CN-122022030-A - Coke oven fire falling time prediction method and related device

CN122022030ACN 122022030 ACN122022030 ACN 122022030ACN-122022030-A

Abstract

The application belongs to a prediction method, and provides a coke oven fire falling time prediction method and a related device, aiming at the technical problems of insufficient prediction precision in the existing fire falling time judgment and prediction method, key variables are screened from coking related variables according to production scenes, the limitation of single factor input of the existing prediction model is broken through, various influencing factors are comprehensively considered, corresponding adjustment can be carried out according to the production scenes, and the comprehensiveness, the effectiveness and the simplicity of the input variables in the current production scenes are ensured. And obtaining a flame falling time prediction result through a coke oven flame falling time prediction model. The coke oven fire falling time prediction model adopts a long-period and short-period memory network structure, and the super parameters of the long-period and short-period memory network are optimized and determined through a sparrow search algorithm, so that the problem that the prediction performance of the prediction model is insufficient due to the fact that the super parameters of the traditional prediction model depend on experience setting is solved.

Inventors

LI SHUYU
LI MIAOMIAO
LI FEIFEI
JING CHAO
CHEN JIAXIN
ZHANG XINGZHONG
CHENG YONGQIANG

Assignees

山西省能源互联网研究院
太原理工大学

Dates

Publication Date: 20260512
Application Date: 20260121

Claims (10)

1. A method for predicting coke oven fire drop time, comprising: according to the production scene, screening key variables from coking related variables; The method comprises the steps of acquiring real-time data of key variables, inputting the real-time data into a coke oven flame falling time prediction model to obtain flame falling time prediction results, wherein the coke oven flame falling time prediction model adopts a long-period and short-period memory network structure, super parameters of the long-period and short-period memory network are determined through optimization of a sparrow search algorithm, and prediction accuracy of the long-period and short-period memory network is measured through mean square error.
2. The method for predicting the falling time of a coke oven fire according to claim 1, wherein the method for optimizing and determining the super parameters of the long-short-period memory network through a sparrow search algorithm comprises the following steps: Initializing sparrow searching algorithm and parameters of long-term and short-term memory network; substituting the hyper-parameter combination of a group of long-short-term memory networks corresponding to the position of each sparrow into the long-short-term memory networks, calculating to obtain a mean square error through a training set consisting of known key variables and fire time, and sequencing the calculated mean square error; Determining the participants and discoverers in the sparrow according to the sequencing result, and randomly selecting the inspector; updating the position of the discoverer, calculating the mean square error after each time of the position updating of the discoverer to determine the optimal position occupied by the discoverer, updating the positions of the enrollee and the scout according to the discoverer occupying the optimal position, and recalculating the mean square error after each time of updating; After each round of updating, sparks corresponding to the minimum value of the mean square error are selected from the jointers, discoverers and investigation persons to serve as optimal individual positions of the current population until the maximum iteration round is reached, and the optimal individual positions obtained during the maximum iteration round are used as super parameters of the long-short-period memory network which are determined in an optimization mode.
3. The coke oven dry time prediction method according to claim 2, wherein the finder's position updating method comprises: Wherein, the A random number of (0, 1), For the maximum number of iterations to be performed, In order to obey a normal distribution of random numbers, Is all 1 for one element The vector of the row is maintained, And Respectively an early warning value and a safety value of the sparrow population position, Is the first The first iteration The sparrow group is at the first The location information in the dimension is used to determine, Is the first The first iteration The sparrow group is at the first Position information in the dimension.
4. The coke oven dry time prediction method according to claim 2, wherein the position updating method of the joiner comprises: Wherein, the Represent the first The best position the finder occupies during the iteration, Representing the global worst-case location of the object, Randomly assigning 1 or-1 to elements A matrix of rows.
5. The coke oven dry time prediction method according to claim 2, wherein the position updating method of the inspector comprises: Wherein, the Is the first The current global optimum position at the time of the iteration, The step size control parameter is indicated as such, Is a random number, and is used for generating a random number, As the current individual fitness value of the sparrow, And The current global optimum and worst fitness values respectively, Is the smallest constant.
6. The method for predicting the falling time of a coke oven fire according to claim 1, wherein the method for screening key variables from coking related variables according to production scenes comprises the following steps: And carrying out association quantitative analysis on coking association variables by adopting grey association analysis, and screening out key variables.
7. The method for predicting the falling time of a coke oven fire according to claim 1, further comprising, after acquiring the real-time data of the key variables: and carrying out normalization processing on the real-time data of the key variables.
8. A coke oven dry time prediction system, comprising: The screening module is used for screening key variables from coking related variables according to production scenes; The prediction module is used for acquiring real-time data of key variables, inputting the real-time data into the coke oven flame falling time prediction model to obtain flame falling time prediction results, wherein the coke oven flame falling time prediction model adopts a long-period and short-period memory network structure, super parameters of the long-period and short-period memory network are determined through optimization of a sparrow search algorithm, and prediction accuracy of the long-period and short-period memory network is measured through mean square error.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the coke oven flame time prediction method of any one of claims 1-7.
10. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the coke oven fire fall time prediction method of any one of claims 1-7.

