center matrix V of the current iteration (e) , where the cluster center is calculated according to the following formula: in, is the membership degree corresponding to the data point in the previous iteration, x i is the time series data of the current iteration, m is the fuzzy coefficient, C is the number of clusters, and n is the number of samples; S3.3: Calculate the data point x using Euclidean distance i To the cluster center The distance matrix D (e) , where the Euclidean distance is calculated as follows: in, is the cluster center point corresponding to the data at the current moment, is the data point x i To the cluster center distance; S3.4: Distance matrix D calculated according to step S3.3 (e) Update the membership matrix U at the current moment (e) , where the membership degree is calculated as follows: in, is the Euclidean distance of the data point at the current moment, is the degree of membership of the data point to the jth cluster center, and k represents the kth clustering; S3.5: The membership matrix U at the current moment (e) and the membership matrix U at the previous moment (e-1) Compare and if the maximum number of iterations E or the minimum preset error is reached, the iteration stops; otherwise, repeat steps S3.2, S3.3, and S3.4 until the iteration stop condition is met, and output the cluster center V and membership matrix U.

5. The method for predicting sintering endpoint temperature by multi-model fusion according to claim 1, characterized in that: S4 includes the following processes: S4.1: Designing the Structure of Long Short-Term Memory Networks The network consists of an input layer, two stacked long short-term memory network hidden layers and a fully connected output layer, where the long short-term memory network hidden layers contain 128 and 64 memory units respectively. The input layer receives the time series data of the historical sintering process, and the output layer provides the predicted value of the terminal temperature. S4.2: Building a time series data processing framework The clustering results are divided into training sets and test sets according to the time window, and the training set data is standardized and time series samples are constructed; the time series samples are used to train the long short-term memory network through the time algorithm through back propagation, and the weights of the long short-term memory network are updated by the gradient descent method. The network parameters are updated by the Adam optimizer until the preset iteration rounds are reached to obtain the trained long short-term memory network model; S4.3: Train multiple long short-term memory networks separately to obtain different temporal feature extraction capabilities; Let the current data be X t , whose temporal association weights under different clusters are w1,w2,…,w n , input the data into each long short-term memory network for multi-step rolling prediction, and obtain the corresponding temperature prediction sequence T1, T2, ..., T n , according to the time series association weight, dynamic fusion is performed, and the final prediction result is, Where T is the predicted sintering end temperature result, avg(T i ) represents the weighted average of the i-th LSTM network prediction sequence, and the weight is adaptively adjusted by the historical prediction error.

6. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method for predicting the sintering endpoint temperature by multi-model fusion as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that: A computer program is stored, and when the program is executed by a processor, the steps of the method for predicting the sintering endpoint temperature by multi-model fusion as described in any one of claims 1 to 5 are implemented.

8. A computer program product, characterized in that The method comprises a computer program or an instruction, which, when executed by a processor, implements the steps of the method for predicting the sintering endpoint temperature by fusion of multiple models as described in any one of claims 1 to 5.

Images