CN-121998289-A - Cabinet position prediction method, equipment and storage medium for converter gas cabinet

CN121998289ACN 121998289 ACN121998289 ACN 121998289ACN-121998289-A

Abstract

The embodiment of the invention provides a method, equipment and a storage medium for predicting the bin position of a converter gas holder, and belongs to the field of bin position prediction of converter gas holders. The cabinet position prediction method comprises the steps of obtaining historical production data and historical cabinet position data of a converter gas cabinet, calculating the association degree between the historical production data and the historical cabinet position data, screening the historical production data with the association degree meeting a preset threshold value, combining the historical cabinet position data and inputting the historical production data into a cabinet position prediction model together, encoding the input data into a global context vector with a fixed length based on the cabinet position prediction model, and generating a multi-step cabinet position prediction sequence according to the global context vector. The method solves the problem of low prediction accuracy of the traditional time sequence prediction caused by the influence of the change of the working condition on the converter gas cabinet, effectively improves the prediction precision and efficiency, and is suitable for intelligent dispatching and safety control of a converter gas system.

Inventors

ZHOU TONG
XUE YINGJIAN
CHEN LAIJUN
LI YANLONG
GUO YU
LI MINGYU
Cheng Mingfan

Assignees

中冶京诚工程技术有限公司
中冶京诚数字科技(北京)有限公司

Dates

Publication Date: 20260508
Application Date: 20251215

Claims (11)

1. The method for predicting the cabinet position of the converter gas cabinet is characterized by comprising the following steps of: Acquiring historical production data and historical cabinet position data of a converter gas cabinet, wherein the historical production data comprises production and elimination user data and cabinet pressure data of the converter gas cabinet; calculating the association degree between the historical production data and the historical cabinet data, screening the historical production data with the association degree meeting a preset threshold value, and jointly inputting the historical production data and the historical cabinet data into a cabinet prediction model, and And based on the cabinet prediction model, encoding the input data into a global context vector with a fixed length, and generating a multi-step cabinet prediction sequence according to the global context vector.
2. The bin level prediction method according to claim 1, wherein before calculating the degree of correlation between the historical production data and the historical bin level data, the bin level prediction method further comprises: and carrying out data elimination filling and normalization processing on the historical production data and the historical cabinet data.
3. The cabinet prediction method according to claim 2, wherein, The data elimination and filling comprises the steps of replacing data which is judged to be abnormal in the historical production data and the historical cabinet data with average value data of the previous moment and the next moment, and/or filling missing data at a certain moment by a linear interpolation method; the normalization processing comprises the steps of carrying out linear transformation on the data after the data elimination filling processing, mapping the data to between [0,1], wherein the calculation formula is as follows: Wherein, the The data after the normalization is represented as such, Representing the data after the data culling and padding process, Represents the maximum value in the data after the data culling and padding process, Representing the minimum value in the data after the data culling and padding process.
4. The cabinet level prediction method according to claim 1, wherein the calculating the degree of association between the historical production data and the historical cabinet level data includes: calculating a difference value between the historical production data and the historical cabinet position data, and determining an extremum in the difference value; and calculating a correlation coefficient according to the extreme value, and calculating the correlation degree between the historical production data and the historical cabinet data based on the correlation coefficient.
5. The cabinet prediction method according to claim 4, wherein the calculation formula of the association coefficient is: Wherein, the Representing the correlation coefficient(s), Representing the historical cabinet level data, The data representing the production of the history is presented, Representing the resolution factor.
6. The cabinet level prediction method according to claim 4, wherein the correlation between the historical production data and the historical cabinet level data is calculated based on the correlation coefficient, and the calculation formula is: Wherein, the The degree of association is indicated as being indicative of the degree of association, Representing the total length of the data sequence, Represent the first Correlation coefficients for each instant.
7. The cabinet prediction method according to claim 1, wherein the cabinet prediction model is a long-short-term memory network model based on an encoder-decoder framework, and comprises an encoder LSTM and a decoder LSTM, wherein a fully-connected layer of the long-short-term memory network model is embedded into a Dropout layer for suppressing overfitting during training.
8. The bin prediction method of claim 7, wherein encoding the input data into a fixed length global context vector, generating a multi-step bin prediction sequence from the global context vector, comprises: sequentially processing the input data according to time steps through the encoder LSTM, and encoding time sequence characteristics in the input data and influence relations among the data to obtain the global context vector with the fixed length; And the decoder LSTM takes the global context vector as an initial state, gradually generates prediction data of each time step, and outputs the prediction data of all the time steps through a time distribution full-connection layer to obtain the multi-step cabinet position prediction sequence.
9. The method of claim 1, wherein after generating the multi-step bin prediction sequence according to the global context vector, the method further comprises performing inverse normalization calculation on the multi-step bin prediction sequence to obtain an actual prediction value.
10. A cabinet level predicting apparatus of a converter gas cabinet, characterized in that the cabinet level predicting apparatus comprises: Memory, and A processor configured to execute instructions stored in the memory to perform the method of predicting a tank level of a converter gas tank according to any one of claims 1-9.
11. A machine-readable storage medium having stored thereon instructions for causing a machine to perform the method of predicting a bin level of a converter gas tank according to any one of the above-described applications.