Description

Coke oven fire falling time prediction method and related device Technical Field The application belongs to a prediction method, and particularly relates to a coke oven fire falling time prediction method and a related device. Background Accurate prediction of the flame falling time is beneficial to a coking operator to judge the mature state of a coke cake in advance, and accordingly, the heating strategy of a subsequent coking process is adjusted, so that the method has important significance in stabilizing the quality of coke and realizing efficient utilization of energy. The duration of the fire drop time is influenced by multiple factors, including flame path temperature, raw gas temperature, water content of the coal entering the furnace, and the like, so that when the fire drop time is judged, the correlation between each related variable and the fire drop time needs to be fully explored. The existing fire drop time judging and predicting technology has a plurality of defects. The traditional method mainly comprises a raw gas color observation method, a raw gas combustion flame color observation method and a raw gas temperature detection method, and has the obvious defects that firstly, the raw gas contains sulfur dioxide, hydrogen sulfide and other toxic gases, gas leakage is easy to occur in the manual observation process, the body health of operators is endangered, meanwhile, the environment is polluted, secondly, the number of coke oven holes is large, the real-time monitoring of the whole oven holes is difficult to realize in manual inspection, the subjective perception of the operators is different, the judgment result of the flame falling time is lack of uniformity, and the error is large. Besides the traditional manual observation method, the traditional flame falling time prediction model has certain limitation that on one hand, the model fails to fully consider the comprehensive effect of multiple factors in the coking production process, only depends on few empirical indexes such as temperature, coal characteristics and the like, omits the coupling influence of multidimensional factors such as heating working conditions, furnace body states and the like, and on the other hand, the super-parameters of the neural network prediction model are mostly set by experience, so that obvious performance bottlenecks exist, prediction precision is limited, and high precision requirements of actual production on flame falling time prediction are difficult to meet. Disclosure of Invention The application provides a coke oven flame falling time prediction method and a related device, aiming at the technical problem of insufficient prediction precision existing in the existing flame falling time judgment and prediction method. In order to achieve the above purpose, the application is realized by adopting the following technical scheme: in a first aspect, the application provides a method for predicting the falling time of a coke oven fire, which comprises the following steps: according to the production scene, screening key variables from coking related variables; The method comprises the steps of acquiring real-time data of key variables, inputting the real-time data into a coke oven flame falling time prediction model to obtain flame falling time prediction results, wherein the coke oven flame falling time prediction model adopts a long-period and short-period memory network structure, super parameters of the long-period and short-period memory network are determined through optimization of a sparrow search algorithm, and prediction accuracy of the long-period and short-period memory network is measured through mean square error. Further, the method for optimizing and determining the super-parameters of the long-term and short-term memory network through the sparrow search algorithm comprises the following steps: Initializing sparrow searching algorithm and parameters of long-term and short-term memory network; substituting the hyper-parameter combination of a group of long-short-term memory networks corresponding to the position of each sparrow into the long-short-term memory networks, calculating to obtain a mean square error through a training set consisting of known key variables and fire time, and sequencing the calculated mean square error; Determining the participants and discoverers in the sparrow according to the sequencing result, and randomly selecting the inspector; updating the position of the discoverer, calculating the mean square error after each time of the position updating of the discoverer to determine the optimal position occupied by the discoverer, updating the positions of the enrollee and the scout according to the discoverer occupying the optimal position, and recalculating the mean square error after each time of updating; After each round of updating, sparks corresponding to the minimum value of the mean square error are selected from the jointers, discoverers and investigation