Description

Cabinet position prediction method, equipment and storage medium for converter gas cabinet Technical Field The invention relates to the technical field of cabinet position prediction of converter gas cabinets, in particular to a cabinet position prediction method, equipment and storage medium of a converter gas cabinet. Background Converter gas is an important secondary energy source in the steel production process, and the efficient recovery and accurate regulation of the converter gas have important significance for reducing energy consumption and carbon emission. The gas cabinet is used as key buffer equipment and bears the functions of storing and adjusting gas, and the accurate prediction of the cabinet position is the basis for realizing gas balance scheduling and energy management. Because the recovery of the converter gas is intermittent operation, the consumption of the converter gas is also affected by the change of working conditions, so that the change fluctuation of the gas quantity of the gas tank is large, and the accurate prediction is difficult. Currently, manual scheduling operation is mainly relied on in the aspect of gas management, and a scheduler usually predicts through experience. However, the method has the problems of hysteresis and lower prediction precision, and the prediction results are easy to be inconsistent due to the experience difference of operators, so that the scheduling efficiency and accuracy are affected. In the prior art, as disclosed in the patent of publication No. CN103942422A, a metallurgical enterprise converter gas holder based on granularity calculation is disclosed, the method adopts a fuzzy C-means method to cluster data, thereby obtaining membership degree and cluster center, then carries out fuzzy reasoning, and predicts by a central defuzzification method. However, the fuzzy C-means clustering method adopted by the method needs to set initial values and cluster numbers manually, has certain subjectivity, has high data requirement quality as unsupervised learning, and often needs to work in severe environments of high temperature, high pressure and high corrosion, so that the sensor is inevitably lost due to the abnormality of similar data, noise exists and the like, and the directly transmitted data is difficult to meet the requirement as time goes on. Secondly, iteration exists in the prediction process of the traditional autoregressive model, the prediction error can be increased along with the iteration, and the prediction precision is not high. In addition, the traditional autoregressive model is accurate in performance under normal working conditions, but when the working conditions change, the prediction result often deviates, and the accuracy is greatly reduced. Disclosure of Invention The embodiment of the invention aims to provide a method, equipment and a storage medium for predicting the bin position of a converter gas bin, which are used for solving the technical problems in the background art, realizing multi-step prediction of the bin position of the converter gas, providing scientific data support for subsequent dispatching and reducing the running cost of enterprises. In order to achieve the above purpose, the embodiment of the invention provides a cabinet position prediction method of a converter gas cabinet, which comprises the steps of obtaining historical production data and historical cabinet position data of the converter gas cabinet, wherein the historical production data comprise production and elimination user data and cabinet pressure data of the converter gas cabinet, calculating the association degree between the historical production data and the historical cabinet position data, screening out the historical production data with the association degree meeting a preset threshold value, and jointly inputting the historical production data into a cabinet position prediction model by combining the historical cabinet position data, and encoding the input data into a global context vector with a fixed length based on the cabinet position prediction model, and generating a multi-step cabinet position prediction sequence according to the global context vector. Optionally, before calculating the correlation degree between the historical production data and the historical cabinet data, the cabinet prediction method further comprises the steps of carrying out data elimination filling and normalization processing on the historical production data and the historical cabinet data. Optionally, the data elimination and filling comprises the steps of replacing the data judged to be abnormal in the historical production data and the historical cabinet data with the average value data of the previous moment and the next moment, and/or filling the missing data at a certain moment by a linear interpolation method; the normalization processing comprises the steps of carrying out linear transformation on the data after the data elimination filling